Columbia Economics Review: Special Issue 2018-2019 by Columbia Economics Review

COLUMBIA ECONOMIC REVIEW SPECIAL ISSUE 2018-2019, VOLUME X

FACULTY ADVISOR Wouter Vergote, Ph.D.

EDITORIAL BOARD Editor-in-Chief & President: Michael V. Crapotta Managing Editor (Journal): William Labasi-Sammartino Publisher: Gabriel Multedo Executive Editor (Online): Mathieu N. Sabbagh Webmaster: Adam Mann Executive Director: Makenzie E. Nohr Art Director: Jessica Lu Treasurer: Michelle Yan Senior Editor: Michael Chu Senior Editor: Spencer Papay Senior Editor: Neel Puri

STAFF EDITORS Lana Awadallah Hannah Buttle Anderson Chan Sinet Chele Chelagat Zachary Cui Ignacio Lopez-Gaffney Michael Gao Devyani Goel Jack Hannah Blaine Helleloid Brian Ling Ho Elizabeth Lee Jeremy Liu Noah Love Edmund Shen Nurassyl Shokeyev Cover Art: Nicholas A. DiCostanzo Cover Design: Michael V. Crapotta

THE TENTH ANNIVERSARY ISSUE

In loving memory of Kirk Wu, CER Staff Editor (2017-2018)

Contents

Letter From the Chair of the Department of Economics

Letter From the Program for Economic Research

Letter From the Editors

Interview 17

If others throw rocks into their harbor, it would be crazy to throw rocks into your harbor

Interview with Professor Jagdish Bhagwati

Articles 23

Trade Interdependence and Peace: Sino-Japanese Relations in the late 20th and early 21st Centuries

Real Exchange Rate Volatility and Economic Growth: A Panel Data Investigation

Amarvir Singh-Bal

Federico Pessina

61 Heterogeneous Effects of Bolsa Familia on Rural and Urban Households

Elena M. Stacy

The Cost of Investment of Legislative Turnover, Immobile Labor, and Corruption: Evidence from the States of India

Pulkit Agarwal

The Determinants of Nationalism and the Effect of Conscription on National Pride Dhivesh Dadlani 117 The Efficacy of Biomedical Research: A Bibliometric Analysis Across Longitudinal Respiratory Disease Data Zoey Chopra

Letter from the Chair of the Department of Economics It gives me great pleasure to welcome you to the Tenth Anniversary edition of the Columbia Economic Review (CER) Journal! I would first like to express my gratitude to Michael V. Crapotta (Editor-in-Chief & President) and his predecessors. Over the past ten years our undergraduates have a done a fantastic job, shaping the journal to ensure it has maintained its place as the leading undergraduate student-run economics journal in the country. The editorial teams have maintained the highest standards of quality, scientific rigor, and fairness and I commend all of their members. The Department is proud to support undergraduate research and to encourage critical thinking in our students. Original research articles form the bulk of the content of the CER journal, and editorials an important sub-section. The methodological quality of CER has improved dramatically over the last ten years, thanks to new guidelines and to the advice of faculty Wouter Vergote during the review process. Such emphasis on methodological rigor is vital to ensure that conclusions reached from articles contained in the journal are reliable. Editorials and commentaries play a key role in exploring more contentious issues in a balanced way. This issue shows the diversity of the questions that shape our field: from questioning the efficacy of biomedical research in the field, to analyzing the costs of legislative turnover, immobile labor and corruption in India, and further assessing the determinants of nationalism and the effects of conscription on national pride. Taken together, I am proud of our studentâ&#x20AC;&#x2122;s efforts. The Columbia Economic Review remains popular amongst readers. The use of electronic media is now a vital component of its dissemination: it provides wider access to download selected articles and allows the Review to give access to detailed appendices to articles. I invite you to visit the CER website (www.columbiaeconreview.com) and take advantage of the online tools created to enrich your reading experience. The online site also provides in-depth coverage of news, events, and development in the world of economics, and it includes information on upcoming events and back issues of the print journal.

Let me conclude by thanking all our submitters for choosing the Columbia Economic Review. Sadly, due to space constraints less than 20% of submissions were eventually accepted for publication. Those whose work was accepted should be proud of their achievement, and I encourage the others to continue to submit to the journal. I very much look forward to your continued contribution to the economic debate. Bernard SalaniĂŠ Professor and Chair Department of Economics

Letter from the Program for Economic Research This Tenth Anniversary edition of the Columbia Economic Review has been over six months in the making, involving a dedicated team of undergraduates at Columbia University. As you will see in this volume, this is a time of exciting change, both in the field of economics and in scholarly research. The articles presented in this issue of CER reflect the high levels of economic research and policy analysis of our students. All of our authors are current undergraduate students or recent graduates in economics or area studies. Our hope in supporting the publication of a student-run journal is not only to produce research, but to highlight the significance of undergraduate research at the intersection of economics, business, politics, and society. This issue consists of two types of writing: articles and editorials. The articles were selected and approved through a peer review process involving undergraduate students. Each article was then reviewed by my colleague and the journalâ&#x20AC;&#x2122;s faculty advisor, Wouter Vergote. The six articles presented here are the result of several rounds of review, editing and collaboration between our authors, reviewers and editors. All the articles in this issue deal with large-scale, global issues in economics from trade interdependence and peace, to real exchange rate volatility and economic growth. The editorials come from the CERâ&#x20AC;&#x2122;s editorial board, the students are responsible for shaping and guiding the creation of this journal. They present unique and sometimes personal perspectives and insights into economic policy planning and practice at home and abroad. Finally, I want to congratulate our students and contributors for sharing their ideas and advancing knowledge and learning at the highest level. David E. Weinstein Carl S. Shoup Professor of the Japanese Economy, Columbia University Executive Director, Program for Economic Research

Letter from the Editor We are excited to present our 2018-2019 Edition of the Columbia Economic Review. This year marks an important milestone for our organization – our tenth anniversary. This special edition Tenth Anniversary Issue represents the hard work and persistence of every member of our organization. With support from the Department of Economics and the Program for Economic Research, we have completed a total redesign of our biannual journal, which has culminated in the publication you hold in your hands. This feat would not have been possible without the support of Sophia N. Johnson, Ph.D., Assistant Director of the Program for Economic Research, to whom we wish to extend special thanks. We would also like to extend a special thank you to Columbia freshman Gabriel Multedo, who took on the responsibilities of publisher this year. In this edition of our journal, we enter a new era of excellence for the Columbia Economic Review – an era in which we are thrilled to have the official support of the Department of Economics. With the backing of the Department, we have been able to explore issues related to current economic and political developments across the world. Select topics in this edition include the dynamics of trade policy, the efficiency of social welfare programs, and the influences of nationalistic ideals. The diverse international perspectives of our published authors make for a unique edition, wherein the aforementioned topics are investigated through a variety of distinct case studies. Given the increased pace and scope of globalization today, we anticipate that these case studies will serve as a strong foundation for further analyses. The Tenth Anniversary Issue contains six submissions drawn from undergraduate institutions across the globe. With the guidance of our newly appointed advisor, Wouter Vergote, Ph.D., we have successfully improved the quality of our publication and have ensured not only aesthetic appeal, but also economic rigor. In this issue, we are also honored to include an exclusive interview with Jagdish Bhagwati, University Professor of Economics, Law, and International Affairs, and former Senior Fellow for International Economics at the Council on Foreign Relations. Professor Bhagwati’s extraordinary insight has helped to elevate our journal to new heights in the realm of contemporary economic discourse. We are indeed grateful for Professor Bhagwati’s support.

Our diverse selection of papers is indicative of our mission to enlighten our readers with the most novel and topical papers from the world’s undergraduate economics students. Amarvir Singh-Bal of Josephine Butler College, Durham University, examines trade between Japan and China in China’s post-autarky period, and the accompanying effects on social and political cohesion between the two countries. Singh-Bal thus provides a historical look at Sino-Japanese relations, the development of which are pivotal to understanding contemporary issues of trade. Federico Pessina of the University of Warwick also investigates historical trends in trade, by analyzing volatility in the real effective exchange rate and associated impacts on growth in developed and developing countries over the past several decades. A pair of selections in our journal focus on case studies of governmental efficiency in Brazil and India. Elena M. Stacy of the University of California, Berkeley, investigates heterogeneity in short-term poverty alleviation through the Bolsa Familia social welfare programs in Brazil, while Pulkit Agarwal of Harvard University provides insight into the effects of governmental hindrances on the usage of capital investments in India. Dhivesh Dadlani, another author from Harvard University, contributes to our selection with a socio-politically driven analysis of nationalism, which investigates the relationship between mandatory conscription policies and metrics of citizens’ nationalistic views. Finally, we are pleased to present Zoey Chopra’s (Columbia University ’18) bibliometric analysis of biomedical research efficacy. In his paper, Chopra estimates the effects of innovation in published research on tangible health outcomes related to multiple respiratory diseases. Zoey’s paper garnered the 2018 Sanford S. Parker Prize, which honors undergraduate economics students at Columbia University who have “demonstrated a boldness of thought and a commitment to excellence” in economic research. Zoey has gone on to pursue a dual M.D./Economics Ph.D. at the University of Michigan. We are proud to represent such immense talent from our own institution in this special edition of our journal. From everyone here at the Columbia Economic Review, it is truly a pleasure to present to you this issue. We are honored to have published our journal for the past ten years and look forward to continue our work in the decades to come. We hope that you enjoy our Tenth Anniversary Edition. Michael V. Crapotta Editor-in-Chief & President, Columbia Economic Review Columbia College ‘19

“IF OTHERS THROW ROCKS INTO THEIR HARBOR, IT WOULD BE CRAZY TO THROW ROCKS INTO YOUR HARBOR”

An Interview with Professor Jagdish Bhagwati Columbia University Free trade is considered to be most vulnerable during economic downturns, when governments seek to (often unfairly) evaluate stresses that contribute to poor outcomes. In the current political climate in the U.S., we are not experiencing such a downturn, but still protectionist views are being promoted and enforced. Today, we are joined by Jagdish Bhagwati – University Professor of Economics, Law and International Affairs at Columbia and former Senior Fellow in International Economics at the Council on Foreign Relations – who will discuss these tensions and their broader implications. Columbia Economic Review: In Why Growth Matters, your 2013 book co-authored with Professor Arvind Panagariya, you provide insight into Indian growth strategy and effectiveness in reducing poverty since Independence. In what ways does your analysis of the Indian experience provide lessons for povertyelimination in other developing countries? Jagdish Bhagwati: We surely think our analysis offers lessons applicable to other countries with much unemployed and underemployed population (ruling out therefore lessons for countries such as Saudi Arabia with abundant oil revenues and underpopulation). Of course, many analysts claim that the Indian experience is unique to her because she is a very big country with multiple ethnicities, languages, religions etc. But these attributes make India’s lessons more, not less applicable, to countries not so endowed with potentially fissiparous factors in their politics. The main lesson we drew from the Indian experience was that the temptation to intervene to accelerate growth, and hence to impact on poverty, was a powerful one since, when you were behind in development, you felt that the government must intervene everywhere to accelerate growth. This seemed like common sense just as steering in the direction you are moving off the road under icy conditions seems like common sense. But it is wrong, of course. Besides, we also discovered that one intervention will often lead to others as

Tenth Anniversary | 17

ProfessorJagdish Bhagwati

new claimants for similar largesse will multiply speedily and cannot be refused politically. The Indian experience showed that the efficacy of growth in lifting the poor out of poverty was clearly undermined by such growth-inhibiting interventionism which was justified by saying that socialism, an “in” word at the time, demanded it! Our reforms, starting in earnest in early 1990s, therefore consisted in dismantling the “socialist” model that had failed to deliver growth and hence poverty-reduction. While some skeptics of the growth strategy for poverty-reduction like Amartya Sen and Joe Stiglitz complained that we were shifting to a right-wing strategy where markets would replace pragmatic interventions, their criticism made little sense since we were moving from a situation where markets were virtually absent and interventions were almost universal, to a pragmatic center. Indian reforms were therefore driven by the realization within India itself (a realization that I was at the center of ) that the old “socialist”, intervention-intensive model was not working, and that growth was slow and poverty-reduction was not being achieved. The leftwing canard was that the market-oriented reforms were forced by Washington; they expected that such a charge would help sabotage the reforms politically. In fact, it was Washington that learned the reforms from Indian economists such as myself! The same was true of the reforms set in motion in Soviet Russia by Gorbachev and in China by Deng; in each case, these countries knew that the old model was not working. If you think in terms of the Harrod-Domar growthmodel, the investment rates were astonishingly high in the Soviet Union but the productivity of these investments was abysmal. Evidently, with growth rates awfully low, the impact on poverty reduction could not be large. So, my work highlighted the critical role of high growth rates, resulting from the reforms, in reducing poverty. The problem was that the left-wing, ideological critics had gained the high ground by calling the “growth strategy” a conservative “trickle down” strategy, reminding one of feudal lords gorging themselves on leg of lamb while crumbs fell to the dogs and the serfs below the table! So, I shifted the terminology to “pull up” strategy where an activist agenda would help accelerate growth and this would “pull up” the poor into gainful employment. CER: As the world has become increasingly interconnected, labor migration has spurred a number of responses – economic, sociopolitical, and otherwise. Referring to labor economist George Borjas’ analysis of immigration in 1998, you call his argument in favor of a preference for skilled immigrants “both morally unacceptable and economically unconvincing.” As a political economist, how do you conceptualize the interaction of morality and economics in prescriptive, policy-oriented research?

18 | Columbia Economic Review

Challenges to Free Trade

JB: To me, the preference for skilled immigrants sounds reasonable if you’re looking at your welfare as the receiving society. If you ask an average American: “whom would you want to benefit [from labor migration]?” I think an American would say, “I would give my last available slot [in the economy] to somebody who is going to benefit more from than a highly-skilled immigrant would.” But, nowadays, we are moving towards giving visas to people who are going to be beneficial for us. That attitude is really the one which is affecting the things which we [economists] try to do. But, how many people have objected to it? We are an aging society, and particularly since we are aging, we do need people to take care of our older people. We see that this is an opportunity that unskilled women take advantage of. It is a good thing for us to be able to access this unskilled and semi-skilled labor and for our women who are typically given this responsibility, but want to join the labor force. So, for us to say unskilled labor is no good, sells this particular problem short. If we get these people, we will be taken care of, our women will be able to participate, and we will also be able to properly train people from around the world. So, we should not knock, or run down, unskilled labor. Really, it is equally useful. Secondly, my feeling also is that if you let in unskilled people, they are more likely to be hungry for success because for them, this is a great opportunity. For them, it’s a big breakthrough to be able to access this opportunity. And that is why ethically I feel we should not discriminate against people at the lower end of the skill spectrum. CER: One of the most popular articles in our classes is your 1976 article “Taxing the Brain Drain.” In regard to the positive effects of semi-skilled labor migration to the U.S., do the sentiments surrounding this article change? JB: One main rationale of “Taxing the Brain Drain” was basically that migration is not free. We’ve got immigration controls. Of course, they leak, but by and large there is a barrier. Those who were lucky enough to go from where they are, which is at a much lower level of income, come to a place where incomes are much higher. That means you get a massive gain of income by being able to get in. So, the idea was simply that one should be able to catch those gains for the benefit of those left behind. The idea was basically to collect money for decent governments and to say “Look, let people go if they want to.” In my view, this was a way of saying, if you’re lucky enough to manage to get from India where there, even at the highest levels or income, you’re still way below what you can earn here, share some of it. So, my idea was to have a surcharge of 1% or 2%, similar to revenue sharing. CER: Let’s pivot a bit toward some recent literature, specifically “The China Shock: Learning from Labor-Market Adjustment to Large Changes in Trade” by Autor, Dorn, and Hanson. Paul Krugman has stated that the main takeaway from the Tenth Anniversary | 19

ProfessorJagdish Bhagwati

paper was that “Ricardo and Heckscher-Ohlin were less relevant to the political economy of trade than the sheer pace of change, which disrupted local manufacturing concentrations and the communities they supported”. We would love to hear some of your thoughts regarding this debate. JB: I must say at the outset that the paper by Autor, Dorn and Hanson claiming to show that Chinese imports correlate with adverse impact on manufacturers in different parts of the US and that therefore competition with China has undermined output and wages in US manufacturing is far from convincing despite the hysteria with which it has been received. Phil Levy of the Chicago Council of World Affairs has pointed out, if safeguard actions have been invoked against imports from China, the effect would surely have been to “shunt” exports from China to other exporters. The exclusive focus on Chinese exports is therefore misplaced. Besides, when one looks at the total amount of wages lost due to decline in employment in the Author-Dorn-Hanson study, it is a minuscule amount: the average worker in their study is estimated to have lost less than 10 dollars a week in nominal income. This is such a small amount that it is likely to have been outweighed by gain from lower cost of imported goods from China. As for the loss of manufacturing to China, it is also worth noting that manufacturing value added has declined virtually everywhere. The notion that trade openness has led to a decline in US manufacturing, by having US manufacturing shift to foreign countries with cheap labor or other natural and government-created artificial advantages in recent years, is erroneous but has nonetheless created a near-frenzy of protectionism by not just President Trump but by many Democratic politicians. As for Paul Krugman, arguments that a rapid pace – actually, the size – of change creates adjustment difficulties, which can create a protectionist backlash, that is of course true but also part of conventional wisdom among international economists. This applies equally to immigration. Thus, if a million Islamic fundamentalists moved into New York, they would tend to overwhelm and undermine local culture. But if a thousand did, the local culture would overwhelm them instead. The problems that President Trump has created for free trade relate rather to his rejection of the traditional view that it is wrong to argue that if other nations are protectionist, we should also be, i.e. that unilateral free trade is harmful. As my Cambridge teacher Joan Robinson famously said: if others throw rocks into their harbor, it would be crazy to throw rocks into your harbor. President Trump rejects this view, of course. But ironically, it was Paul Krugman who undermined the traditional view and argued that unilateral openness in the face of protectionism by your trading partners could harm you! So, one might as well call it the Krugman-Trump doctrine! 20 | Columbia Economic Review

Challenges to Free Trade

Again, President Trump often argues that he wants to use US tariffs to “pry open” foreign markets, an argument he has made especially with regard to China. Ironically, this argument goes back to Adam Smith and I wrote a whole book on this issue, with a cartoon on the cover designed by me, showing Uncle Sam prying open oysters! The problem is that, once you allow for retaliation, it is no longer clear that you would win from such a tariff policy. In fact, the “war of tariffs” between China and the US is indicative of how tricky such a policy can be. Besides, where President Trump is clearly mistaken is in his belief that the openness in trade of a country can be inferred from whether it runs a payments deficit. There is no necessary relationship between trade openness and external payments deficits. President Trump has however correctly argued that China’s practice preventing access to its market by foreign firms unless they invest in China, is predatory and forces these firms to share their technology with Chinese entities. This is tantamount to de facto intellectual property theft. So, tariffs are imposed and lead to “tariff-jumping” inflow of foreign investment and associated diffusion into China of IP. But the US has also used protectionism to attract foreign investment, but in a different way. The US administrations have used the threat of protectionism to induce foreign firms to transplant production into the US, in a tactic used against Japan and christened by me as “Quid Pro Quo” foreign investment: if the foreign firm invests in the US, it will be allowed to enter the market with production elsewhere. E.g. Toyota will be allowed to retain its access to the US market for Toyotas produced outside the US provided Toyota invests in producing some Toyotas in the US. There is in fact a huge literature on this. Just as tariff-jumping influx of foreign investment leads to induced influx of IP in the Chinese case, influx of Quid Pro Quo investment also does this. E.g. surely we learned how to produce small cars this way since our DNA was for producing large cars. CER: Today, economists seem to have reached consensus on the case for free trade. But, at the policy level, there is skepticism by many. There is also the danger that we may end up with escalating trade disruptions as rival trading nations like the US and China wind up with confrontational policies. What do you think? JB: The future is not easy to predict. Let me illustrate it with China. Will it be more amenable to changing its predatory behavior or will it reject the US administration’s demands? The authoritarian Chinese economy is in a bad shape, facing a sharply reduced growth rate. It also faces possible containment from a US-led coalition of four major democracies: Japan, India, Australia and the US, as it flexes its military muscle and confronts nations in East Asia and South Asia. Tenth Anniversary | 21

ProfessorJagdish Bhagwati

Its human-rights record on Tibet and the Muslim areas is also increasingly challenged internationally. Its Belt and Road Initiative, originally welcomed as adding to funds available to developing countries, is now seen widely as a debt-trap that would lead to surrender of critical sovereignty in these countries. Will this ongoing reversal of fortune, as it were, prompt a proud civilization to dig its heels in or will it lead to more accommodating behavior? Only time will tell.

22 | Columbia Economic Review

TRADE INTERDEPENDENCE AND PEACE: SINO-JAPANESE RELATIONS IN THE LATE 20TH AND EARLY 21ST CENTURIES

Amarvir Singh-Bal Josephine Butler College, Durham University Abstract: This paper argues that trade between Japan and Communist China in its post-autarky period (from 1978 to the present) led to increasing social and political cohesion between the two countries. Throughout this period there was an unprecedented growth in trade and foreign direct investment (FDI) between the two countries. Trade increased from US$568m in 1978 to US$1.2tn in 2014 (International Monetary Fund 2017), making it the third largest bilateral trade relationship in the world (Jain 2006). Despite these deepening economic and financial ties, the majority of the existing literature on the Sino-Japanese relationship is focused on geopolitics. Attention has been diverted to moments of political importance such as a leaders’ visits to the Yakashuni Shrine, or student-led anti-Japan demonstrations in the spring of 2005 (Mori 2006). This paper will seek to reverse this trend. The behavior of exporters and investors is a lesser known, but no less important, historical force of conciliation between China and Japan.

his paper articulates an alternative to the neo-realist claim of Whiting (1992), Geng (2011), and Vencaleck (2015) that socio-political interactions are driving commerce in Sino-Japanese relations, and not vice versa (Keshk, Pollins, Reuvenv 2004). The paper proposes a neo-liberal interpretation of Sino-Japanese relations, claiming the diminution of tensions arises from increased economic interdependence (2015). Private sector interests increased the economic wellbeing of the two countries, facilitating laxer trade policies and increasing interdependence and cooperation. Increased economic contact fostered a regularized social relationship, increasing the probability of political co-operation and decreasing the probability of conflict between the two nations (Hilpert 2002). Academics who argue in a similar vein, including Armstrong, have maintained that in the past, the “economic relationship prevented the intensification of political tensions” (2014). An almost identical position is taken by Hilpert, who maintains that an “increasing volume of trade shows a tendency for bilateral cooperation” (2002). Burns takes a more general approach and suggests that “economic relations [in general] ... provided a balm in times of trouble, and marked the way to formal diplomatic relations” (2000). As such, this paper offers three analyses which in conjunction make an argument for the positive effect of trade and investment on social and political relations between Tenth Anniversary | 23

Amarvir Singh-Bal

the two countries. The three analyses are i) that Japan’s private sector sought to trade with the People’s Republic of China (PRC) because of its Open-Door Policies; ii) that Japanese firms that originally traded with China eventually relocated their manufacturing bases there, which led to increasing social cohesion; and iii) that the increasing social cohesion as well as mutual economic gains, derived from Japanese FDI, spurred diplomatic political cohesion in Sino-Japanese relations. Japan’s private sector sought to trade with the PRC because of its OpenDoor Policies. Contrary to neo-realist claims, it is evident that the Japanese private sector sought to develop economic relationships with the PRC as a result of China’s Open-Door Policies. The Japanese private sector generated economic integration at a sub-national level, which bolsters the neo-liberal argument that this went on to cause social and political cohesion at sub- and state levels. This section will show that increasing consumer confidence, as a result of China’s Open-Door Policies, incentivized the Japanese private sector to trade with the PRC. Consequently, the existing level of cost competition associated with bilateral trade led to an increasing inflow of Japanese FDI into China. This analysis is necessary to explain how Sino-Japanese FDI led to increasing social political cohesion between the two nations. Creating Confidence Between 1978 and 2014, a series of liberal economic policies spearheaded by Deng Xiaoping led Japanese firms to realize that China welcomed the idea of trade. This encouraged the Japanese private sector to establish trade links with China. The creation of Special Economic Zones (Dixin 1981) and the reopening of the Shanghai Stock Exchange in 1990 are examples of such Open-Door Policies through which China welcomed trade relationships. By the end of 1981, the construction of 898 trading warehouses throughout the zones (Stoltenberg 1984) allowed China to create an image of economic competency (Naughton 2008). This was done in the hope of establishing China as a host for long term trading relationships on the global stage. Even after Deng Xiaoping’s death, such Open-Door trade policies continued to gain momentum in the 21st century, enticing more Japanese private sector firms to trade with China. The result was unintentional inflows of FDI into the PRC by the Japanese private sector.

24 | Columbia Economic Review

Trade Interdependence and Peace

Source: World Bank

Conforming to Anderson and Cross’s interpretation (2004), China’s accession to the World Trade Organization (WTO) gave the Japanese private sector even more confidence (Armstrong 2014). No other member of the WTO had been required to make as many legal, economic, and political changes as did the PRC during its admission in 2001 (Drysdale & Kalirajan 2000, Brandt & Rawski 2007). Leveraging this position, the Chinese were able to project an image of stability and credibility. As seen in figure 1.1, the most significant tariff reductions did in fact occur in the 1990s, but trade, as a proportion of GDP, did not rise significantly until around the time of WTO accession when it increased by 24% in 4 years, the single largest percentage increase in the period (Armstrong 2014). The significance of China joining the WTO was profound: Japanese exporters saw China as a credible investment destination for the first time. Japan’s trade with China led to an increase in FDI Despite the large increase in Japan’s trade with China, several factors increased competition for Japanese firms in Chinese markets. To overcome the competition, Japanese firms increased FDI within mainland China. The Open-Door Policies throughout the period fostered extensive interaction with the outside world, opened China to foreign capital, and promoted the export of Chinese products (Burns 2000). The economic relationship between China and Japan was therefore complementary. Sino-Japanese bilateral trade grew more than tenfold between 1979 and 1999. By 1992, China was Japan’s fifth largest trading partner; a year later, China overtook Germany, Taiwan and South Korea to become Japan’s second largest trading partner after the United States (Burns 2000). The Sino-Japanese economic relationship stands out because of two structural and complementary

Tenth Anniversary | 25

Amarvir Singh-Bal

characteristics. First, China exploited its comparative advantage of possessing relatively low-wage production processes within a large growing market. Second, Japanese firms employed their factor endowments in exporting their physical capital and technologically sensitive products (Hilpert & Nakagane 2002). Despite the increase in bilateral trade, the relatively high cost of Japanese products competing with Chinese firms incentivized the Japanese private sector to increase FDI in the PRC and thus to overcome cost competition. Increasing FDI in the PRC also allowed Japanese firms to reap the same factor endowments, such as cheap labor, from which competing Chinese firms also benefitted.

Source: www.fxtop.com

As such, Japanese firms sought to engage in FDI initiatives across China’s Special Economic Zones (SEZs), as a means to overcome three domestic cost issues. First, unlike Japan, China was capable of supplying labor-intensive products, which were in high-demand in Japan’s industrial and consumer markets (Hilpert & Katsuji 2002). Secondly, China’s import market was protected by high tariffs and other trade barriers, such as government bureaucracy (Xinxin 2002), which incentivized Japanese firms to increase FDI in the Chinese market (Hilpert 2002). Thirdly, the appreciation of the yen relative to the Chinese RMB— held up by economic historians such as Xing (2006) as the most important factor—also led to increasing flows of Japanese FDI into China. As seen in figure 1.2, the 779% increase in real-exchange rate appreciation of the Japanese currency reduced the competitiveness of Japanese trade exports to China. In order to overcome these three obstacles to competitive trade, the Japanese private sector increased FDI across the PRC. Doing so also allowed Japanese firms to re-export their products back to Japan at

26 | Columbia Economic Review

Trade Interdependence and Peace

a lower manufacturing cost. As such, the PRC seemed to be a highly attractive location for manufacturing in Japan’s immediate vicinity (Hilpert 2002). This analysis is a first step towards understanding how Japan’s trade with China led to increasing social and political cohesion between the two nations. It has been established that the Japanese private sector traded with the PRC as a result of China’s Open-Door Policies and that China’s comparative advantage led to Japanese FDI. On this basis, we can show that, at a microeconomic level, Japanese trading firms engaged in FDI to expand their operations. Japanese firms that traded with China eventually relocated their manufacturing bases there, which led to increasing social cohesion Japanese trading firms increasingly relocated to China, and indeed relied on the growing Chinese market to expand their businesses. The success of Japanese FDI trading in China led to increasing social cohesion for both nations. This social cohesion produced increasing political cohesion at the sub-state and state level in Sino-Japanese political relations in the period. Increasing reliance of the Japanese private sector on the Chinese market Financial statements of Japanese firms in China can help us see the success experienced by these firms at the time. Instances include the Beijing Nantsune Meat Machinery Joint Venture Co. Ltd in the 1980s. The firm’s sales were reported to be RMB 0.38 million in 1986 when they operated within Japan. However, after relocating to the PRC, the organization’s profits increased to RMB 24 million in 1999 (Xinxin 2002). This is only one example, but it illustrates the fact that Japanese firms which previously exported to China reaped greater economic rewards when they relocated. TF Tianjin Industrial Co Ltd. also profited from this arrangement, generating a profit of RMB 5.21 million in 1999 versus a loss of RMB 0.73 million in 1990 (Xinxin 2002). Such relocations of Japanese firms to China are but a microcosm of a broader historical trend throughout the period. Increasing reliance of the Japanese private sector on the Chinese market Financial statements of Japanese firms in China can help us see the success experienced by these firms at the time. Instances include the Beijing Nantsune Meat Machinery Joint Venture Co. Ltd in the 1980s. The firm’s sales were reported to be RMB 0.38 million in 1986 when they operated within Japan. However, after relocating to the PRC, the organization’s profits increased to RMB 24 million in 1999 (Xinxin 2002). This is only one example, but it illustrates the fact that Japanese firms which previously exported to China reaped greater economic rewards when they relocated. TF Tianjin Industrial Co Ltd. also profited from this arrangement, generating a profit of RMB 5.21 million in 1999 versus a loss of RMB 0.73 million in 1990 (Xinxin 2002). Tenth Anniversary | 27

Amarvir Singh-Bal

Such relocations of Japanese firms to China are but a microcosm of a broader historical trend throughout the period.

Source: JBICI

A survey conducted by the Japanese Bank for International Cooperation (JBIC) illustrates that between 1996 and 2014, a larger proportion of Japanese trading firms relied on the Chinese market than on any other single national market to maintain their operations. As seen in figure 2.1, between 1996 (when the survey was established) and 2013, the Japanese exporting firms increasingly depended on the Chinese market to manufacture their products. Japanese firms supplying FDI favored China over any other nation as a manufacturing base. JBIC surveys consistently reported that this was largely due to the low-cost production processes available in the PRC. Due to the success of the Japanese firms with bases in China, other Japanese firms followed suit. Because of this, FDI from Japanese firms that originally only traded within China increased further. This generated a flood of Japanese FDI in the PRC (Knickerbocker 1973, Armstrong 2014). This trend is shown by surveys conducted by the JBIC (figure 2.2). Japanese export firms began to regard China as the most promising destination to expand medium-term operations given the formerâ&#x20AC;&#x2122;s increasing reliance on Chinese subsidiaries. Increasing optimism led to the bandwagon effect, as described by Knickerbocker, and consequently led to further increases of Japanese inflows of FDI in China, as Japanese firms followed their rivals. Between the years 1983 and 2014 FDI flows increased from US$0.6bn to US$8.35bn (Statistical Year 28 | Columbia Economic Review

Trade Interdependence and Peace

Source: JBICI

Book of China, viewed 2017). The survey conducted by the Exim Bank of Japan, released in 1994, also suggests a correlation between success and reliance of Japanese firms on their established FDI in China. Increasing inflows of Japanese FDI led to increasing social cohesion within Sino-Japanese relations The inflow of Japanese export-orientated FDI in the PRC has led to increasing social cohesion within Sino-Japanese relations at the organizational and the cultural level. At the organizational level, as a direct result of the increase of Japanese FDI, a rise in social cohesion was demonstrated in both strategic management and corporate responsibility. Within joint venture partnerships, Japanese and Chinese employers and employees became increasingly acquainted with each otherâ&#x20AC;&#x2122;s different management styles. Lang concluded that Sino-Japanese joint ventures throughout the late 20th and early 21st centuries gave birth to a fusion of two different cultural-management techniques (Lang 1998). The firms at the forefront of this trend include the aforementioned Beijing Nantsune Meat Machinery Joint Venture Co. and TF Tianjin Industrial Co Ltd. (Xinxin 2002). Additionally, Austin and Harris observed that Japanese firms diverted their profits to address community social needs (Austin & Harris 1999) by supporting a variety of public policy initiatives, such as sponsoring the Sino-Japanese Software Research Institute in conjunction with the Chinese Tenth Anniversary | 29

Amarvir Singh-Bal

Academy of Sciences (Keji Ribaor 1994), as well as supporting local government on urban planning initiatives (Xinhua 1997). Evidence of such increasing social cohesion was often reported by national columnists, especially in China Daily (Business Weekly), which ran four articles by an official of the Ministry of Foreign Trade and Economic Co-operation on the establishment of social relations between Japanese firms and the local Chinese population (Liu 1997, Austin & Harris 1999). These are all clear signs of social cohesion. Upon establishing itself in Chinese markets, Japanese FDI was used to supply Japanese goods to the domestic Chinese consumer, which increased socio-cultural cohesion. From a neo-liberal perspective, Vyas and Baoguo’s research demonstrates that this took place through the consumption of Japanese products and media, which stimulated the exchange of tourism and language between the two nations. Chinese consumers wanted to buy Japanese products and emulate Japanese lifestyles, because throughout this historical period they were considered attractive due to their high quality and reliability (Vyas 2011). Between the 1990s and 2014, the influence of Japanese culture expanded within Chinese households, increasing sales of Japanese goods that came directly from FDI plants, namely Sony, Panasonic, Honda, and various food companies (Baoguo 2010). According to Baoguo, Japanese workers situated in the SEZs imported and spurred on the consumption of Japanese film, anime, and video games. Chinese national television increased its screening of Japanese TV from 0 hours in 1987 to 123.83 hours in 1993. Baoguo argued that this “drastically transformed the image of the ‘devils’ that the Chinese had long harbored toward the Japanese people, into that of people who have a sense of justice and are to be loved” (2010). He concludes that such an acceleration of consumption led to “improved bilateral social relations” as well as “bringing together emotional ties and understanding between the peoples of the two countries” (2010). The resulting communication within firms, as well as exposure to Japanese products and media by Chinese consumers, led to the creation of the Japan Foundation. This facilitated social cohesion by accentuating Sino-Japanese cultural similarities (JATA 2007). The Foundation was funded by Japanese investors. It hosted exhibitions and encouraged cultural rapprochement by airing issues of common interest, such as the protection of Sino-Japanese film industries against Hollywood (Japan Foundation 2008). Between 2010 and 2014, 200 Japanese language workshops were conducted nation-wide for Chinese employees within SEZs as well as locally. The Foundation also subsidized cultural exchange programmed which spurred on social cohesion at a grass roots level. The shared social and cultural identity that resulted from the inflow of Japanese FDI led to increasing political cohesion at the sub-state and state level in Sino-Japanese political relations during the period. This will be explored in more depth in the next section. 30 | Columbia Economic Review

Trade Interdependence and Peace

The increasing social cohesion and mutual economic gains, derived from Japanese FDI, spurred diplomatic political cohesion in Sino-Japanese relations The increasing social cohesion which resulted from net inflows of Japanese FDI paved the way for increasing political cohesion within Sino-Japanese relations. This section demonstrates that the resulting mutual economic gains, as well as pan-Asian sub-national political movements, prompted Japanese state leaders to implement strategies for integrating with their Chinese neighbors. This third analysis conclusively supports the neo-liberal claim that foreign relations were developed through economic interdependence. Mutual macroeconomic gains as a result of Japanese firms relocating to China As the situation now stands, both China and Japan have a vested interest in their economic interdependence and increased social contact (Austin & Harris 1999). Historically, Japan reaped considerable macroeconomic benefits from its export industries which relocated to the PRC. As Wakasugi observes, such outward flows of FDI limited the prospect of the existing domestic inflation and yen appreciation. The outbound Japanese FDI reduced overwhelming trade surpluses (Wakasugi 1996). Moreover, by making use of lower production costs, Japanese FDI in China, when re-exporting products back to Japan at a lower price, also diminished inflationary pressures (Ernst 1997 & Kwan 2002). The Chinese economy had also benefited in terms of an increase and improvement in technological capital formation (Chulai 1997), human capital (Deng 1997), export led growth (Nakagane 2002), and tax revenues (Bende-Nabende 1999). Throughout the late 20th century, FDI contributed to the PRC’s Gross Domestic Product (GDP). According to the East Asia Analytical Unit, FDI between the late 20th and early 21st centuries contributed significantly to China’s export growth as well. The share of China’s exports from foreign investments rose from 0.11% in 1980 to 41% in 1995 (Canberra 1997). Examples of export-led growth include the re-export of Japanese manufactured goods. According to a 1995 MITI survey, 29% of Japanese production in China was exported back to Japan (Burns 2000). This generated further economic growth (Wang & Swain 1995). Such empirical evidence of inbound Japanese FDI driving export-led growth within the period is vital in illustrating the positive effects this has had on the national economy of China. The subsequent governmental tax revenue from Japanese export firms relocating to China also generated 15% of China’s industrial and commercial tax revenues (Research Team on Environment and Policy for Foreign Investment in China 2000 & Xinxin 2002). Therefore, as a result of FDI generated by Japan’s trade with China, both nations needed one another politically and economically. Tenth Anniversary | 31

Amarvir Singh-Bal

Political Cohesion at The Sub-State Level Mutual economic gains and social cohesion, resulting from Sino-Japanese trade, followed by FDI, led to the rise of pan-Asian nationalist movements in Japan. The movements prompted state leaders to become politically integrated with China. This was largely due to two reasons. First, the cultural distance of the Japanese public from the west was widened. Secondly, economic benefits that Japan reaped from FDI increased incentives for cooperation. Firstly, as Wang maintains, Japanese citizens wished to detach themselves from the West and converge politically with China. This was due in part to the opposition towards close military relations with the U.S., which were viewed as compromising Japan’s independence (Wang 2005). The wish to detach from the West was also prompted by the belief that Japan’s destiny lay in closer relations with China, which had more in common with Japan from a social and cultural perspective. Secondly, Japanese firms’ dependence on the US market was reduced, due to relocations to China. This was exacerbated by the North American Free Trade Agreement (NAFTA) and the erection of EU tariffs which resulted in Japanese industries losing ¥400billion of their global market share. GDP fell by ¥620 billion, and 32,000 jobs were lost (Hideo 2004). As Johnston argues, these issues motivated Japanese firms to lobby political parties in favor of pro-China policies (Johnston 1986). The Japanese electorate, lobbyists, and politicians began to campaign for increasing Sino-Japanese integration. For instance, Japanese Liberal Democratic Party back benchers, who acknowledged they were backed by Japanese investors, relocated to China and forced Prime Minister (PM) Takeo Fukuda to complete the Treaty of Peace and Friendship between Japan and the PRC in 1978. This treaty stated that “the contracting parties shall develop relations of perpetual peace and friendship” and “endeavor to further develop economic and cultural relations between the two countries and to promote exchanges” (Treaty of Peace and Friendship Between Japan and the People’s Republic of China, 1978). It is therefore clear that from the grass roots level, increasing social cohesion and mutual economic gains propelled political cohesion onto a state level. Political cohesion at the Diplomatic Political Level As China grew more powerful throughout this period and the Japanese economy became more intertwined with that of its neighbor, both nations had reached out to each other in other non-economic spheres (Burns 2000). This was reflected by PM Koizumi’s speech at the Boao Forum for Asia, “I see the advancement of Japan-China economic relations [...] as an opportunity to nurture new industries in Japan and to develop their activities in the Chinese market. Our integrated efforts for economic reform in both

32 | Columbia Economic Review

Trade Interdependence and Peace countries should advance the wheel of economic [and political] relations” (Koizumi 2002).

Building on its own economic power, as well as its economic relationship with China, Japan led a political crusade to integrate the PRC within the international community (Burns 2000). Japanese efforts to foster political cohesion have been evident throughout the late 20th and early 21st centuries. An example of Japan befriending China in order to increase its integration in international markets include Japanese politicians persuading the leading OECD Development Assistance Committee members to admit China under the category of “developing country” (Deng 1997). The Japanese also assisted China’s fellowship in the World Bank and the International Monetary Fund, as well as urging other members to renew China’s membership of international institutions after it was revoked following the Tiananmen Incident (Austin & Harris 1999). Japan also facilitated the PRC’s involvement within the 1987– 1992 Uruguay Round of the General Agreement of Trade and Tariff conference, and helped China’s attempt to enter the WTO in 2001 (Faust & Kornberg 1995). Increasing demand for political cohesion was also evident in the 2008 anniversary of the original Treaty of Sino-Japanese Peace and Friendship. The Japanese PM Taro explicitly advocated building on the increasing diplomatic political cohesion within Sino-Japanese relations, as a result of the existing social and economic gains which both nations derived from their economic partnership. In doing so the PM also stated: “We should not constrain ourselves in the name of friendship between Japan and China. Rather, sound competition and active cooperation will constitute a true mutually beneficial relationship based on common strategic interests. Confucius said, ‘At thirty, I stood firm.’ In the same way, Japan and China must now stand atop the international stage and work to spread to the rest of the world this spirit of benefiting together” (Aso 2008).

Therefore, the market competition and cross-border diplomatic cooperation stated by PM Aso, shows that increased Japanese trade with China, which eventually spurred FDI flows in the Chinese mainland, led to increased social and political cohesion at the sub-state and diplomatic levels. Conclusion This paper argues that Japan’s trade with China in the late 20th and early 21st century led to increasing social and political cohesion between the two nations. This paper has been conducted specifically within an economic context, and therefore an allusion to the increased social and political cohesion can only be made in a one-dimensional manner. Nevertheless, the Sino-Japanese social-political relationship is still currently developing. Additionally, the historical weight and contextualization this paper has provided supports the neo-liberal claim. The increasing tensions in the Sino-Japanese political Tenth Anniversary | 33

Amarvir Singh-Bal

relationships have not derailed, but rather, have increasingly become dominated by the economic relationship. Trade, and the resultant FDI, increased exponentially throughout the period, and this improved social cohesion. Despite the political differences and resulting difficulties between both governments, each has realized that their need for one. The relocation of Japanese exporting firms within SEZs allowed citizens of both nations to build on their existing shared cultural heritage. This undermines the neo-realist claim that socio-political interactions drove commerce in Sino-Japanese relations, and not vice versa. REFERENCES Armstrong, S., ‘Japanese FDI in China: Determinants and Performance’ in Asia Pacific Economic Papers (2009), 378 (1). Armstrong, S.,‘Economics Still Trumps Politics Between Japan and China in Kotusai Mondai: (International Affairs) (2014), 634 (1) Aso, T., ‘My Personal Conviction regarding Japan-China Relations’ in At the Reception to Commemorate the Thirtieth Anniversary of the Conclusion of the Treaty of Peace and Friendship Between Japan and the People’s Republic of China inGreat Hall of the People, Beijing, China (October 24, 2008) at http://www.mofa.go.jp/region/asia-paci/ china/address0810.html. Austin, G. and Harris, S., Japan and Greater China: political economy and military power in the Asian century (London: Hurst and Company, 1999). Baoguo, C., Comment: Japanese TV Drama in China in ‘19th JAMCO Online International Symposium’ (2010) at: http://www.jamco.or.jp/en/ symposium/19/6/. Bende-Nabende, A., FDI, Regionalism, Government Policy and Endogenous Growth. A comparative Study of the Asean-5 Economies, with Development Policy Implications for the Least Developed Countries (Aldershot: Puffins, 1999). Brandt, L., Rawski, T. And Zhu, X., ‘International dimensions of China’s long boom: trends, prospects and implications’, in China and the Balance of Influence in Asia, (Pittsburgh: University of Pittsburgh, 2007). Burns, K.G., ‘China and Japan Economic Partnership and Politics Ends’ in Economic Confidence-Building and Regional Security (Washington DC: Stimson Centre, 2000), pp. 27–58. C. Chulai, Foreign Direct Investment and Trade: an Empirical Analysis of the Evidence from China’, Chinese Economics Research Centre (1997).

34 | Columbia Economic Review

Trade Interdependence and Peace

Chunlai, C., ‘Foreign Direct Investment and Trade: an Empirical Analysis of the Evidence from China’, in Chinese Economics Research Centre (1997) at https://www.researchgate.net/publication/23736875_Foreign_Direct_ Investment_and_Trade_An_Empirical_Investigation_of_the_Evidence_ from_China. Cross, K.H., ‘China’s WTO accession: economic, legal, and political implications’, in Boston College International and Comparative Law Review (2004), 27 (2). Deng, Y., ‘Chinese Relations with Japan: Implications for Asia-Pacific Regionalism’, in Pacific Affairs (1997), 70 (3), pp. 373–391. Drysdale, P.H., Huang, Y and Kalirajan, K.P.,‘China’s trade efficiency: measurement and determinants’, in Drysdale, P., Zhang, Y. and Song, L. (ed.), APEC and liberalisation of the Chinese economy, (Canberra: Asia Pacific Press, 2000). East Asia Analytical Unit, China Embraces the Market: Achievements, Constraints and Opportunities, (Canberra: Dept. of Foreign Affairs and Trade, 1997). Ernst, D., ‘Partners for the China Circle? The East Asian Production Networks of Japanese Electronics Firms,’ in Naughton, B. (ed.), The China Circle: Economic and Technology in the PRC, Taiwan, and Hong Kong (Washington, DC: Brookings Institution Press, 1997), pp. 210–254. Faust, J.R. and Kornberg, G., China in World Politics (London: Lynne Rienner Publishers, 1995). Geng, Y., Perspectives and Biases of Chinese and Japanese Youth on China-Japan Relations, in The Kiessling Papers: The Influence of Social Identity in Trudeau Centre for Peace and Conflict Studies (Toronto: Trudeau Centre, 2011). Hideo, O., ‘The Impact of China’s Rise on Sino-Japanese Economic Relations,’ in Rrosei, K., and Jisi, W. (ed.) (Tokyo: Japan Centre for International Exchange, 2004), pp. 175–193. Hilpert, H.G. and Nakagane, K.,‘Economic relations: what can we learn from trade and FDI?’, in Soderberg, M. (ed.), Chinese-Japanese Relations in the Twenty-first Century: Complementary and Conflict (London: Routledge, 2002). Hilpert, H.G., ‘China and Japan: Conflict or Cooperation? What does Trade Data Say’, in Hilpert, H.G. and Haak, R. Japan and China: Cooperation, Competition and Conflict (New York: Palgrave Macmillan, 2002). Hirschman, A.O., ‘National Power and the Structure of Foreign Trade’ in Political Science Quarterly (1946), 61 (2).

Tenth Anniversary | 35

Amarvir Singh-Bal

http://www.fxtop.com ‘Exchange Rate Historical Data’ at http://fxtop.com/ en/historical-exchange-rates.php?A=1&C1=JPY&C2=CNY&YA=1&D D1=01&MM1=01&YYYY1=1970&B=1&P=&I=1&DD2=16&M M2=01&YYYY2=2017&btnOK=Go%21. International Monetary Fund, ‘China-Japan Bilateral Trade Statistics’ (2017) at https://www.imf.org/en/Data. International Monetary Fund, at ‘investment flows’ https://www.imf.org/en/ Data. Jain, Purnendra, ‘Forging New Bilateral Relations: Japan-China Cooperation at the Subnational Level’, in Lam Peng Er (ed.), Japan’s Relations with China: Facing a Rising Power (London: New York, Routledge. 2006). Japan Bank for International Cooperation Institute (JBICI) Review, 1996–2016, at http://www.jbic.go.jp/english/research/report/review/ index/php. Japan Foundation, ‘Survey report on Japanese language education abroad’ (2008) at http://jpf.go.jp/e/japanese/survey/result/index.html (2008). JATA ‘Kaigai ryokosha no ryokosaki (ukeirekuni tokei) (Foreign Travellers’ Destinations (receiving countries statistics))’ (2007) at http://www. jata-net.or.jp/tokei/004/2007/05.htm. JATA ‘Kuni chiikibetsu honichi gaikokujinsu no suii (Change in the number of visitors to Japan by country and region)’ (2007) at http://www. jata-net.or.jp/tokei/004/10.htm. Johnston, C., (1986), ‘The Pattern of Japanese Relations With China, 1952–1986’, in Pacific Affairs (1986), 59 (3), pp. 402–428. Keji ribaor, ‘Japanese Business Boost Social Enterprise’ (15 Nov. 1994). Keshk, O.M., Pollins, B. M., & Reuveny, R., (2004), ‘Trade still follows the flag: The primacy of politics in a simultaneous model of interdependence and armed conflict’ in Journal of Politics (2004), 66 (4), pp. 1155– 1179. Knickerbocker, F.T., Oligopolistic Reaction and the Multinational Enterprise (Cambridge, MA,: Harvard University Press, 1973). Koizumi, J., “Asia in a New Century – Challenge and Opportunity” Boao Forum for Asia (Boao, Hainan Island, People’s Republic of China, 2002) http://www.mofa.go.jp/region/asia-paci/china/boao0204/speech.html. Kwan, C.H.,‘The Rise of China as an Economic Power: Implications for Asia and Japan’, in Hilpert, H.G. and Haak, R. (eds.) Japan and China: Cooperation, Competition and Conflict (New York, Palgrave Macmillan, 2002). Lang, N.A., Intercultural management in China: strategies of Sino-European and Sino-Japanese joint ventures (Wiesbaden: Deutscher Universitäts-Verlag, 1998).

36 | Columbia Economic Review

Trade Interdependence and Peace

Liu, S., ‘Official: National Industry, FDI Insuperable’, China Daily (Business Weekly) (1997). Mori, K., Japan-China Relations: from Post-War to a New Era (Tokyo: Iwanami Hoten, 2006). Nakagane, K., ‘Japanese Direct Investment in China: It’s Effect on China’s Economic Development’, in Hilpert, H.G. and Haak, R. (eds.) Japan and China: Cooperation, Competition and Conflict (New York: Palgrave Macmillan, 2002). Naughton, B., (2008), ‘A Political Economy of China’s Economic Transition in China’s Great Transformation’, in Brandt, L.; Rawski, G. (ed.), China’s Great Transformation, (Cambridge: Cambridge University Press). Naughton, B.,‘A Political Economy of China’s Economic Transition in China’s Great Transformation’, in Brandt, L.; Rawski, G. (ed.), China’s Great Transformation, (Cambridge: Cambridge University Press, 2008), pp. 91–135. News Release, ‘The outlook of Japanese foreign direct investment: Exim Japan 1994 survey’, in The Export–Import Bank of Japan (21 December 1994). Polacheck, S. W., ‘How trade affects international interactions’ in Economics of Peace & Security Journal (2007), 2 (2), pp. 737-761. Pollins, B.M, (1989), ‘Conflict, Cooperation, and Commerce: The Effect of International Political Interactions on Bilateral Trade Flows’ in American Journal of Political Science (1989), 33 (3), pp. 737–761. Research Team on Environment and Policy for Foreign Investment in China, ‘Comparison of Cases of Some Multinational Companies’, in Foreign Investment in China, vol. 80 (7) (2000). Rosen, D., Behind the Open Door: Foreign Enterprises in the Chinese Market Place (Washington DC: Institute of International Economics, 1999). Statistical Year Book of China, ‘Statistical Database’, at http://www.stats.gov.cn/ ENGLISH/Statisticaldata/AnnualData/. Stoltenberg, C.D., (1984), ‘China’s Special Economic Zones’, Asian Survey XXXIX (1984), 6 (1), pp. 637–54. Treaty of Peace and Friendship Between Japan and The People’s Republic of China (Aug. 12 1978), Article 1 at http://www.mofa.go.jp/region/ asia-paci/china/treaty78.html. Vencalek. T.E., (2015), ‘Hot Economics, Cold Politics: the Influence of anti-Japanese protests on Japanese Foreign Direct Investment in China’, in KU Scholar Works (University of Kansas, 2015), pp. 1–47.

Tenth Anniversary | 37

Amarvir Singh-Bal

Vyas, U., Soft Power in Japan-China Relations: State, sub-state and non-state relations (Wiltshire: Routledge, 2011). Wakasugi, R.,’Japan’s Trade and Investment Policies towards Asian Countries’, Paper to an international workshop on Japan and China in the Asia Pacific Region (Canberra: The Southeast Asia Dimension, 1996). Waltz, K., Theory of International Politics, (Long Grove: Waveland Press, 2010). Wang, M., The Truth is that Chinese People Admire Japan (Tokyo: PHP Kenkyusho, 2005). Wang, Z. and Swain, N., (1995), ‘The Determinants of Foreign Direct Investment in Transforming Economies: Empirical Evidence from Hungary and China’, in Weltirtschaftliches Archive (1995), 131 (1), pp. 359–82 . Whiting, A.S.,(1992), ‘China and Japan: Politics versus Economics’ in The Annals of the American Academy of Political and Social Science: China’s Foreign Relations (1992), 519 (1), pp. 39–51. Xing, Y. and Wan, G., (2006), ‘Exchange Rates and Competition for FDI in Asia’ in The World Economy (2006), 29 (4), pp. 419–434. Xinhua, ‘Japanese join with Local Authorities to Build City’ (5 March 1997). Xinxin, H, ‘Japanese firms in China what problems and difficulties are they facing?’ in Soderberg, M. (ed.), Chinese-Japanese Relations in the Twenty-first Century: Complementary and Conflict (London: Routledge, 2002). Xu Dixin, (1981) ‘China’s Special Economic Zones,’ Beijing Review (1981), 24 (50). Zhang, D.D., China’s Relations With Japan in an Era of Economics Liberalisation (Commack, New York: Nova Science Publishers, 1998).

38 | Columbia Economic Review

REAL EXCHANGE RATE VOLATILITY AND ECONOMIC GROWTH: A PANEL DATA INVESTIGATION

Federico Pessina University of Warwick Abstract: The study aims to investigate the impact of REER volatility on economic growth for a set of 33 developed and developing economies, using panel dataset ranging from 1970 to 2016. Stemming from a precise measure for exchange rate volatility, results of various Fixed Effects and System GMM models suggest that increased (decreased) Real Effective Exchange Rate (REER) volatility, controlling for trade and misalignment and contingent on diverse model specifications, leads to a negative (positive) effect on economic growth for developing countries. A relationship cannot be ascertained for developed countries. In addition, a significant impact of the REER level and its interaction with volatility is found, whilst neither a significant interaction of volatility with trade nor terms of trade shocks is found.

he choice of exchange rate regimes is one of the most debated variables within macroeconomic policy. Exchange rate regimes, volatility and misalignment of the exchange rate are inherently interlinked (Eichengreen 2007). A growing body of literature, focused in particular on RER (real exchange rate) misalignment, has developed to investigate their effects on economic growth. Whilst misalignment is a critical aspect, I argue that RER volatility, which causes uncertainty in trade, should be equally considered given its potential effects on growth. Theoretical and empirical literature regarding the advantages and disadvantages of different regimes is generally inconclusive, as the proponents of diverse systems indicate contrasting effects of volatility on macroeconomic variables. In the same manner, literature has not been able to ascertain a relationship between volatility and growth or its components (Eichengreen 2007). As noted by Vieira et al. (2013), utilisation of standard deviation as a measure of volatility implies clear limitations, which will be explored further on. Inconsistent methodologies in RER estimation have also led to diverse results. Literature on exchange rate misalignment has typically been a divisive issue, due to measurement inaccuracy caused by the necessity of estimating an equilibrium exchange rate, and the contrast between macroeconomic instability and outward-oriented growth strategies, both potentially caused by misalignment in the exchange rate. Tenth Anniversary | 39

Federico Pessina

The motivation of this paper is to complement these studies with an investigation of the effects of Real Effective Exchange Rate (REER) volatility on economic growth and its components. I draw from previous results the model specification and control variables. In addition, the paper strives to go beyond previous literature in various other manners. Firstly, I utilize REER instead of RER, a broad measure of the countryâ&#x20AC;&#x2122;s competitiveness. I use monthly figures instead of annual, which are calculated with a consistent methodology for the whole panel, implying a higher level of preciseness when modelling annual REER volatility. Secondly, I overcome the limitations of using standard deviation as measure of volatility by modelling it with a GARCH process. Thirdly, I focus on the relationship between REER volatility, trade and terms of trade (TOT). Differentiating whether the economy is developing or not, I investigate whether the impact of volatility is more pronounced if trade represents a larger share of the economy and whether it is experiencing any TOT shocks. The paper is based on a panel dataset ranging from 1970 to 2016 for 33 different countries, concentrating therefore on the post-Bretton Woods period. Stemming from a precise measure for exchange rate volatility, results of various Fixed Effects, Between-Within and System GMM Models suggest that increased (decreased) REER volatility, controlling for trade and misalignment and contingent on diverse model specifications, leads to a negative (positive) effect on economic growth for developing countries. A relationship cannot be ascertained for developed countries. A significant impact of the REER level and its interaction with volatility is found, whilst no significant interaction of volatility neither with trade nor TOT shocks is found. Exchange Rate Volatility and Economic Growth Following the fall of the Bretton-Woods system, exchange rate volatility has increasingly become a topic of interest given its potentially critical implications for economic policy. Theoretical and empirical literature has attempted to investigate the relationship between exchange rate volatility and economic growth, augmenting it with further research regarding the volatilityâ&#x20AC;&#x2122;s effects on other macroeconomic indicators such as trade, investment and so on. Results of both theoretical and empirical studies have been inconclusive due to contrasting results. As noted by Eichengreen (2007), these disagreements are generally caused by diverse model specifications and control variables, high dependency on the sample period considered and, to conclude, inconsistent and simplistic methodologies when measuring exchange rate volatility. I will focus on literature regarding economic growth, which is directly relevant to my paper.

40 | Columbia Economic Review

RER Volatility and Economic Growth

Levy-Yeyati & Sturzenegger (2002) indicate how the relationship between growth and exchange rate volatility has traditionally been disregarded by economists due to the classical money neutrality argument: nominal variables do not affect the long-run equilibrium growth of an economy. On the other hand, it is accepted that the relationship described above is contingent upon the type of exchange rate regime adopted (Eichengreen 2007). The debate on the advantages and disadvantages of different exchange rate regimes provides useful insights. Advocates of flexible exchange rate argue that, in cases of real shocks on the economy, flexibility facilitates the adjustment process (Levy-Yeyati & Edwards 2004, Friedman 1953). Ghosh et al. argue that fixed exchange rate may create RER misalignment, leading to erroneous price signals and thus inefficient allocation of resources. Alternatively, supporters of fixed exchange rates suggest that the stability of the regime promotes trade and investment, through a decrease in uncertainty of prices and real interest rates that firms face, improving the efficiency of price mechanisms on the international scale: this will incentivise economic growth (Mundell 1973, De Grauwe 1998). Similarly, empirical literature has been inconclusive. For example, Ghosh (1997) analysed the links between exchange rate regimes, inflation and growth utilising data on 136 countries over a period of 30 years. The study does not find any statistically significant effect of RER volatility on economic growth but highlights the importance of controlling for trade and investment when isolating the effects of exchange rate stability on growth. Flexible exchange rates are subject to market forces, theoretically rendering misalignment less possible, and enhancing the development of export industries which, due to stronger competition, are associated with higher productivity growth. The paperâ&#x20AC;&#x2122;s focus on the nominal exchange rate, whilst aiming to assess changes in long-run economic growth, has been criticised (Eichengreen 2007). Developing countries may have different priorities when assessing exchange rates. Bleaney & Greenaway (2001) investigate the impact of RER volatility on investment and growth on a panel of 14 sub-Saharan African coun tries over 1980-1995. They find a significant negative effect on investment but not on economic growth. The paperâ&#x20AC;&#x2122;s emphasis on developing countries provides useful insights: their dependence on the export of primary products is based on the idea that RER volatility should particularly affect economies relying on specialisation in primary products. Vieira et al. (2013) investigate the impact of RER volatility on long-run economic growth for a set of 82 advanced and developing economies, using a panel dataset ranging from 1970 to 2009. They find a significant negative impact of RER on long-run economic growth, even when controlling for country-specific idiosyncrasies. A further interesting result is provided by the paper: when including volatility, level of exchange rate nor the misalignment Tenth Anniversary | 41

Federico Pessina

are statistically significant. The authors thus suggest that pursuing a stable exchange rate may be a better strategy than exchange rate misalignment when aiming for long-run economic growth. This conclusion ties with the issues in literature regarding exchange rate misalignment briefly highlighted in Section I. Similarly, Dollar (1992), focusing on 95 developing economies, finds a negative impact of exchange rate volatility on economic growth. The author includes degree of openness as control variable, finding a relationship between it and the impact of volatility on growth. Literature has aimed to investigate the relation by looking at the components of economic growth, in particular trade: no conclusive evidence has been found on both theoretical and empirical level (Vieira et. al. 2013, Franke 1991, Peree & Steinherr 1989). Whilst exchange rates movements are assumed to be random in the short run, literature recognises that RER does depend on macroeconomic variables in the medium and long run. TOT shocks may drive RER changes, as changes in export and/or import prices will affect nominal exchange rates, inflation and trade (Coudert et. al. 2008). Overall, Eichengreen (2007) provides an insightful summary of literature by arguing that whilst the empirical results have not been conclusive, ensuring a stable, competitive and aligned RER may not directly cause long-run economic growth but will allow a country the possibility of exploring their economic potential thanks to, amongst other reasons, reduced uncertainty and higher competitiveness. In my paper, I aim to build upon and fill gaps in the literature addressed above. As mentioned in the introduction, the contribution of this paper rests upon tackling several criticisms of the papers reviewed. Conflicting empirical results are caused by contrasting model specifications, high dependency on the sample period considered, and naĂŻve volatility measures (Eichengreen 2007). Following Vieira et al. (2013) and Ghosh (1997), the model will be based upon the key macroeconomic indicators and components of GDP. In addition, I utilise trade as a proxy for degree of openness and investigate its relationship with exchange rate volatility, hypothesising that the impacts of volatility increase together with trade as a share of GDP. TOT has typically been disregarded in the RER-volatility literature. The paper posits a larger impact of volatility when the economy is experiencing large TOT shocks, which are assumed to be exogenous (Pryor 1966). The paper will focus on RER and disregard its nominal counterpart, in an attempt to isolate its effects on economic growth. In contrast to Vieira et al. (2013), economic growth will not be modelled around 5-year averages. The model will instead be augmented with lagged trade and other indicators, as further investigation of the interaction between trade and volatility, as found by Dollar (1992). To conclude, the use of consistent methodology when calculating REER and its volatility, together with the focus on the post-Bretton Woods 42 | Columbia Economic Review

RER Volatility and Economic Growth

Woods period and the inclusion of a large number of countries, aims to reduce differences between papers, leading to comparable results and higher external validity. Modelling REER Volatility In this paper, I utilise REER instead of RER. REER measures the “development of the real value of the country’s currency against the basket of the trading partners of the country” (Bruegel 2017). Alternatively, the REER can be defined as the average of the bilateral RER between the country in question and its trading partners, weighted by the respective trade shares of each partner (IMF 2007). Intuitively, the advantages of REER are clear. Changes in REER provide useful insights for this analysis, indicating the equilibrium value of the currency, thus permitting us to estimate, albeit simplistically, the level of misalignment, and changes in competitiveness of the whole economy, driven by either change in exchanges rates or in relative inflation (price or cost competitiveness). REER permits the isolation of the competitiveness of the single country, reducing the noise of the other countries. This paper models REER volatility using the dataset “Real Effective Exchange Rates for 178 Countries” developed by Bruegel. The dataset includes both annual and monthly data for a large number of countries. I will utilise monthly REER data from 1970 to 2016 for 33 countries (Appendix 1). Modelling REER volatility is one of the causes of contrasting results in literature (Eichengreen 2007, Siregar & Rajan 2004). A contribution of this paper is to use REER calculated with consistent methodology and, consecutively, utilise an innovative measure for volatility coherent with the papers reviewed previously.

Literature has modelled volatility through standard deviation, therefore looking at how the exchange rate fluctuates around its mean overtime (Schnabl 2007). Utilising standard deviation as measure of volatility is approximate. Firstly, standard deviation has a skewed distribution. Secondly, standard deviation doesn’t capture the information in the previous periods (Jansen 1989). Exchange rates are determined, to an extent, by a random process; utilising standard deviation leads to ignoring the level of volatility in the previous period, Tenth Anniversary | 43

Federico Pessina

implying that volatility in one period is unaffected by volatility in the previous one. The paper proceeds by utilising Generalised Autoregressive Conditional Heteroskedasticity (GARCH) process, developed by Bollerslev (1986). As noted by McKenzie (1999), exchange rates best follow the GARCH process; GARCH, compared to ARCH, captures the past values and thus permits higher accuracy. I model the variance w_t following a GARCH (1, 1) process on the monthly REER data, for each country, from 1970 to 2016.

The method presents several advantages compared to SD. As mentioned, I am able to capture the effects of volatility of one period on the successive one, differentiating between the random and the predictable elements of the exchange rate determination process (Ebaidalla 2013, Arize et. al. 2000). To conclude, variance is conditional on the past values, better reflecting how volatility is perceived by agents. After obtaining monthly figures for REER volatility, I average them to obtain annual volatility figures for each country from 1970 to 2016. This is done to match the availability of data for macroeconomic indicators. Final output is 1551 observations in total. Descriptive statistics are shown in Table 2.

Empirical Strategy The aim of this paper is to estimate the effect of REER volatility on economic growth, commencing by following this initial model:

The choice of control variables follows growth literature and the papers previously reviewed. (Eichengreen 2007, Vieira et. al. 2013). Trade and government expenditure are denoted as share of GDP. Theoretical literature suggests that inflation may have negative effect on GDP growth due to distortion in price signals; government expenditure is ambiguous, as some argue

44 | Columbia Economic Review

RER Volatility and Economic Growth

that it may increase aggregate demand (Keynesian Economics), whilst others remind of its crowding-out effect, which leads to lower investment and thus growth in the long run; trade is expected to increase GDP growth. Inflation is used as a proxy for macroeconomic stability, whilst trade represents degree of openness. In conclusion, developing and developed countries will be analysed separately (Appendix 1), leading to more insightful results, and potentially limiting endogeneity issues by analysing countries with similar macroeconomic conditions together.

Further models include:

1) TOT Shocks, and an interaction variable between REER volatility, trade and TOT shocks. The paper hypothesizes that the impact of volatility may be different when the country is experiencing TOT shocks, given that they drive RER, controlling at the same time for the importance of trade in the economy. A separate interaction variable between trade and REER volatility is investigated, hypothesising that impact on growth is larger when trade represents a larger share of GDP. 2) Interaction terms between volatility and GDP per capita; literature suggests that the higher the GDP, the lower the GDP growth. Thus, I control for the variable and investigate any additional effects of volatility. 3) Dummy variables to control for possible regional differences. 4) I also posit whether lagged independent variables such as government expenditure and trade may explain variation in GDP Growth. Given that I am looking at annual data, theoretical literature suggests that government expenditure may have a multiplier effect. Whilst this is not the aim of the paper, including such terms may be insightful. 5) Diverse functional forms, such as including lagged or squared terms of the independent variable REER volatility. 6) REER misalignment from its equilibrium value.

The model proposed clearly holds endogeneity issues, where some of the variables are determined within the model. One form of endogeneity present is omitted variable bias, inherent to the macroeconomic nature of the model, and indicated by the existence of inconclusive empirical literature. Omitted variables leads to OLS bias, implying larger coefficients and lower standard error for the variables included in the model, and thus wrongful conclusions. In this case, the direction of the bias on coefficients is ambiguous. Firstly, the paper attempts to reduce the bias by estimating various fixed effects models. Fixed Effects exploits within-group variation. The model involves controlling for the average difference across the countries in any observable/unobservable predictors, which are assumed to be time-invariant. Therefore, threat of omitted variable bias is reduced. As a price, between-group variation is lost. The paperâ&#x20AC;&#x2122;s main variables of interest are not time-invariant; thus, the model is useful. Choice of Fixed Effects model rests upon data Tenth Anniversary | 45

Federico Pessina

data consistent with the theory and assumptions needed to implement it; in addition, Hausman tests confirms its suitability (Appendix 3). Tests are implemented, which confirm the presence of autocorrelation, cross-sectional dependence and heteroskedasticity. Their presence was highly predictable given the nature of the data and variables utilised. Dependent variable may be affect by unexpected variation or shocks. The model thus controls for time-fixed effects, further reducing omitted variable bias and cross-sectional dependence (Appendix 3). Following the tests, I proceed by utilising Driscoll and Kray errors (Driscoll & Kray 1997). Secondly, I estimate various within-between effects model, in an attempt to investigate continent dummies, as they’re time invariant and cannot be discerned in the fixed-effects model, and between-group variation. As mentioned, results from the Fixed Effects model are assumed to be more robust to endogeneity issues. Thirdly, the paper investigates dynamic panel data models, through system Generalised Method of Moments (GMM), as developed by Arellano-Bover (1995)/Blundell-Bond (1998), as a further potential solution to endogeneity issue. In this technique, I utilise lagged levels (differences) of the endogenous variables as instruments for the regression in differences (in levels), together with the already-specified exogenous instruments (Vieira et. al. 2013, Roodman 2009a). Two empirical methods will be employed to reduce the number of instruments. Firstly, the collapse option of the command xtabond2 will be used. Secondly, restrictive lag limits will be imposed. An excessive number of instruments would overfit the model, leading to wrongful conclusions and a still present endogeinety issue (Roodman 2009b, Windmeijer 2005). Time-fixed effects (year-dummies) are included to reduce autocorrelation, logically present in macroeconomic data, and miitigate the effects of any trends, although non have been discerned through data inspection. In conclusion, GMM should lead to more robust estimates of the coefficients, and permit the investigation of the dynamic relationship between the variables. Data Sources Data for all macroeconomic variables has been taken from World Development Indicators (2017) dataset, developed by the World Bank (National Accounts data). REER data has been taken from the dataset “Real Effective Exchange Rates for 178 Countries” developed by Bruegel (2017). Descriptive Statistics Descriptive statistics have been reported in Table 1. Inspection of the data suggests the existence of an outlier: Bolivia’s inflation with a reported

46 | Columbia Economic Review

RER Volatility and Economic Growth

value of 11749%. The paper proceeds without the elimination of the data point, given the potentially useful information it contains. Logarithmic transformation of variables reduces the effect of the outlier, and a sensitivity analysis is completed to ensure the validity of the results. Overall, the statistics show broad results and diverse countries on the full spectrum of economic development. Detailed information regarding REER volatility has been covered in â&#x20AC;&#x153;Modelling REER Volatility.â&#x20AC;?

Data is compiled together to obtain a balanced dataset, with no missing observations, comprising annual data from 1970 to 2016 for 33 countries. Definition of all variables can be found in Appendix 2. Annual data permits such a large time span; quarterly data would have excessively restricted the years and variables available, potentially not permitting an efficient control of eventual shocks and trends in the data. The study is focused on the post-Bretton Woods period, with the emergence of floating exchange rates. Quick inspection of the data highlights how developing countries experienced on average higher REER Volatility, GDP growth rates and inflation. Tenth Anniversary | 47

Federico Pessina

Developing countries tend to show larger dispersion in their values, alluding to their more instable macroeconomic situations. Higher GDP growth rates are expected due to convergence. To permit an effective investigation, the paper controls for this effect, for example through the division of the analysis between developing and developed countries. A large range of years permits to mitigate the business-cycle effects and reduce the potential influence of them on volatility in the exchange rates. Further insights have been explored in â&#x20AC;&#x153;Empirical Strategy.â&#x20AC;?

Empirical Results Table 3 reports the main results. The empirical strategy involves estimating the model highlighted in Section IV, and then proceed by investigating various interaction variables and control variables, such as the level of REER, the relationship between macroeconomic stability and volatility and an interaction variable between GDP level and volatility. The results are confirmed throughout all models. All regressions, for developing countries, indicate significant, negative coefficients for REER Volatility on economic growth. A significant squared term has been estimated with fixed effects, suggesting a concave relationship for developing countries, although successive models were not able to confirm this. Regarding developed countries, the results have been inconclusive, and no relationship can be ascertained. System-GMM indicates that a 1% increase in REER volatility leads to around 0.63 percentage-points fall in GDP growth rate. Whilst coefficients vary between models, thus a specific value should not be concluded, the negative direction of the effect has been confirmed by various model specifications.

48 | Columbia Economic Review

RER Volatility and Economic Growth

Progressing through the estimators, it can be seen that macroeconomic controls central to growth theory, such as government expenditure and inflation, become increasingly significant in explaining economic growth. I expect such results due to the estimators increased ability to control for endogeneity, in particular omitted variable bias in this case. Given the limited number of variables used, and a general risk of endogeneity when explaining the economy on a macroeconomic scale, the use of more robust estimators is thus essential to ensure appropriate results.

Tenth Anniversary | 49

Federico Pessina

Analysis of interaction and control variables provides further insights. No significance is found for the relationship between volatility and trade and TOT shocks, where the paper had hypothesised a potential increased impact of volatility. A positive impact of the interaction variable between the level of REER and REER volatility is found for developing countries, whilst the more robust GMM estimators evidence a further significant impact of the REER level. Similar result is found for developed countries with the GMM estimators, although the study interprets these results as less robust, unless its limitations are addressed. This suggests the importance of a REER both competitive and stable. In conclusion, contrasting results have been found for REER misalignment; the study though did not investigate misalignment fully, seen as a separate issue from volatility, and thus no concrete conclusion should be inferred. Regarding the control variables, results show a high degree of robustness, with general significance and coefficients of the same sign. Trade, the proxy for degree of openness, has a positive impact on economic growth, with significance always found for developed countries. Inflation, the proxy for macroeconomic stability, has instead a negative impact. Government expenditure is, by definition, a divisive issue in literature. Overall, the robustness of the control variables indicates a generally efficient model. Results of the paper follow what has been found by papers with similar methodologies, although some stark differences remain. In contrast to Vieira et. al (2013), for example, the REER has found to be generally significant with a positive effect on economic growth. This paper posits that it may be over-simplistic to discard the variable and that instead, most importantly, the coexistence of a competitive and, at the same time, stable REER is a powerful determinant of economic growth. This accords with a general summary of the literature, as noted by Eichengreen (2007). Endogeneity is still present and addressed in limitations and extensions. In contrast to the R2,where high values were not expected due to the macroeconomic nature of the data, the diagnostic testing shows a general validity of the results, with normally distributed errors (Appendix 5). The robustness of the control variables is another positive indication. Regarding the dynamic estimators, the inclusion of time-fixed dummies permits the rejection of the AR(2) test, highlighting no problems of second-order autocorrelation. The Hansen over-identifications tests are generally positive, suggesting the validity of the instruments. In some models, I have an instrument-proliferation problem, caused by limitations in the data. Conclusion, Limitations and Extensions Literature has not reached a consensus regarding the role of REER/RER volatility on economic growth. The paper aimed to provide further evidence

50 | Columbia Economic Review

RER Volatility and Economic Growth

through the use of a consistent and innovative methodology to measure volatility, and an investigation of the interaction of volatility with trade and TOT shocks. The study is an empirical investigation of the relationship between REER volatility and economic growth on a panel dataset of 33 countries from 1970 to 2016. Overall, after controlling for endogeneity and countryspecific characteristics through Fixed Effects and GMM estimators, results indicate the existence of a negative impact of REER volatility on economic growth for developing countries, whilst no effect is evidenced for developed countries. No significance is found for the effect of TOT shocks and degree of openness on the impact of volatility on economic growth, whilst a significant positive effect of the level of REER is found. Results confirm the attention exchange rates given to them by literature. The choice of exchange rate regime has been a critical decision of governments and policymakers when attempting to promote prolonged and sustainable economic growth. The results thus suggest policy-wise that maintaining low REER volatility, together with a competitive REER, should be beneficial for economic growth in the case of developing countries. Advising on the choice of regime is beyond the scope of the study; whilst a fixed regime may, by definition, have the lowest volatility, it may not permit the existence of a competitive REER or be unsustainable, thus being overall detrimental. Managed floating exchange rate, for example, may be a viable alternative. The limitations of the study stem primarily from the data. In the data collection process, there was a trade-off between the inclusion of countries and variables with missing observations, or their exclusion in order to maintain a balanced panel data. The study has followed the second road. A possible extension of the research is, therefore, the inclusion of new variables and countries; in such a case, in order to maintain a balanced panel dataset, the time range will be much more limited. On the other hand, if the time range is reduced, quarterly data could be used, which may evidence short-run dynamics and shocks which are lost on an annual scale or, alternatively, control more effectively for business cycle and trends in GDP growth. A second limitation due to sample size regards the methodology, where the paper was unable to utilise the more efficient two-style estimator, which gives asymptotic efficiency for system GMM. Thirdly, the channels through which volatility affects economic growth are not modelled. A large degree of endogeneity is still present. Further research should, therefore, aim to explore these limitations and most importantly, investigate the reasons why contrasting results have been found between developing and developed countries. Identification of the channels through which exchange rate volatility affects economic growth may suggest the causes of such findings. In addition, they may indicate more effective policy recommendations. Secondly, identifying the source of REER volatility could provide an insightful analysis in the dynamics of the relationship explored Tenth Anniversary | 51

Federico Pessina

by this paper. Thirdly, the potential inclusion of capital flows, current account, inflation differentials and interest rates may highlight the flow of capital across countries and enhance the terms-of-trade analysis of this paper which has been inconclusive.

REFERENCES Arellano, M., & Bond, S. (191). Some Tests of Specification for Panel Data: Monte Carlo Evidence and an Appliacton to Employment equations. Review of Economic Studies(58), 277-297. Arellano, M., & Bover, O. (1995). Another Look at the Instrumental-Variable Estimation of Error-Components Models. Journal of Econometrics(68), 29-51. Arize, A., Osang, T., & Slottje, D. (2000). Exchange Rate Volatility and Foreign Trade: Evidence from Thirtenn LDCs. Journal of Business and Economics Statistics, 18(1), 10-17. Bleaney, M., & Greenaway, D. (2001). The Impact of Terms of Trade and Real Exchange Rate Volatility on Investment and Growth in Sub-Saharan Africa. Journal of Development Economics, 65, 491-500. Blundell, R., & Bond, S. (1998). Initial Conditions and Moment Restrictions in Dynamic Panel Data Models. Journal of Econometrics(87), 115-143. Bollerslev, T. (1986). Generalised Autoregressive Conditional Heteroscedasticity. Journal of Econometrics(31), 307-327. Bruegel . (2017). Real effective exchange rates for 178 countries: a new database. Retrieved from Bruegel [Dataset]: http://bruegel.org/publications/data sets/real-effective-exchange-rates-for-178-countries-a-new-database/ Coudert, V., Couharde, C., & Mignon, V. (2008). Do Terms of Trade Drive Real Exchange Rates? Comparing Oil and Commodity Currencies. CEPII. De Grauwe, P. (1998). Exchange rate variability and the slowdown in growth of international trade. IMF Staff Papers 35(1), 63-84. Dollar, D. (1992). Outward Oriented Developing Countries Really Do Grow More Rapidly. Economic Development and Cultural Change(4), 523-554. Driscoll, J., & Kraay, A. (1997). Consistent Covariance Matrix Estimation with Spatially-Dependent Panel Data. Review of Economics and Statistics. Ebaidalla, M. E. (2013). Impact of Exchange Rate Volatility on Macroeconomic Performance in Sudan. Economic Research Forum. Eichengreen, B. (2007). The Real Exchange Rate and Economic Growth. University of California, Berkeley.

52 | Columbia Economic Review

RER Volatility and Economic Growth

Engle, R. (1982). Autoregressive Conditional Heteroscedasticity with Estimates of the United Kingdom Inflation. Econometrica(50), 987-1008. Franke, G. (1991). Exchange Rate Volatility and Internation Trading Strategy. Journal of International Money and Finance(10), 292-307. Friedman, M. (1953). The case for flexible exchange rates. Essays in Positive Economics. Ghosh, A. R. (1997). Does the Nominal Exchange Rate Regime Matter? Working paper 5874. NBER Working Paper Series. Ghosh, A., Ostry, J., Gulde, A., & Wolf, H. (1997). Does the Exchange Rate Regime Matter for Economic Growth? IMF Economic Issues(2), 1-19. IMF. (2007). Why Real Exchange Rates. Retrieved from IMF - Finance and Development: http://www.imf.org/external/pubs/ft/fandd/2007/09/ba sics.htm Jansen, D. (1989). Does Inflation Uncertainty Affect Output Growth? Further Evidence. The Federal Reserve of Bank of St.Louis Review, 43-54. Levy-Yeyati, E., & Edwards, S. (2004). Flexible exchange rates as shock absorbers. European Economic Review, 1-2. Levy-Yeyati, E., & Sturzenegger, F. (2002). To Float or to Fix: Evidence on the Impact of Exchange Rate Regimes on Growth. American Economic Review. McKenzie, M. D. (1999). The Impact of Exchange Rate Volatility on International Trade Flows. Journal of Economic Surveys(13(1)), 71-106. Mundell, R. (1973). Uncommon arguments for common currencies. The Economics of Common Currencies. Peree, E., & Steinherr, A. (1989). Exchange Rate Uncertainty and Foreign Trade. European Economic Review(33), 1241-1264. Pryor, F. (1966). Economic Growth and the Terms of Trade. Oxford Economics Papers, 18(1), pp. 45-57. Roodman, D. (2009a). How to do xtabond2: An Introduction to Difference and System GMM in Stata. Stata Journal(9), 86-136. Roodman, D. (2009b). A Note on the Theme of too many Instruments. Oxford Bulletin of Economics and Statistics(71), 135-158. Schnabl, G. (2007). Exchange Rate Volatility and Growth in Small Open Economies at the EMU Periphery. Siregar, R., & Rajan , R. (2004). Impact of Exchange Rate Volatility on Indonesiaâ&#x20AC;&#x2122;s Trade Performance in the 1990s. Journal of the Japanese and International Economies(18), 218-240. The World Bank. (2017). World Development Indicators [2017] [Dataset]. Retrieved from http://data.worldbank.org/data-catalog/world-develop ment-indicators

Tenth Anniversary | 53

Federico Pessina

United Nations. (2014). Country Classification . Retrieved from http://www. un.org/en/development/desa/policy/wesp/wesp_current/2014wesp_ country_classification.pdf Vieira, F. V., Holland, M., Gomes da Silva , C., & Bottecchia, L. C. (2013). Growth and Exchange Rate Volatility: a Panel Data Analysis. Applied Economics(45), 3733â&#x20AC;&#x201C;3741. Windmeijer, F. (2005). A Finite Sample Correction for the Variance of Linear Efficient Two-step GMM Estimators. Journal of Econometrics, 126(1), 25-51. Zsolt, D. (2017). Real Effective Exchange Rate for 178 Countries. Bruegel.

54 | Columbia Economic Review

RER Volatility and Economic Growth

APPENDIX

Tenth Anniversary | 55

Federico Pessina

56 | Columbia Economic Review

RER Volatility and Economic Growth

Definitions taken by the official World Bank definitions, maintaining consistency with the source of our data.

Tenth Anniversary | 57

Federico Pessina

58 | Columbia Economic Review

RER Volatility and Economic Growth

Tenth Anniversary | 59

Federico Pessina

60 | Columbia Economic Review

HETEROGENOUS EFFECTS OF BOLSA FAMILIA ON RURAL AND URBAN HOUSEHOLDS

Elena M. Stacy University of California, Berkeley Abstract: I employ cross-sectional data from the Brazilian Consumer Expenditure Survey and inverse propensity score weighting to investigate the presence of heterogeneous effects on short term poverty alleviation of the Bolsa Familia program on rural and urban households. The results suggest that there is heterogeneity in favor of rural household’s quantity of food, and not in favor of rural household’s disposable income per capita. I discuss potential explanations for these findings and their implications.

he field of development and poverty alleviation is concerned with improving the lives of the world’s poor. In the modern day, there is constant debate over which methods are actually the most effective at doing so. A concept that has been more popular in recent decades is the conditional cash transfer (CCT). A CCT is a program that gives money to the poor on the condition that they do a specified activity which is believed to have positive effects on an outcome of interest. These conditions have typically consisted of activities such as children’s school attendance, vaccinations and health checkups, or nutritional goals. In addition to affecting the outcome of interest linked to the conditional activity, the act of giving money to the poor itself is supposed to provide short term poverty alleviation, and potentially have long term effects as well. Oportunidades/Progresa was the first of many of these governmentimplemented conditional cash transfer programs which targeted the poor. Its large success led to the uptake of similar programs across Latin America, including the country of focus for this paper, Brazil, which went on to launch the largest CCT program in the world (de Brauw et. al. 2015). The Bolsa Familia program gives cash transfers to families below the poverty line—in 2008, defined as those with an income of less than R$137 for “poor” and an income of less than R$69 for “extremely poor”—on the condition that their children attend school and get vaccines, as well as regular health check-ups. Those defined as ‘poor’ are eligible to receive a conditional monthly transfer of R$20 per child in the household, and those defined as ‘“extremely poor” are eligible to receive an unconditional monthly transfer of approximately R$69 in addition to the conditional per child transfer.

Tenth Anniversary | 61

Elena M. Stacy

Bolsa Familia has been lauded as one of the most successful CCT programs, and has been a model and inspiration for other government CCT programs. For this reason, the program has been extensively studied, and while much literature has found that Bolsa Familia has been successful in many ways, this paper seeks to answer the question of whether there is a presence of heterogeneous effects of households receiving Bolsa Familia transfers in urban regions as compared to rural regions on their short term poverty status, with the hypothesis that those in rural areas will have a negative bias due to lack of resources and infrastructure that their urban counterparts have access to. I will be making use of the Brazilian Consumer Expenditure Survey, also known as the Pesquisa de Orçamentos Familiares from the period 20082009. Since the program was not randomly assigned, I use propensity score weighting as a proxy for constructing a comparison group that is similar to the treatment group. This method has been used in much of the literature on Bolsa Familia and will be discussed more in subsequent sections. Following successful balancing of observable characteristics across the treatment and control groups, weighted least squares regression was used on regressions for the outcomes of interest with an interaction term on a dummy for rural/ urban to detect heterogeneity. Key results showed that there was significant positive bias towards rural households in sufficiency of quantity of food, while there was a significant negative bias against rural households in disposable income per capita, and no significant effects on ability to purchase preferred types of food. These results did not support my initial hypothesis, and I will present possible alternate explanations for this. The paper proceeds as follows. “Literature Review” provides an overview of literature that has both informed this study and will be relevant to the reader. “Data & Methodology” provides a detailed description of the dataset and methodology employed, including construction of treatment and comparison groups and identifying assumptions. “Analysis & Results” presents the results of the analysis and discusses them, as well as presenting alternate explanations for results. “Conclusion” discusses potential expansions and implications of these findings. Literature Review The first important paper that I will point out is “The Impact of Bolsa Família on Women’s Decision-Making Power” (de Brauw, et. al). While this paper is interested in a completely different outcome than this paper, it provides a detailed description of important parts of the design of Bolsa Familia targeting. This assists with the question of how to approach estimating treatment effects in the context of this program. De Brauw et. al point out that each municipality has a different maximum allowable number of Bolsa Familia recipients. That is, 62 | Columbia Economic Review

Heterogeneous Effects of Bolsa Familia

in some municipalities, there may be more eligible people for Bolsa Familia than can actually be allowed to receive it, for no other reason than the maximum number of recipients for that municipality being reached. This suggests that in many cases, those who are registered in the eligible beneficiary pool who did not receive Bolsa Familia benefits may be comparable to those who did receive benefits in observable and unobservable characteristics much of the time. To correct for this even further, De Brauw et. al. employ propensity score weighting to the comparison group in their data, which is said to give more precise estimates for potential differences in observed/unobserved differences across the two groups (Hirano and Imbens 2001, Keisuke et. al. 2003, Austin 2011, de Brauw et. al. 2014). This method has been widely used in Bolsa Familia research (de Brauw et. al. 2015, de Brauw et. al. 2014, Teixeira 2010). Another relevant paper, “Evaluating the Impact of Brazil’s Bolsa Familia” (Soares, Ribas, et.al.) is similarly looking at the effects of Bolsa Familia, but on the broad outcome of “poverty reduction”, more along the lines of my own question, but with a lack of interest in heterogeneity in rural vs. urban areas. An important piece of information that Soares and Ribas point out is that while Bolsa Familia has been widely criticized in the literature for using unverified means testing to determining eligibility, there is a cross-checking process that is carried out that corrects any significant differences between the federal database and what households report, since there is incentive to under-report one’s income in order to receive benefits (Soares, Ribas, et. al. 2010, Teixeira 2010). Despite this research, there is still evidence that households above the income threshold do in fact receive benefits, which undermines Soares’ and Ribas’ findings. A third paper of relevance that is interested particularly in heterogeneous effects of Bolsa Familia is, “Heterogeneity in Bolsa Familia Outcomes” (Barrientos, Debowicz, and Woolard 2016). This paper focuses specifically on heterogeneous effects, particularly across municipalities, and across boys and girls. The paper takes on a different methodological approach, however. Barrientos et. al. employ a quantile regression on municipalities to estimate these heterogeneous effects, allowing for a more detailed treatment of distributional effects across quantiles. Additionally, they estimate the effects of the program at the municipality level, arguing that assignment into the treatment group — receiving Bolsa Familia transfers — is exogenous at the municipal level, since treatment levels depend on pre-program poverty levels as pointed out by de Brauw et. al. I reference this paper to point out that while this is an ideal situation, the data I make use of do not have municipal level identifiers, and thus it is While the papers that I have referenced above are only a small sampling of the literature on Bolsa Familia effects, I have yet to come across any literature that focuses specifically on the poverty outcomes with particular focus on

Tenth Anniversary | 63

Elena M. Stacy

heterogeneity between rural and urban households. This specific focus is important for estimating whether some communities may have more benefit from cash than others, shifting the focus away from the outcomes of interest that the conditions lead to—education and health, namely—and moving it towards the effect of the cash itself. This may provide broader implications for both conditional and unconditional cash transfers, and the types of communities that will benefit from them the most. Data & Methodology I make use of the Brazilian Consumer Expenditure Survey, also known as the Pesquisa de Orçamentos Familiares from the period 2008-2009. This is a government survey conducted by the Brazilian Institute of Geography and Statistics (IBGE) every six to seven years(Anonymous). The survey consists of a detailed questionnaire on household and individual level characteristics such as education levels, race, sewage, household consumption and expenditures, and household income and government benefits. The data are cross-sectional, and thus we are only able to observe a single period of time. In 2008-2009, the sample consisted of 55,970 households, and a total of 190,159 individuals. The sampling was done in strata that consisted of geographical regions, and each region was defined as either rural or urban, which was constructed into the rural/urban dummy of interest for detection of heterogeneity. For the purposes of this paper, the sample was restricted to households that had a net income per capita of below R$200 per month, for whom the head of the household was not missing, income data was not missing, and only included individuals above 16 years old who report contributing to the household income—to avoid missing values for key variables. This left a sample of 27,733 observations. The reasoning behind choosing a sample up to and including households with monthly income per capita of R$200 is the nature of the targeting methods of the Bolsa Familia program (de Brauw et. al. 2014). In fact, although the threshold is R$137 for this sampling period, there are beneficiaries who report a household income per capita above this limit, because eligibility for Bolsa Familia is self-declared rather than means-tested, so households are able to underreport their income in order to be ‘eligible’ for the program, as discussed in Section II (Teixeira 2010, Soares, Ribas, et.al. 2010). I present a graph in Figure 1 to illustrate that while the likelihood of receiving Bolsa Familia benefits is significantly lower as income increases, the mean of treatment status is far from zero once the cutoff of R$137 is passed. Furthermore, the mean of treatment status is also not equal to 1 before the income cutoff is passed, indicating that many people are eligible but do not receive the program. This can be attributed to the beneficiary maximums that many municipalities have, which cause some eligible people to miss out on the program despite being 64 | Columbia Economic Review

Heterogeneous Effects of Bolsa Familia

effectively the same as other beneficiaries in every other way, also as discussed in Section II (de Brauw and others 2015, 303-316). For the reasons outlined above, using a regression discontinuity design is not feasible in the context of Bolsa Familia, because the eligibility cutoff is not effective enough in restricting treatment, although otherwise a discontinuity design would be an attractive strategy given that the data have detailed income information. There is also evidence that the â&#x20AC;&#x2DC;conditionalâ&#x20AC;&#x2122; part of Bolsa Familia is not treated strictly, as less than 2% of Bolsa Familia beneficiaries end in cases of termination from the program due to non-compliance, despite the fact that non-compliance occurs, so it is not a reasonable assumption that those who did not receive the program are likely to have been terminated from it (de Brauw et.al. 2014). This again returns us to the explanation of the municipality maximums as the main hypothesis for individuals below the eligibility cutoff not receiving treatment.

Figure 1: Monthly household per capita income before Bolsa Familia. Dots represent mean value of Bolsa Familia status of respective income bins, with reference lines at R$69 and R$137, the eligibility cutoffs for the basic unconditional transfer and the variable, per child conditional transfer, respectively.

Tenth Anniversary | 65

Elena M. Stacy

The outcome variables that I chose to investigate are: quantity of food, type of food, and disposable income per capita. Quantity of food is a variable that comes from a survey question on whether the household feels that they have enough food to eat on average, and takes on numerical values of (1) for never having enough food, (2) for sometimes having enough food and, (3) for always having enough food. Type of food is a variable that takes on numerical values of (1) for always having the desired types of food, (2) for sometimes having the desired types of food, and (3) for rarely having desired types of food. Disposable income per capita is a variable that takes into account a household’s income after taxes and other deductions, and is its effective spending capability. These outcome variables were chosen in light of the fact that my outcome of interest is “short term poverty alleviation”. Short term poverty alleviation can mean many different things, but in this case I chose to focus on basic needs being met (having enough food), the ability to not only meet basic needs but also to cater to preferences (type of food), and general spending capacity (disposable income per capita). Explanatory variables include: a treatment indicator that equals 1 if a household received Bolsa Familia in the reference period or 0 if it did not; rural/ urban indicator that equals 1 if a household is defined to be in a rural region; an interaction term defined as treatment*rural to capture heterogeneous effects; a large list of controls including for income, number of children, years of schooling, race, access to a credit card, access to a checking account, whether one is the account holder on either the credit card or checking accounts, whether the home has piped water, whether the household water comes from the government line, number of bedrooms in the house, material of the floor and walls, number of people living in the residence, number of families living in the residence, whether the household owns the residence, how long the household has been renting the residence, the type of rental contract the household has, individual literacy, and educational controls such as completion of last course, highest grade level completed, etc. which are either used in the logit model for constructing propensity scores, or in the analysis to act as controls in the regressions on outcome variables. The methodology used in this paper is inverse propensity score weighting. This method has been widely used in analysis of Bolsa Familia, as discussed in “Data & Methodology”. Due to the non-random nature of Bolsa Familia benefit distribution, we cannot simply use those who did not receive Bolsa Familia as a comparison group. Although some households may be effective comparison groups, as they could be eligible yet unable to receive benefits due to their municipality reaching the benefit quota, other households may not present effective comparisons, as they may have differences in either observable or unobservable characteristics. 66 | Columbia Economic Review

Heterogeneous Effects of Bolsa Familia

By using propensity scores as weights in the regression analysis, we can effectively weight those households with higher likelihoods of being treated more heavily, and weight those households with lower likelihoods of being treated less heavily. Households with similar characteristics will thus be given similar weights, and this allows an estimation of an unbiased comparison group (Keisuke et. al. 2003, Austin 2011). In order to use this method, assumptions that must hold include: sufficient overlap of propensity scores across the treatment and comparison groups, achievement of balance of key observable characteristics after applying propensity score weights, all propensity scores fall between 0 and 1, the propensity scores are relatively accurate representations of the real probability of treatment, and the average outcomes of those who received Bolsa Familia transfers would have been the same as those who did not receive the program. I believe that these assumptions were all reasonable in this context. I will later present the distribution of propensity scores across treatment and comparison groups, as well as balance of observables to support these assumptions in Analysis and Results, and as robustness checks that my propensity score model was specified accurately. I employ a weighted least squares regression on each outcome variable of interest. The final weight on this regression will be the propensity score weights described above, multiplied by the sampling weights used in the data. The construction of the propensity scores is as follows: observations in the treatment group receive a weight of 1/p, and observations in the comparison group will receive a weight of 1/(1-p) (Austin 2011). The basic regression for analysis will be: where Yhi is the outcome of interest for the given regression (either quantity of food, type of food, or the natural log of disposable income per capita), â&#x20AC;&#x153;hiâ&#x20AC;? are household and individual identifiers. The regression will start at the basic level, with nothing filling the ellipses after the interaction term, and will progressively have more controls added which will be specified in each regression table. The model for estimating the propensity scores will be a logit model: where each X represents an observable characteristic. The observables included in the logit model were: literacy, last grade completed, school attendance, school level attending, highest level of school attended, total years of study, race, ownership of a credit card, ownership of a checking account, wall Tenth Anniversary | 67

Elena M. Stacy

material, floor material, rooms, bedrooms, ownership of the residence, number of bathrooms, running water in the home, the source of the household water, the household waste destination, and the number of families living in the residence. Analysis & Results To begin, I present a table of the mean of key observable characteristics of the treatment and control groups in the final sample used for analysis. Note that almost every single characteristic is significantly different at between the treatment and control group, giving an indication that the two groups are not comparable at this point without significant bias. For this reason, I proceed with estimation of propensity scores using the logit model specified in â&#x20AC;&#x153;Data & Methodologyâ&#x20AC;?. In order to determine whether the groups have reasonable overlap in their propensity score estimates, I present a kernel density graph of the distribution of propensity scores across the treatment and comparison groups respectively in Figure 2. That is, if there is reasonable amount of overlap across the two groups without undergoing the process of propensity score matching, we have some evidence that propensity score weights may be effective in balancing the observable characteristics (de Brauw et.al. 2015). Figure 2 shows that the propensity scores across the two groups are fairly evenly distributed and contain overlapping trends in their distributions, and thus may be effective in balancing the observable characteristics. In Table 2, I present the new values of the differences between observable characteristics when using the inverse propensity weights described above as probability weights in the balance regressions (the observable characteristic regressed on treatment status, once for each characteristic). I find that none of the characteristics have significant differences across the two group after using the inverse propensity score weights, and can proceed with detection of heterogeneity across rural and urban households. The results of the regression analysis on quantity of food, type of food and disposable income per capita are presented in Tables 3, 4, and 5, respectively. In all of these tables, the main coefficient of interest is the interaction, Rural*BolsaFamilia, which captures the difference between the effect of Bolsa Familia on rural and urban households. In Table 3, we find a consistently positive coefficient on each regression with varying levels of controls. The result is not significant in column (1) but becomes significant at the 5% level once adding an increasing number of controls in the following regressions. The lack of significance in column (1) could suggest an omitted variable that was causing a downward bias. This downward bias was accounted for when adding in relevant controls. The significance in the subsequent regressions is evidence of a positive bias on the effects of Bolsa Familia transfers on householdâ&#x20AC;&#x2122;s quantity of food in rural areas relative to urban areas. 68 | Columbia Economic Review

Heterogeneous Effects of Bolsa Familia

Figure 2

Tenth Anniversary | 69

Elena M. Stacy

70 | Columbia Economic Review

Heterogeneous Effects of Bolsa Familia

Tenth Anniversary | 71

Elena M. Stacy

72 | Columbia Economic Review

Heterogeneous Effects of Bolsa Familia

In particular, rural families have approximately a 12% higher effect from Bolsa Familia on their ability to purchase enough food for the household. However, something troublesome to note in this table is that the coefficient on Bolsa Familia alone is consistently negative and significant, suggesting that receiving Bolsa Familia benefits is associated with having lower rates of sufficient quantity of food compared to not receiving Bolsa Familia benefits. This could be an unfortunate consequence of the program, but this result could also signal that there is something out of balance with the sample despite the propensity weights. In Table 4, results are not as promising. This table shows that Bolsa Familia has no significant heterogeneous effects on rural householdâ&#x20AC;&#x2122;s ability to purchase their preferred type of food relative to urban households, as none of the coefficients for the interaction term are statistically significant in any column. As all of the coefficients are negative and with smaller values representing a higher satisfaction with the food type purchased, a positive bias is represented. However, since these coefficients are not significant, we cannot statistically differentiate them from zero, so this is taken as a zero effect. Table 5, on the other hand, is highly significant in the first column (1), and still significant at the 1% level in subsequent columns on the interaction term. Column (4) gives the most precise estimate since it has the inclusion of the most relevant controls, the results show that in fact rural households have a negative bias on their disposable income per capita. Another notable result from this table is that all coefficients on receiving Bolsa Familia are negative and highly significant. This result indicates that Bolsa Familia transfers are actually associated with a reduction in household per capita disposable income rather than an increase. Furthermore, this result has a larger, more negative, impact on rural households compared to their urban counterparts. These results are problematic, because the results on ability to purchase sufficient quantity of food were positively biased towards rural households, while results on per capita disposable income were negatively biased towards rural households, and both results were statistically significant. These combinations of results make it difficult to have a definitive answer on the true heterogeneity of effects on rural and urban households. However, a main component of the disposable income calculation that is subtracted from total income is taxes. Thus, a potential hypothesis for the negative relationship between Bolsa Familia and disposable income per capita could be that those who receive Bolsa Familia tend to have higher tax rates than households with similar incomes who are non-recipients due to the benefits. Further research could suggest reasons why this may affect rural households more heavily than urban households, but regional tax rates may be at fault. Tenth Anniversary | 73

Elena M. Stacy

The positive bias on rural households on having sufficient quantity of food is contrary to my initial hypothesis, which was that rural households would have a negative bias. However, further consideration in light of these results has suggested that the reason behind this positive bias is due to price differences across regions. While Bolsa Familia transfer values are exactly the same across regions, those recipients in different regions may face different costs of living. It is reasonable to assume that urban areas have a higher cost of living than rural areas, and thus, rural households are left with greater purchasing power, despite receiving the same exact cash transfer as a similar household in an urban area. This could be an explanation for reporting higher rates of having sufficient quantity of food for the entire household, and thus a positive rural bias in this outcome. Potential problems that this analysis may have is endogeneity of the propensity score specification or of the regression specification. I argue that since there is evidence that noncompliance is not a significant issue with Bolsa Familia, we have no evidence that schooling is endogenous to this model. Recipients and non-recipients may have similar characteristics regardless of treatment status despite the conditions of the transfer program. Furthermore, the income that is used to calculate the propensity scores and as a control in the regression models employed is “Bolsa Familia eligible income”, i.e. net income before Bolsa Familia transfers, which also makes it an exogenous variable. These are the main variables that I found to be potentially threatening to the models used. Another potential issue of this model is that propensity score weights can lead to underestimates of standard errors. However, the robustness check that is used to regulate this issue is the addition of covariate controls that were used in the estimates of the propensity score calculation. This is seen with the progression of adding covariates to the columns in each regression table, and standard errors actually become more precise as this correction is added to the regression specification, so that by column (4) we can be quite sure that the standard errors are correct. Conclusion This paper was concerned with the presence of heterogeneous effects of the Bolsa Familia program. These results can help inform which groups derive the most benefit from similar government programs, and thus can potentially lead to policy implications. Using inverse propensity score weighting on all final regressions, the results indicated that heterogeneity was present in a household’s quantity of food and disposable income, but not on a household’s ability to buy preferred types of food. The result on household quantity of food was a positive bias towards rural households, and the result on household disposable income per capita was a negative bias against rural households. Surprisingly, on average, Bolsa Familia 74 | Columbia Economic Review

Heterogeneous Effects of Bolsa Familia

transfers are associated negatively with both having sufficient food, and log of disposable income per capita. This finding was surprising enough to cause worry of model misspecification, and I would recommend further study on this topic using alternate models and data sources which could confirm or deny them. However, at the very least, we find that the resulting heterogeneity of quantity of food was contrary to my hypothesis, and could potentially be explained by the fact that Bolsa Familia transfers do not account for differences in price of goods across regions. A potential policy implication could relate to considerations of an approach to accounting for differences in regional pricing and cost of living when dispersing government benefits. Although it’s reasonable to believe that regional price differences are present in many different countries, and these results would be the most useful in the context of applications to CCT programs around the world, results are particular to Brazil’s Bolsa Familia program, and may not be externally valid to other countries and other welfare programs. REFERENCES Austin, Peter C. “An Introduction to Propensity Score Methods for Reducing the Effects of Confounding in Observational Studies.” Multivariate Behavioral Research 46, no. 3 (May 31, 2011): 399- 424, http://www.tandfonline.com/doi/abs/10.1080/00273171.20 11.568786. Barrientos, Armando, Dario Debowicz, and Ingrid Woolard. “Heterogeneity in Bolsa Familia Outcomes.” Quarterly Review of Economics and Finance 62, (Nov 1, 2016): 33-40, https://search.pro quest.com/docview/1822390989. de Brauw, Alan, Daniel O. Gilligan, John Hoddinott, and Shalini Roy. . The Impact of Bolsa Família on Schooling. Vol. 70, 2015, http://www.sciencedirect.com/science/article/pii/S0305750X1500025X. “The Impact of Bolsa Familia on Women’s Decision-Making Power.” World Development 59, (Jul 1, 2014): 487-504, https://search.proquest.com/ docview/1518748989. Higgins, Sean and Claudiney Pereira. “The Effects of Brazil’s Taxation and Social Spending on the Distribution of Household Income.” Public Finance Review 42, no. 3 (May 1, 2014): 346-367, https://search.pro quest.com/docview/1516061308.

Tenth Anniversary | 75

Elena M. Stacy

Higgins, Sean, Nora Lustig, Whitney Ruble, and Timothy M. Smeeding. “Comparing the Incidence of Taxes and Social Spending in Brazil and the United States.” Review of Income and Wealth 62, no. S1 (Aug, 2016): S46, http://onlinelibrary.wiley.com/doi/10.1111/roiw.12201/ abstract. Higgins, Sean and Nora Lustig. “Can a Poverty-Reducing and Progressive Tax and Transfer System Hurt the Poor?” Journal of Development Economics 122, (Sep 1, 2016): 63-75, https://search.proquest.com/ docview/1813889775. Hirano, Keisuke and Guido Imbens. “Estimation of Causal Effects using Propensity Score Weighting: An Application to Data on Right Heart Catheterization.” Health Services and Outcomes Research Methodology 2, no. 3 (Dec, 2001): 259-278, https://search.proquest.com/ docview/228926421. Keisuke Hirano, Guido W. Imbens, and Geert Ridder. “Efficient Estimation of Average Treatment Effects using the Estimated Propensity Score.” Econometrica 71, no. 4 (Jul 1, 2003): 1161-1189, http://www.jstor.org/ stable/1555493. “Household Surveys: Pesquisa De Orçamentos Familiares.” http://www.econ. puc-rio.br/datazoom/english/pof.html. Soares, Sergei, Rafael Perez Ribas, and Fábio Veras Soares. “Targeting and Coverage of the Bolsa Família Programme: Why Knowing What You Measure is Important in Choosing the Numbers.” Working Paper, International Policy Centre for Inclusive Growth 71, (2010), https:// www.econstor.eu/handle/10419/71798. Teixeira, Clarissa Gondim. “A Heterogeneity Analysis of the Bolsa Família Programme Effect on Men and Women’s Work Supply.” Working Paper, International Policy Centre for Inclusive Growth 61, (2010), https:// www.econstor.eu/handle/10419/71807.

76 | Columbia Economic Review

THE COST TO INVESTMENT OF LEGISLATIVE TURNOVER, IMMOBILE LABOR, AND CORRUPTION: EVIDENCE FROM THE STATES OF INDIA

Pulkit Agarwal Harvard University Abstract: This paper studies the effect of quasi-randomly occurring legislative and administrative turnovers in Indian states on the likelihood of capital investments getting stalled and shelved. For stalling, it shows a positive effect significant at the 10% level that is compounded by the interaction with higher inflation. For shelving, no coherent causal effect is observed. It further presents suggestive evidence to show that variation in labor market reforms explain more heterogeneity in this effect across states than variation in corruption levels. These results bear important implications given increasing political competition in India and differing patterns in the occurrence of turnovers across states.

his paper studies the relationship between legislative turnovers in state governments and the stalling of concurrently initiated building projects in India. This is a question that is relevant to both Indian policymakers and investors; a positive causal effect of turnovers on the likelihood of projects being stalled would suggest that newly elected governments must become more concerned with maintaining continuity in relationships with prior investors in the state and that businesses should be justifiably circumspect of raising capital expenditures in states with uncertain impending elections. A negative effect, on the other hand, would indicate that frequent turnovers are conducive to and perhaps even instrumental in stimulating investment. The estimated results show that quasi-randomly occurring turnovers have a positive causal effect on the probability of a project announced in the same year being stalled, but this relationship is especially pronounced in times of fiscal- and price-volatility. I go on to provide suggestive evidence to show that heterogeneity in labor market reforms across states may account for some of the variations in this effect. In particular, I find that states with more reformed labor laws see a greater positive effect of turnovers on the likelihood of projects being stalled. Additionally, firms that employ contract labor are able to almost fully counter this effect. Finally, I test whether accounting for variation in corruption levels across states affects this relationship, and do not find significant results for the interactions. Tenth Anniversary | 77

Pulkit Agarwal

Background Literature and Institutions In India’s system of parliamentary democracy, legislative turnover is entirely coincidental with administrative turnover. As is pointed out in Horowitz et al. (2009), much of the existing literature on the importance of turnover suffers from the confounding effect of multiple political institutions. The absence of the possibility of a ‘divided government’ in India allows for a more consistent definition of turnover, which is useful in measuring its importance for economic outcomes such as capital investment. However, all such turnovers may not represent similar ideological shifts in governance. As such, this variation in the nature of turnovers is one that this paper attempts to control for by focusing on within-state effects of turnovers, as discussed in “Data”. Past research has been cautious of commenting on the overall desirability of turnovers, which, when occurring infrequently, can be an obstruction to policy change and political competition and, when too frequent, can lead to irresponsible and unaccountable legislatures (Peltzman 1976 and Van Herzeele 2017). A developing thread in the literature has considered this relationship in emerging markets and found favorable effects of consistency in governance: Brender and Drazen (2008) find a positive relationship between re-election and economic growth, while Hu et al. (2010) concludes that turnover in top management at Chinese state-owned enterprises is negatively associated with firm performance. More recent literature has also called the rationality of turnover into question: Cole et al. (2012) argue that Indian voters punish politicians for natural disasters beyond their control, while Leigh (2009) reports that voters reward elected officials for global rather than national economic growth. his paper will consider close elections - which would have been difficult to call ex-ante - to study whether, irrespective of what causes the turnover, it has a bearing on the continuity of investment. Ever since the liberalization reforms of 1991, state governments in India have played a crucial role in enabling or impeding investment given particular constitutional authorities with which they are charged. Pre-1991, in the age of the ‘license raj’, political considerations weighed heavily on how the central government awarded industrial licenses to states (Biswas et al. 2010). Gradually since, however, there has been increasing and fairer competition among state governments to win over investment by providing incentives of taxation, environmental and labor compliance, and land allocation. Furthermore, Vadlamannati (2011), following liberalization, investors and state governments are required to file all but one application - the Industrial Entrepreneurs Memorandum - with central authorities in New Delhi upon the agreement of a new investment project. Thus, finalizing of business proposals and soliciting of all required permissions occurs at the state level prior to eventual clearance in the national capital, which highlights the potential

78 | Columbia Economic Review

Turnover, Labor, and Corruption

disruption that changes in state administrations may cause to ongoing business activity. An important sphere of local power is in the area of labor relations where states enjoy a full monopoly of legislative and executive authority and exhibit substantial heterogeneity. An OECD paper (2007) compares state governments in India based on the number of reforms they have made to the national labor code, and whether these reforms have reduced or increased transaction costs for new businesses. It finds, and finds that states such as Bihar and Chhattisgarh have been rather slow in reforming their labor markets, while Uttar Pradesh and Gujarat have made considerable strides in the direction of favoring market-based regulation. A similar if older study, Besley et al. (2004), finds that states that have made amendments to the Industrial Disputes Act of 1947 in the “pro-worker” as opposed to “pro-market” direction have experienced slower growth. It is plausible, therefore, that a disruptive effect of unexpected legislative turnover on investment may also vary in magnitude or direction based on the preexisting level of reform in a state’s labor code; I test this relationship later in “The Labor Market Teory”. A last but notable aspect of state governments that may influence investment climates is their level of corruption. Amirapu et al. (2014) has found, through a combined study of labor regulations and corruption levels across Indian states, that the negative effect of restrictive labor laws on firm growth is compounded by the case of corrupt state governments. Relatedly, an OECD study notes that, across countries, corruption occurs most often in the procurement of public contracts (2014). Given the role of India’s state governments in delivering such contracts corruption is likely to be an important element of a state’s investment machinery. State governments have incentives to be both less corrupt, in order to attract investment from businesses looking to compete on a level playing field, and more corrupt, to attract businesses seeking preferential access to licensing contracts or resources under public control. As such, it is plausible that preexisting arrangements of corruption be disrupted as a result of turnover in state governments, which may contribute to the slowdown of incoming investment; I test this hypothesis in “The Labor Market Teory”. Data The primary source of data for this paper is the Database of the Lifecycle of New Investments, CapEx, from the Centre for Monitoring Indian Economy, a private economic research firm based in Mumbai. It tracks all new projects initiated by public and private sector firms in India through their lifecycles – from the date of announcement to the date of completion—and enlists their costs, industry, intended purpose, location, among several other associated information. The detailed nature of the data allows for a study of the heterogeneity of the effect of interest on firms by industry, by private or public control, and by Tenth Anniversary | 79

Pulkit Agarwal

state. After removing missing entries in the data, I consider 60,162 observations, each of which represents an individual investment project launched after 1991. The data on state elections is sourced from the Election Commission of India and spans elections across twenty-five Indian states that account for 92% of the national population. Of the 136 state elections that have occurred since 1991, 87 resulted in a legislative turnover. I limit my sample to elections in which the largest post-poll alliance of parties received no more than a 3-percentage point greater vote-share than the second largest post-poll alliance; this is done to isolate elections that were close enough to represent a quasi-random, or unanticipated, change in government. This leaves a sample of 48 elections. Conducting empirical analyses using a causal inference model of the kind employed in this paper often represents a challenge of accounting for sufficient fixed-effects - such as state, year, and industry - while ensuring that the power of the coefficient of interest is not absorbed entirely by such controls. In this paper, I will be using state-fixed effects instead of year-fixed effects for a variety of reasons: one, states in India show substantial variation in growth rates of output and investment elasticity to output, for both of which trusted data are difficult to find; two, given that there is a fewer number of large states with many investments than there are years, the estimation model is likely to hold more power with state-fixed effects than with year-fixed effects; and three, since years are closely related to the chief independent variable - a year with an election that resulted in a quasi-random turnover - fixed effects may absorb some of the effect of turnover itself. In lieu of time-fixed effects, the paper employs data on national inflation, annual growth rates, and real interest rates - from the World Bank, OECDâ&#x20AC;&#x2122;s National Accounts, and the St. Louis Federal Reserve Bank respectively. Furthermore, employing industry-fixed effects in the model is crucial given the variation in the applicability of licensing and labor laws to firms across sectors. This is demonstrated by Bhagwati and Panagariya in chapter 8 of their book through an exposition of various protections afforded to businesses in â&#x20AC;&#x153;coreâ&#x20AC;? industries, the reservation of all labor-intensive manufacturing exclusively for small enterprises until recently, and the sectoral differentiation in state laws pertaining to the use of contract labor (2014). Table 1 shows how characteristics of projects that were or were not stalled differ by their industry. Between healthcare and education, two sectors under direct state supervision and of inarguably great developmental importance, the former shows relative consistency by cost and public-control across projects that were or were not stalled, whereas in the case of the latter, projects that got stalled were on average almost three-times as costly as those that were not. The paper will also draw from data on state-wise labor regulations and corruption measures as compiled by researchers from the OECD, National 80 | Columbia Economic Review

Turnover, Labor, and Corruption

Council of Applied Economic Research in New Delhi (NCAER), Asian Development Bank, and the Delhi School of Economics to test mechanisms that could be driving the main results.

Empirical Framework The estimation strategy used in this paper is of the following general form: Pr(Ypist | X)= αs+Γi+β1 Xst+Vp+Wt+ξpist In the above model, left-hand side variable Y is the conditional probability that equals 1 if a project ‘p’, in industry ‘i’, in state ‘s’, announced in year ‘t’ gets stalled, and 0 otherwise; αs represents state-fixed effects; Γi stands for industry- fixed effects; Xst is the main independent variable which equals 1 if the given state had an election with a quasi-random turnover in the year ‘t’; β1 is the coefficient of interest, interpreted as the causal effect of a quasi-random turnover on the likelihood of a project announced in the same year being stalled at some point in its lifecycle; Vp denotes a vector for control variables by project characteristics such as its cost, whether it was owned by a public or foreign company, and whether it was eventually discarded; Wt accounts for time-varying national level controls such as the inflation and real interest rates; and finally ξ is the residual. I can think of three potential sources of bias in this model: one, variables unaccounted for in the regression that may be both correlated with a quasi-random turnover and determinants of a project being stalled, such as a short-term shock in the economic performance of a state resulting from natural disasters or an atypical monsoon season; two, if there is a mechanism for reverse causality such that as a result of projects being stalled in a state, it becomes more Tenth Anniversary | 81

Pulkit Agarwal

likely for an incumbent government to lose a close election; and three, if projects that are announced in the year of such a turnover are stalled prior to the occurrence of the election. It is difficult to address the first of these concerns given that state-fixed effects combined with national indicators for economic performance cannot fully account for short-term, regional economic shocks. However, it is likely that projects in certain industries are more susceptible to such shocks than others â&#x20AC;&#x201D; for example those that may be reliant on agricultural inputs â&#x20AC;&#x201D; in which case industry-fixed effects should help counter some of this bias. The reverse causality concern is likely addressed through the use of close elections as the chief independent variable; although this is not as robust as a regression discontinuity, the occurrence of turnovers after close elections is arguably random enough to not be determined by the stalling of investments that were initiated in the same year. It is also worth mentioning that because the model considers the stalling of projects merely announced in election years - not necessarily ones that were also stalled in the same year - it is unlikely that the dependent variable would be a determinant of the independent. Lastly, and relatedly, if there are projects that were announced and stalled prior to the occurrence of such elections, and special business interests involved lobbied or mobilized voters to oust incumbent governments as a result, then the estimated effect would indeed be biased; this is unlikely, however, for this would require substantial and prompt coordination on part of investors whose projects are stalled, and for them to exercise considerable influence over votersâ&#x20AC;&#x2122; preferences to overturn a close election in favor of the opposition alliance. An important limitation of this approach is that it does not allow for the studying of the effect that turnovers may have on the stalling of projects announced in the years prior to that of a turnover. There are two reasons for why this qualified approach is employed: one, projects announced well before elections may have passed their initial phases of implementation, which is where the possibility of stalling is likely to be highest and most relevant to study; and two, it is difficult to distinguish projects launched in previous years based on whether they were initiated under a government of the same or different political party as the one that assumes incumbency following a turnover. In other words, restricting the sample to projects announced in the same year as a turnover guarantees that the governing party changed soon after the projects were launched. The rationale for employing a probit over a linear probability model is twofold. Firstly, my response variable is binary (1 for a project being stalled, and 0 otherwise), and the probit is more suited to such an analysis as it estimates probabilities of projects with a certain set of baseline characteristics falling in one or the other category. Secondly, it is difficult to reasonably assert that the marginal effect of the independent on the response variable is likely to be linear; 82 | Columbia Economic Review

Turnover, Labor, and Corruption

for example, it may be significantly more likely for projects in industries such as steel - which are on average more likely to be stalled (Table 1) — than in renewable energy to be stalled as a result of close turnovers. A linear model would instead forcibly, and likely wrongly, assume these marginal effects to be constant across baselines. Summary Statistics This section presents two tables of summary statistics. The first one classifies projects based on whether or not they were stalled at some point in their life-cycle, and lists averages for certain characteristics (Table 2).

It is interesting to note that a similar share of projects across both categories were announced in the year of a turnover, or a close turnover, which suggests on the surface that it may not be any more likely for projects announced around changes in government to be stalled than those that were announced at another time; the causal effect estimated later in the paper will shed more light on this relationship. Further, among projects that were stalled, 0% go on to eventually be shelved, whereas 3% of projects that are never stalled are later abandoned. This may suggest that the two occurrences are unrelated: Perhaps projects that are stalled are done so as a result of short-term shocks or temporarily unfavorable market conditions, whereas those shelved are deemed altogether unfit to be carried forward. If this is the case, it may be worth including the indicator for ‘Was shelved’ in the original model to control for projects that have poorer chances of being completed. Similarly, it is possible that being stalled has no effect on the chances of a project being shelved later on, which I test later in “Main Results”. Projects that are stalled are also, on average, considerably costlier than those that are not. This may be because they are more uncertain, require greater investments, and are hence also likelier to run into unforeseen issues that force them to be set aside. It also, at the very least, points Tenth Anniversary | 83

Pulkit Agarwal

to the great economic cost and uncertainty that stalled projects may impose on the financial system: The total cost of all stalled projects in this sample amounts to over 700 billion dollars over 28 years. Lastly, a relatively small share of projects promoted by publicly- or foreign-owned companies are stalled, which could either be because they receive greater financial or regulatory support, or because they are better managed in general. Notwithstanding, this makes a strong case for including both ‘Public’ and ‘Foreign’ indicators as controls in the main regressions. Table 3 reports characteristics for projects by state, based on whether or not they were stalled; a few key statistics are salient.

Firstly, publicly-owned projects have significantly lower chances of being stalled than private projects in Gujarat and, to a lesser extent, in Uttar Pradesh (UP) than in Madhya Pradesh (MP) and Sikkim. Notably, Gujarat and UP also have a smaller share of total projects that are initiated by the public sector, and based on the OECD labor reform index, have more pro-market labor regulations. Interestingly, the relative cost of stalled projects to other projects is substantially higher in Uttar Pradesh than in all other states, and as is the coincidence of turnover with the year of announcement of such projects. It is clear, thus, that due to variation in the frequency of turnover, share of publicly-owned projects and a variety of unobserved characteristics across states, it would be imperative to employ state- fixed effects in the regression models. Main Results Table 4 presents the probit estimates from regressions of the primary response variables - likelihood of a project being stalled or shelved - on the occurrence of a legislative turnover. Column 1 shows results from a naive OLS regression where the explanatory variable is an indicator for a legislative turnover having occurred in the year of a project’s announcement. Given that the 84 | Columbia Economic Review

Turnover, Labor, and Corruption

explanatory variable in this case is not randomly assigned, these results, besides not being statistically significant, also do not represent causal effects. Columns 2, 3, and 4 show results from OLS, probit, and logit estimates for regressing the occurrence of a quasi-random legislative turnover in a state on the likelihood of a project - announced in the same year - being stalled in its lifecycle, without accounting for state- and industry-fixed effects. The positive coefficient on the explanatory variable is significant at the 5% level in each case, which appears to support the hypothesis that the occurrence of turnover indeed slows down the lifecycle of investments. When I add fixed effects to the probit model in Column 5, results remain significant albeit at the 10% level. In Column 6, I test for whether a similar effect is observed on the likelihood of a project being shelved, but the results do not turn out to be statistically significant. Per the discussion earlier in the paper, in Column 7 I test for whether a projectâ&#x20AC;&#x2122;s having previously been stalled - conditional on it having originally been announced in a year of a random turnover - affects the probability of it being shelved; once again, the coefficients for both the main explanatory variable and interaction variable remain statistically insignificant. At this point, I fail to reject the hypothesis that a random legislative and administrative turnover has no effect on increasing the chances of investment projects being abandoned. Finally, I question whether the notably substantial effect of turnover on the stalling of projects is in any way affected by macroeconomic conditions; while controls such as the inflation rate and discount rate are included in each previous model exhibiting significant results, it is worth testing whether the causal relationship in question is altered by the confounding effect of a concurrently poor macroeconomic outlook. In order to include a suitable measure of economic uncertainty, I interact the explanatory variable, â&#x20AC;&#x2DC;quasi-random turnoverâ&#x20AC;&#x2122;, with the average annual national inflation rate. There are four reasons for why this measure is selected: one, since the Reserve Bank of India reformed its monetary policy framework to include inflation targeting only as late as August 2016, for much of the time period covered in the data, inflation remained a somewhat unbiased and flexible measure of fiscal stability in India; two, as per Mohaddes et al. (2014), variability in inflation levels in India has been found to be an important determinant of long run growth rates of output - in particular, persistently high inflation has been found to hurt economic growth; three, inflation in developing countries often has the potential to stir social unrest (Brookings 2011) which could worsen the investment climate; and four, given the dearth of consistent CPI measures across most Indian states, I can only use national indicators for price volatility.

Tenth Anniversary | 85

Pulkit Agarwal

The results shown in Column 8 shed new light on the relationship between turnovers and the stalling of projects. The coefficients on both the main explanatory variable and the interaction are found to be statistically significant; the former is so at the 5% level and the latter at the 1% level. The sign of the coefficient on ‘quasi-random turnover’ however is negative, as opposed to in all previous models. In fact, the marginal effect of a concurrent turnover under average inflation conditions is to reduce the likelihood of a project being stalled by 0.027 (reported in Appendix A); this represents a 61 percent decrease in the probability, contrary to the 25 percent increase reported earlier. Interestingly, the coefficient on the interaction variable has a positive sign, which implies that a close turnover accompanied by one standard deviation rise from the average inflation rate would decrease a project’s chance of being stalled by 0.013, or 29 percent. Since inflation at the state-level can be more volatile, there is a possibility that the interaction term is biased downward - i.e. a similar rise in within-state inflation may have a more substantial counteracting effect on the observed negative relationship between turnovers and the stalling of projects. On the other hand, it may also be possible that rise in national inflation is a more concerning statistic for investors about the overall desirability of an investment destination; in such a case, perhaps, more localized inflation may not have as substantial an effect on the slowdown of investment into a state as the estimates report in Column 8. It is not possible to test either theory without data on state- level inflation, unfortunately. The Labor Market Theory As discussed in “Background Literature and Institutions”, considerable attention has been paid by economists to the variability of labor laws across 86 | Columbia Economic Review

Turnover, Labor, and Corruption

Indian states, and their consequences on investment and output growth. It is plausible that one of the more crucial ways in which new state governments may impact the rate of inflowing investment is by executing labor laws from such codes as the Industrial Disputes Act of 1947 and the Factories Act of 1952. Given the differentiated rate of reforms made to these labor codes across states, I consider whether they impact the relationship between government turnover and slowdown in the completion of investment projects. My initial hypothesis is that states with more reformed labor laws are likely to leave incoming governments with little authority to halt ongoing projects by interfering in business-labor relations, and are therefore likely to see a smaller effect of turnover on the increase in probability of contemporaneous projects being stalled, than states with less reformed labor codes. In order to conduct a thorough analysis of the importance of this mechanism, I draw on data from an index developed by Gupta et al. (2009) that places states into three buckets based on the relative level of reforms they have made to the Industrial Disputes Act - 0 being neutral, 1 being most pro-market reforms, and -1 being most pro-worker reforms. In Figure 1, I plot this index against the OLS coefficient on â&#x20AC;&#x2DC;quasi-random turnoverâ&#x20AC;&#x2122; for each state. The resultant trend line does not present a strong relationship between the two metrics, which undermines the initial hypothesis.

Tenth Anniversary | 87

Pulkit Agarwal

To further test the viability of this theory, I use data on an index developed by the OECD that scores states based on both the business-friendliness of their labor reforms, and the enforcement and adjudication of industrial disputes. I also use a state-level metric specifically for labor mobility, compiled by the National Council for Applied Economic Research and the Foreign and Commonwealth Office of the British High Commission, India (2016). A concern with each of these sources is that they compare the relative levels across states at a fixed point in time rather than tracking changes in their labor regulations across the period covered in this dataset. For a more accurate measure of labor market flexibility, I also employ project-level data on the use of contract labor from the original CapEx database. The use of contract labor in India has been found to be a way for employers to circumvent restrictive labor laws, most of which do not apply to temporary workers (Bertrand et al. 2015). It is plausible therefore that projects that employ contract labor are less susceptible to changes in state governments because of the limited disruption this may cause to their labor relations. I consider each of these variables in interaction with the main explanatory variable and report results in Table 5. Column 1 shows that there is no significant relationship between the interaction of the explanatory variable and a stateâ&#x20AC;&#x2122;s NCAER labor mobility score, and the likelihood of a project getting stalled; in fact, turnover by itself is also not significant when one controls for this interaction. In Column 2, while Pi remains positive and significant, the interaction with the state-wise labor reform score in Gupta et al. (2009) is found to be insignificant. This suggests that the effect of turnover on stalling of projects is independent of the variability 88 | Columbia Economic Review

Turnover, Labor, and Corruption

in labor reforms across states. Column 3 reports the coefficients for â&#x20AC;&#x2DC;quasi-random turnoverâ&#x20AC;&#x2122; and its interaction with the OECD index for labor markets to be significant; in terms of its net marginal effect, a quasi-random turnover is estimated to reduce the probability of a project being stalled by 0.006, or 13.6 percent, for a state with the mean OECD reform index. When the given state, holding all else constant, improves its score in the index by one standard deviation, the probability of stalling increases by 0.012, or 27%. While this might seem somewhat counterintuitive, one could consider the risk that projects in states with more reformed labor laws have to contend with regards to changes towards more restrictive enforcement in the future, especially if the likelihood of restrictive changes in more reformed states is higher than the

Tenth Anniversary | 89

Pulkit Agarwal

likelihood of liberal changes in states with less reformed laws. Finally, in Column 4, I report that in the absence of contract labor, turnovers on average increase the likelihood of a contemporaneous project being stalled by 0.041, or 93 percent, while in the case that a project does employ contract labor, the probability is reduced by 0.036, or 81 percent; notably, each of these results are significant at the 1% level. This is consistent with results in Menon et al. (2013) and Bertrand et al. (2015), as well as the arguments made in Bhagwati et al. (2014) about the role played by contract labor in enabling certain industries to allocate inputs more resourcefully. It is worth highlighting that while the composite labor market indices indicate mixed evidence for any clear effect of labor market reforms on the relationship of turnovers with investment slowdowns, the use of contract labor itself is dictated by regulatory reforms that vary across industries and states. Therefore, while all labor reforms may not have a coherent effect on this relationship, it certainly appears so that allowing firms to hire contract workers may enable them to counter some of the negative effect of legislative turnovers. The â&#x20AC;&#x2DC;Putting Brakes on Corruptionâ&#x20AC;&#x2122; Theory Another theory that I posit to explain the average negative effect of turnover on contemporaneous projects as found in Columns 3 and 5 of Table 4 is focused around corruption. In particular, I hope to test whether the effect of a turnover is more prohibitive on investment projects in states where governments are known to be more corrupt than in less corrupt states. This could be possible as administrative turnover may disrupt the progress of corrupt arrangements between investors and state governments, which in turn may lead to delays and the stalling of projects. In order to study this mechanism, I use data from an NCAER report that scores states on a variety of governance indicators such as the prevalence of e-governance - which in theory enhances transparency and curbs corruption - the share of all crimes in a state that are classified as economic offences, and a composite indicator for good governance. In Figure 2, a slightly negative relationship is shown between the effect of turnovers on the likelihood of projects getting stalled and the performance of a given state in tackling corruption and economic crimes. In other words, in states that are less corrupt as per this metric, projects launched in a year of turnover are on average less likely to be stalled as a result. I conduct regression analyses to further test this theory and report estimates in Table 6. Interestingly, none of the interaction coefficients in Columns 1, 2, or 3 are found to be statistically significant, which weakens the theory that corruption strengthens the negative effect of turnovers on the stalling of projects. Since the coefficient on the main explanatory variable remains positive and significant at the 10% level in Columns 1, and 3, it can be argued that accounting for corruption does little to affect the observed results. This 90 | Columbia Economic Review

Turnover, Labor, and Corruption

lends to two possibilities: either the theorized effect of corruption on the relationship between turnovers and the stalling of projects is negligible, or that corruption affects this relationship through competing mechanisms. For example, while a disruption to corruption may be an important channel through which turnovers slow down investment, if corruption continues to persist after a turnover â&#x20AC;&#x201C; as may well be the case â&#x20AC;&#x201C; then it could also enable a smooth transition to newer, similarly corrupt arrangements between investors and incoming governments. As such, a scenario in which these two effects are almost equal in magnitude and cancel each other is not wholly surprising.

Tenth Anniversary | 91

Pulkit Agarwal

Since the NCAER governance indices are only available for a specific time, it is also possible that time-series data on corruption uncover a significantly different effect on the relationship between turnovers and the stalling of projects from what I have shown. However, in the absence of such data, I fail to reject the hypothesis that corruption has no role in driving the observed effects of the explanatory on the response variable. From the evidence presented in this paper, it is clear that legislative turnovers in India’s state governments can play a role in determining the likelihood of recently launched projects getting stalled. While I find, along initially expected lines, that this effect is on average positive, results from specific interactions reveal certain nuances in this relationship. The coincidence of quasi-random turnovers with high levels of inflation can make their disruptive effect on capital investment substantially more pronounced. Turnovers do not appear to have a similar predictive effect on the chances of building projects being eventually shelved, which suggests that most projects affected could have positive Net Present Values. These results are in line with Uppal (2010), which argues that excessive turnovers result in the misallocation of public investments; I show that there are also prohibitive effects for private capital investments. When one considers the stark variation across Indian states in the occurrence of incumbent turnovers -0.77 in Punjab as opposed to 0.42 in West Bengal - it is possible to see why this may be consequential in the considerations of businesses making such investments. I find mixed evidence for whether this effect of turnover is driven by states’ interference in labor markets; while the general level of labor reforms undertaken in a state does not appear to impact this relationship, accounting for the level of enforcement of restrictive labor laws within states does explain some variation. Furthermore, this paper shows support for the claims in Bhagwati et al. (2014) and Bertrand et al. (2015) that contractual labor arrangements in India have helped firms circumvent certain restrictions in labor mobility. Finally, this paper finds no causal evidence for whether this effect of turnover is compounded by corruption across states. There is a strong case for India’s state governments to actively support the continuity of investments upon the occurrence of legislative turnovers. Given rising frequency in the incidence of turnovers (Cole et al. 2012) - driven by greater competition in Indian political markets - ignoring this effect may hurt the economy’s long-term accumulation of fixed capital. Finally, businesses looking to expand capital investments should be wary of states with uncertain prospects for turnovers, especially in the case that they have relatively pro-market labor regulations. Further research should focus on policy reforms that can be undertaken at both the center- and state-levels to ensure that greater political competition in India fosters, rather than stifles, investment.

92 | Columbia Economic Review

Turnover, Labor, and Corruption

REFERENCES Ahsan, Ahmad and Carmen Pages. (2007). “Are all Labor Regulations Equal? Assessing the Effects of Job Security, Labor Dispute and Contract Labor Laws in India,” World Bank Policy Research Working Paper 4259. Amirapu, Amrit and Michael Gechter. (2014). “Indian Labor Regulations and the Cost of Corruption: Evidence from the Firm Size Distribution,” Boston University Working Paper. Asher, Sam and Novosad, Paul. (2017). “Politics and Local Economic Growth: Evidence from India,” American Economic Journal: Applied Economics. 9. 229-273. Besley, Timothy and Burgess, Robin. (2004). “Can Labor Regulation Hinder Economic Performance? Evidence from India,” The Quarterly Journal of Economics, 119, issue 1, p. 91-134. Bhagwati, J. and Arvind Panagariya. (2014). “Why Growth Matters,” Public Affairs. Biswas, Rongili, Sugata Marjit and Velayoudom Marimoutou. (2010). “Fiscal Federalism, Sate Lobbying and Discretionary Finance: Evidence from India,” Economics and Politics, 22: 1. 68-91. Brender, Adi, and Allan Drazen. (2008). “How Do Budget Deficits and Economic Growth Affect Reelection Prospects? Evidence from a Large Panel of Countries,” American Economic Review, 98 (5): 2203-20. Cole, Shawn and Andrew Healey. (2012). “Do voters demand responsive governments? Evidence from Indian disaster relief,” Journal of Development Economics, vol. 97, issue 2. 167-181. Dougherty, S. (2008). “Labour Regulation and Employment Dynamics at the State Level in India”, OECD Economics Department Working Papers, No. 624, OECD Publishing, Paris. The Economist, “Rigging the Bids,” November 19, 2016. Gupta, Poonam, Rana Hasan and Utsav Kumar. (2009). “Big Reforms but Small Payoffs: Explaining the Weak Record of Growth and Employment in Indian Manufacturing,” Munich Personal, Paper No. 13496. Hasan, Rana and Karl Robert L. Jandoc. (2012). “Labor Regulations and the Firm Size Distribution in Indian Manufacturing,” Columbia University Academic Commons, https://doi.org/10.7916/D88G8TVH. Horowitz, Shale, Karla Hoff, and Branko Milanovic. (2009). “Government turnover: Concepts, measures and applications,” European Journal of Political Research, 48. 107-129. Hu, Fang and Leung, Sidney. (2010). “Top Management Turnover, Firm Performance and Government Control: Evidence from China’s Listed State-Owned Enterprises,” The International Journal of Accounting. 47. 10-39. Tenth Anniversary | 93

Pulkit Agarwal

International Monetary Fund, Interest Rates, Discount Rate for India [INTDSRINM193N], retrieved from FRED, Federal Reserve Bank of St. Louis; https://fred.stlouisfed.org/series/INTDSRINM193N, April 26, 2018. Knutsen, Carl and Wig, Tore. (2015). “Government Turnover and the Effects of Regime Type: How Requiring Alternation in Power Biases Against the Estimated Economic Benefits of Democracy,” Comparative Political Studies. 48(7). 11-77. Leigh, Andrew and Mark McLeish. (2009). “Are State Elections Affected by the National Economy? Evidence from Australia,” Economic Society of Australia, 85: 269. 210-222. Marianne Bertrand, Chang-Tai Hsieh, and Nick Tsivanidis. (2015). “Contract Labor and Firm Growth in India,” University of Chicago Working paper. Menon, Nidhiya and Yana van der Meulen Rodgers. (2013). “Labor Regulations and Job Quality: Evidence from India,” ILR Review, Vol. 66, No. 4.933-957 Mohaddes, K. and Mehdi Raissi. (2014). “Does Inflation Slow Long-Run Growth in India?” IMF Working Paper, WP/14/222. National Council for Applied Economic Research, “The NCAER State Investment Potential Index,” March 2016. OECD, “Regulatory Management and Reform in India”, 2008. OECD, “India: Economic Surveys”, 2007. Peltzman, S. (1976). Toward a more general theory of regulation. Journal of Law and Economics, 19 (2). 211-240. Prasad, Eswar. (2011). “The Impact of Inflation on India and Other Developing Nations,” The Brookings Institution. Sekeris, Petros and Litina, Anastasia & De Luca, Giacomo. (2013). “Growth- Friendly Dictatorships,” Journal of Comparative Economics. 43. 10-46. Tiruneh, Gizachew. (2006). “Regime type and economic growth in Africa: A cross-national analysis,” The Social Science Journal. 43. 3-18. Uppal, Yogesh. (2010). “Does legislative turnover adversely affect state expenditure policy? Evidence from Indian state elections,” Public Choice. 147.189-207. Vadlamannati, KC. (2011). “A Race to Compete for Investment Among Indian States,” National Bureau of Economic Research Working Paper. Van Herzeele, Brecht. “The Influence of Legislative Turnover on Policy Innovation,” European Consortium for Political Research, Oslo, September 2017.

94 | Columbia Economic Review

THE DETERMINANTS OF NATIONALISM AND THE EFFECT OF CONSCRIPTION ON NATIONAL PRIDE

Dhivesh Dadlani Harvard University Abstract: This paper aims to empirically explore the variation in nationalism between individuals and countries around the world. This study depends heavily on the findings of the World Value Survey, a global research project designed to explore people’s values and beliefs, and their consecutive impacts on society and politics. This data is used to explore the determinants of nationalism and understand the effect of conscription on nationalist sentiments. On the individual level, I find that nationalism is positively correlated with age, confidence in one’s government, religiosity, interest in politics, communalism, armed forces enrollment, and marriage; it is negatively correlated with education and generalized trust. On the national level, I find that it is positively correlated to regional conflict; it is negatively correlated to trade and taxes. Where conscription is concerned, I find that the effect of removing mandatory conscription on nationalistic sentiments is positive. Hence, countries with mandatory conscription are less nationalistic.

ationalism refers to an individual’s feelings of patriotism and loyalty towards their nation. Much of the literature in economics surrounds the relationship between the nation-building process and the economy. Nation-building is defined as a process which leads to the formation of countries in which the citizens feel a sufficient amount of commonality of interests, goals and preferences, which inhibits desires to separate from each other (Alesina and Reich, 2015). However, there exists little empirical evidence documenting how the process of nation-building works and the tools that governments employ to influence nationalism in a country. This raises the question: Why examine nationalism and nation-building? There are cases of nation-building being utilized in efforts to resolve a variety of social problems, especially in societies with ethnically fragmented populations (Ahlerup and Hanson, 2011). Ethnically fragmented countries tend to exhibit poor economic and political performance. Studies accredit negative outcomes related to ethnic diversity, such as distorted provision of public goods, enablement of corruption and consequently, poor economic growth. Employing nation-building as a policy tool allows governments to shift from unity within ethnic groups and create unity in a country. Thus, such a policy that is capable

Tenth Anniversary | 95

Dhivesh Dadlani

of reducing the fragmentation caused by ethnic diversity and influencing nationalistic sentiments is worth studying. In this paper, I aim to provide empirical evidence about the determinants of nationalism and the effect of conscription on nationalism. The following null hypotheses will be tested:

1. Men who are forced to serve their nation due to conscription laws become more nationalistic throughout their years of service.

2. On an individual level, income and education negatively correlate with nationalistic sentiments; trust, confidence in oneâ&#x20AC;&#x2122;s government, age and political interest positively correlate with nationalism.

3. On the national level, international wars, average confidence in governments, and average trust levels positively correlate with nationalist, while GDP per capita, openness of economies, civil wars, and conflicts negatively correlate with nationalism.

I will begin by reviewing the existing economic literature on nationalism. This precedes an examination of the data used for analysis on an individual, in particular, examining Spain to understand the effects of conscription on nationalism. Literature Review Two studies the quantitative aspect of nationalism. Ahlerup and Hannson (2011) empirically assess the effects of nationalist sentiments on government effectiveness, using a cross-section of countries. They use the aforementioned World Value Survey to measure the intensity of nationalistic sentiments through a series of questions about national pride, which I will also employ in my study. Their paper has two main findings. There is a hump-shaped relationship between nationalism and government effectiveness (Ahlerup and Hanson, 2011), which they use to substantiate the claim that various factors impact different levels of nationalism. They explain that nationalism has positive effects on government effectiveness at low levels of nationalism, but has negative effects at high levels of nationalism. Additionally, they find that nationalism can erode the negative association between ethnic fractionalization and government effectiveness (Ahlerup and Hanson, 2011). The second study that empirically assesses effects use the same measure of nationalism as well as data from the International Social Survey Program (ISSP). Shulman (2003) is interested in testing the two hypotheses that richer countries with richer individuals, as well nations with less economic inequality amongst individuals and ethnic groups, are more nationalistic. He finds that relative and absolute wealth, as well as economic equality, are not positively related to nationalism. Specifically, only 6 out of the 20 countries he studied yielded significant relationships between nationalism and income. 96 | Columbia Economic Review

The Determinants Of Nationalism

In these six countries, the relationship is negative, which means that poorer people are likely to be more nationalistic (Shulman, 2003). Even after considering various controls—such as class, settlement size, education, ethnicity, religiosity, age and sex—he found few statistical relationships, all of which negative. From a theoretical approach, Alesina and Reich (2015) study nation-building across political regime and in transition periods. They are interested in understanding the reasons why some dictatorships, as well as European elite class and the Soviet Union, wanted to homogenize their populations, and why certain colonies engaged in nation-building after independence. They argue that in non-democracies, rulers are motivated to homogenize populations when they fear democratization as 1) it allows people in charge to implement their own preferred policies even if democracy prevails and 2) homogenization may reduce any ill feelings towards their rule, reducing incentives to overthrow them. Through historical evidence, he argues that homogenization and indoctrination through nation-building are thus used as a means of making people less averse to dictators’ rule (Alesina and Reich, 2015). Although I do not analyze these issues in my study, it is worth understanding that countries used nation-building to homogenize populations prior. While the above paper explains why some countries employed nation-building policies, Barry Posen (1993) writes about the close relationship between nationalism and warfare. He argues that nationalism causes countries to undertake foreign policies that initiate war, or prolong and intensify war, through various mechanisms. The argument is that the intensification of warfare through nationalism is accomplished by allowing the state to mass mobilize the “creative energies and the spirit of self-sacrifice of millions of soldiers”, as nationalism is used as a tool by states to enhance their military capabilities (Posen, 1993). He also uses historical evidence, particularly of France and Prussia/Germany, to show how conscription rules changed overtime, and the affect these changes incurred on the building of a mass army and nationalism. In some instances, conscription intensified solidarity among youth, while in other cases, where exemptions to conscription were possible, it failed to intensify nationalistic sentiments (Posen, 1993). It is worthy to note that Posen focuses on the role of education in building nationalistic sentiments. For example, he notes that in France, children were taught patriotism through understanding that their main duty was to defend their nation, and army personnel were just like them (Posen, 1993). With regards to conscription, Alesina, Reich and Riboni (2017) write about how states switched from mercenaries to mass armies via conscription in the late 18th century. Though people faced punishments if they did not comply and defected, the authors claim that wars cannot be won with unmotivated Tenth Anniversary | 97

Dhivesh Dadlani

soldiers. Thus, the elites reduced rents and provided public goods to make citizens voluntarily comply with conscription. This created an understanding amongst citizens and soldiers that losing war meant losing public goods, “which they learned to appreciate because of nation-buiding” (Alesina, Reich and Riboni, 2017). Essentially, increasing the value of public goods through instilled patriotism and indoctrination motivated soldiers to fight for their country. However, this paper focuses on citizens who voluntarily comply with conscription, and does not consider draw a link between conscription and nationalism, or nation-building. The latter is the relationship explored in my paper. My paper is similar to the first quantitative papers described in this section in that it will explore the observable variation in nationalistic sentiments within and across countries. However, the majority of the papers discussed only explore the determinants and advantages of nation-building and nationalism. Their conclusions heavily depend on historical arguments, leaving much of the conclusions theoretical. They fail to provide the empirical evidence, which is what I aim to provide through my study and analysis. Data Description The primary source of data for individual analyses comes from the World Value Survey (WVS) longitudinal dataset. This dataset contains responses to surveys conducted in 100 countries over the time-period of 1981 to 2014. The primary variable used as a proxy or indicator of Nationalism is the answer to question G006: How proud are you to be [Nationality]? This variable is referred to as Nationalism or National Pride. In exploring the determinants of nationalism on an individual level, I run regressions on the ten following variables: income, education, a dummy variable for active in the military, gender, age, confidence in government, general trust levels, religiosity, interest in politics, and marital status. These variables arise from different questions from the WVS. Appendix 1 provides a summary of these questions. Although each of the questions is taken directly from the survey, the variables in my study are coded differently from the original WVS coding. While 1, 2, 3 and 4 respectively represented “Very Proud”, “Quite Proud”, “Not Very Proud” and “Not at All Proud” in the original documentation of G006, my study employs reverse coding for the purpose of easier diagram interpretation. Further, all missing values for these questions are dropped. Values could be missing due to the question not being asked, the respondent not knowing the answer to the question, or the respondent declining to answer the question. Data for country level analyses come from WVS, World Development Indicator (WDI), UCDP Monadic Conflict Onset and Incidence Dataset, Major Episodes of Political Violence and Conflict Region (1946-2016), the Polity IV project, Conscription as Regulation (Mulligan and Shleifer, 2005), 98 | Columbia Economic Review

The Determinants Of Nationalism

and The Economic Consequences of Legal Origins (LaPorta, Lopez-de-Silanes and Shleifer, 2008). Appendix 2 provides a summary of the key variables used in each of the datasets cited above. For conscription data, Mulligan and Shleifer (2005) was the primary source of data for country-year pairs before 2000. Conscription was coded 1 if there was more than 1 draft month in that year. Due to data limitations, conscription is coded 1 when conscription is selective, lottery-based, or compulsory. Additionally, since Mulligan and Shleifer only provided data for years 1970-2000 in five year intervals, I cross-referenced any observations beyond the scope of their dataset with the CIA World Factbook and www.globalsecurity.org to check for any change in conscription laws in the respective country during that time-period. For example, according to Shleifer and Mulligan, Albania had 12 months of conscription in 1995. Since the observation in my dataset is for year 1998 and Shleifer and Mulligan do not provide an estimate of year 2000, I again used the cross-referencing method. This allowed me to identify that conscription laws were removed in Albania in 2010; thus, conscription was coded as 1 for Albania-1998 in my dataset. In the case of any missing information from those data sets, conscription is left as a missing value. Consequently, that country-year pair is omitted from my analysis. Individual Level Determinants of Nationalism To understand the determinants of nationalism, I use all the data from 100 countries and in all waves of the WVS. The repeated cross-section regression framework is shown below: Yict = λt + γc + X’ictβ + εict where Yict is the nationalism level of individual i in country c at time t, γc and λt are the country and year fixed effects respectively, and Xict are the independent variables income, education, gender, age, confidence in the government, general trust levels, interest in politics and dummies for marital status as well as whether an individual is a member of the armed forces. To understand the data and variation in nationalism, I refer to Figure 1. This figure shows a clear positive relationship between National Pride and age in 4 difference countries. The strength of the relationship clearly varies across countries. For instance, in Turkey, every person surveyed between the age of 79 to 83 in Wave 4, responded Very Proud to the question about national pride. In the above regression framework, I control for the national differences and time trends through the country and year fixed effects, thus only comparing individuals within a country in a specific year, and repeating this for every country and every year in the dataset.

Tenth Anniversary | 99

Dhivesh Dadlani

Table 1 below shows the results of the above regression on various independent variables. Income seems to have a marginal effect on nationalism, both in a bivariate regression as well as regressions with controls. Education levels are negatively correlated with nationalism. Both sex and trust do not have a significant effect. Age, government confidence, religiosity and interest in politics have strong positive significant relationships with nationalism. Being married and serving in the military are also positively correlated. Most of these results are not shocking. Older people tend to be prouder of their country. Military personnel are either serving because they are patriotic or have become patriotic due to military service. Having an interest in politics means that one would be interested in affairs on a national level, which could explain the relationship observed. On the other hand, there are some results that are unexpected. I would fully expect that if people trusted others in their country, there would be a higher chance that they would be patriotic. A possible explanation is that when people are asked, â&#x20AC;&#x153;Generally speaking, would you say that most people can be trusted or that you need to be very careful in dealing with people?â&#x20AC;?, they recall people they spend time with rather than people in general. Though people could trust their close circle, they may not necessarily be proud of their nation at large, in which case we cannot say much about the correlation between these two variables. Another interesting finding is the relationship between education

100 | Columbia Economic Review

The Determinants Of Nationalism

and nationalism. It seems that as people pursue further education, they lose a sense of pride for their nation. Perhaps, this is because students typically learn about their country before high school and as they proceed, they forget their education about their nation and focus more on specific topics that interest them. However, an in-depth analysis of this finding is out of the scope of this paper. Nevertheless, from a nation-building policy perspective, governments can consider revisiting their syllabus and introducing topics that are meant to build a sense of national unity in students.

Tenth Anniversary | 101

Dhivesh Dadlani

Ben Enke (2018) writes about the role of morals in the US 2016 presidential elections. He describes the fine line between individualizing values and communal values. Individualizing values are “prescriptive judgements of justice, rights and welfare pertaining to how people ought to relate to each other” (Enke 2018). Communal values on the other hand are concerned with loyalty and obedience to a hierarchy. It relies heavily on notions of “us” and “them” and “applications of moral principles depend on context” (Enke 2018). Concretely, harm, care, fairness and reciprocity are individualizing values while respect, authority, loyalty and in-group such as family or country are communal values. Table 2 presents the relationship between nationalism and certain cultural values. In some sense, nationalism can be seen as the ultimate form of moral communalism and we expect there to be a positive relationship between those who show moral communalism and those that are nationalistic. We can see that there is indeed a positive relationship between moral communalism and nationalism from columns 1, 2 and 3. People who believe in the importance of family as well as duties to parents and children are on average more nationalistic. Conscription and Nationalism - Evidence from Spain Removal of conscription in Spain makes for a good natural experiment to understand how conscription affects nationalism. Though the ideal setting would be to study a country that decided to introduce conscription at some point, the data from the WVS does not contain any country that introduced

102 | Columbia Economic Review

The Determinants Of Nationalism

conscription between the survey period. However, in 2001, Spain decided to abolish the law that mandated young men to serve for nine months in the military. The Defense Minister gave a speech in 2001 telling the public that by December 31st 2001, men would no longer need to serve their nation. Military service would be completely voluntary and men can declare themselves objectors until one day before their official service period (Elmundo, Wikipedia). As context, women were never required to serve in the military. Men who turn 18 before 2001 would have needed to serve in the military, but men who turned 18 in or after 2001 would never be mandated to, except for times of national emergency (CIA World Factbook). Given that I have data on Spain in 2007, I can compare the national pride of groups that were affected by this law and groups that were not. Specifically, men aged 24 and below in 2007 would never have been required to serve. Men aged 25 and above would have been required to serve. On the other hand, females of all ages were never required to serve. Given this, using data from 2007, I use the following difference-indifference framework to analyze the effect of not having to serve in the military, on nationalism: Yi = β0 + β1Malei + β2Below25i + β3(Below25i * Malei) + εi where Yi is the nationalism level of individual i, Malei is a dummy variable indicating if an individual is male and Below25i is a dummy variable indicating if an individual is below 25 years of age in 2007. Essentially, we are taking the difference between females aged 25 and above and females 24 and below, and comparing it to the difference between males aged 25 and above and males 24 and below. I restrict the sample to males between the age 21-29 so that the groups I compare are almost similar in age. An underlying assumption here is that this difference should not exist, in the absence of this law.

Tenth Anniversary | 103

Dhivesh Dadlani

As we can see from the Figure 2, males are on average, in 2007, more patriotic if they are part of the younger group than the older group while the evidence for females is consistent with the earlier finding that age tends to correlate positively with nationalism. To test the assumption that there should be no difference between the differences in males and the differences in females, in the absence of this law, I apply the same framework using data in 1995 and 2000. Figure 3 and Table 3 present the findings.

104 | Columbia Economic Review

The Determinants Of Nationalism

The results from Figure 3 and Table 3 illustrate two important points. First, from Figure 3, we can see that the difference across males and females have followed the same trend from 1995 to 2000. Second, there are no significant differences between differences in females and differences in males, in both 2000 and 1995. As such, under the assumption that there would be no difference in the absence of the law, there is a positive casual effect of not having to serve on nationalistic sentiments. This does not provide evidence that conscription always negatively impacts nationalism. Rather, it provides evidence that if a country already requires people to serve, then getting rid of that law could increase nationalistic sentiments of people who were going to serve, but did not have to due to the law. This distinction is extremely important. I argue that the following factors caused this change in sentiments. Conscription takes away the right to choose if one wants to serve. Many conscripts never actually want to serve but do so because they must. Once potential conscripts expect that they have to serve, they get conditioned to it and when they are told that they no longer will need to, they would feel more patriotic towards their nation. Further, for people just above the cutoff, there is a sense of resentment that they were not â&#x20AC;&#x153;sparedâ&#x20AC;? and were made to serve. This could lead to them feeling less proud of their country. These two effects combined could cause the difference to become larger, thus explaining the significant positive effect of removing conscription on nationalistic attitudes. Country Level Determinants Data from the WVS allows us to understand the variation of nationalism in individual countries. Given that there is data about different economic indicators and political events, we can analyze how nationalism correlates with other factors on a country level. Again, to understand the data from a country perspective, I refer to Figure 4. These figures show the relationship between various factors (trade openness, general trust levels, general confidence in the government and religiosity) and average national pride in a country.

Tenth Anniversary | 105

Dhivesh Dadlani

Table 4 shows the results of simple regressions of national pride on some economic factors. GDP per capita does not have any effect on nationalism. Trade has a small negative effect on nationalism. I argue that if countries are more globalized and economically integrated, citizens tend to have pride for the world at large rather than pride for their country. Further, much evidence points to the fact that when countries have more nationalistic sentiments, they tend to introduce protectionist policies to promote job growth within their country. Taxes have a slightly positive effect on nationalism, though this effect is marginally significant and not very large. Table 5 shows us the relationship between nationalism and incidences of conflict, both civil and international. As we can see, national pride is correlated with all types of conflict. The relationship between international violence and nationalism can partially be explained by the fact that the nation rallies together against the enemy. Further, Alesina et al. (2017) write that in times of war, states typically engage in nation-building activities as well as negative propaganda against the enemy, thus creating a stronger sense of nationalism. On the other hand, the strong positive relationship between civil war and intrastate conflict with the government is surprising. I offer two explanations for this. First, governments may try to influence nationalism and create a national identity as a response to civil conflict. Second, individuals answering the questions may be proud of only the side that they are for and consider that side their nation, instead of looking at the nation at large. An in-depth analysis of this is, unfortunately, out of the scope of this paper. 106 | Columbia Economic Review

The Determinants Of Nationalism

Table 6 presents evidence of relationships between nationalism and political factors. There is no significant relationship between the strength of a democracy or legal origins on nationalism. There seems to be a strong negative relationship between trust and nationalism, as well as conscription and nationalism. Where average confidence in the government and religiosity are concerned, coefficients are strong and positive.

Tenth Anniversary | 107

Dhivesh Dadlani

108 | Columbia Economic Review

The Determinants Of Nationalism

Table 7 presents the evidence of the relationship between nationalism and cultural factors as discussed in the individual level analysis. Figure 5 depicts these relationships. We can see that the correlation between moral values and nationalism still holds when we exploit variation at the country level. Before explaining any of these findings, let us understand how these effects change when we control for certain factors. Table 8 presents the results of multivariate regressions controlling for wave effects. We can see that the relationship for all conflict indicators, income or trust, are only either marginally significant or not significant at all, indicating that it was the correlation with other factors that was explaining the previous relationships. However, the coefficients on other factors such as trade, religiosity, confidence in government and conscription are extremely significant and are in the same direction as the correlation estimates. An interesting finding here, that provides support to the analysis on individuals, is that nationalism in negatively correlated with conscription. However, it is important to note that the interpretation here is very different. In this analysis, I find that countries that have mandatory conscription are on average, less patriotic than countries that do not have conscription. This finding is interesting because it is not a natural experiment that gives light to this finding; we are essentially comparing average nationalism across countries controlling for other factors and wave effects. Surely, the hypothesis about peopleâ&#x20AC;&#x2122;s expectations of conscription cannot play a role here. I argue that nationalism is a function of the freedom one gets in their country and their exposure to nation-building activities. As individuals are given less freedom to decide how they can spend their life, when they can pursue further education and where they can go, they tend to love their nation less. However, when Tenth Anniversary | 109

Dhivesh Dadlani

individuals are exposed to nation-building activities like conscription, propaganda, national day celebrations and compulsory singing of national anthems, they tend to feel prouder of their nation. Mandatory conscription is an event where these two effects are opposing each other. On one hand, individuals are being forced to train and fight for their country and on the other, they are being stripped of the liberty to decide how they would spend those years otherwise. Clearly, the effect of having less liberty dominates for the evidence above. In Spain, people were given back their liberty and they became more patriotic. Further, countries where men have this liberty and more patriotic than those that do not.

110 | Columbia Economic Review

The Determinants Of Nationalism

Conclusion Relationships between nationalism and a variety of economic, cultural and political factors are strong and significant. There is a lot of variation in nationalism both on the individual as well as the country level. Nationalism can be beneficial for a country, especially in times of war, as much of the literature has discussed. It could also be harmful and lead to economically damaging policies such as protectionism. In this paper, I make no claim about whether nationalism is beneficial or harmful, but I believe that if policy makers are looking to engage in any sort of nation-building policy, then it is worth understanding what high and low levels of nationalism are associated with. Further studies could explore the reason behind why nationalism is positively related with factors like religiosity, age and political interest, and negatively related with trust and trade openness. As of now, we only know that empirically, this is the way they correlate. Where conscription is concerned, I find that the removal of conscription in fact increases national pride and I argue that the mechanism is through the liberty one gets when they do not need to serve. The evidence from Spain’s natural experiment is consistent with evidence from the country level data. My first hypothesis that conscription increases nationalism has turned out, in some ways to be false. As for the other hypotheses, aside from income, trust and conflict, the rest of the correlations have been confirmed by the data. Further research will seek to explore why trust and conflict are correlated with national pride. REFERENCES Ahlerup, Pelle, and Gustav Hansson. “Nationalism and Government Effectiveness.” Journal of Comparative Economics 39, no. 3 (2011): 431-51. doi:10.1016/j.jce.2011.05.001. Shulman, Stephen. “Exploring the Economic Basis of Nationhood.” Nationalism and Ethnic Politics 9, no. 2 (2003): 23-49. doi:10.1080/13 537110412331301405. Alesina, Alberto, and Bryony Reich. “Nation Building.” 2013. doi:10.3386/ w18839. Posen, Barry R. “Nationalism, the Mass Army, and Military Power.” International Security 18, no.2 (1993): 80. doi:10.2307/2539098. Alesina, Alberto, Bryony Reich, and Alessandro Riboni. “Nation-Building, Nationalism and Wars.” 2017. doi:10.3386/w23435. Mulligan, Casey, and Andrei Shleifer. “Conscription as Regulation.” 2004. doi:10.3386/w10558.

Tenth Anniversary | 111

Dhivesh Dadlani

Enke, Benjamin. “Moral Values and Voting: Trump and Beyond.” 2018. doi:10.3386/w24268. “Contact CIA.” Central Intelligence Agency. April 01, 2016. Accessed May 03, 2018. https://www.cia.gov/library/publications/theworld-factbook/ SemiColonWeb. “GlobalSecurity.org.” GlobalSecurity.org. Accessed May 03, 2018. http://www.globalsecurity.org/. World Development Indicators — DataBank. Accessed May 03, 2018. http:// databank.worldbank.org/data/reports.aspx?source=WorldDevelop ment-Indicators. “WVS Database.” WVS Database. Accessed May 03, 2018. http://www.worldvaluessurvey.org/. SemiColonWeb.“GlobalSecurity.org.” GlobalSecurity.org. Accessed May 03, 2018. http://www.globalsecurity.org/. Polity IV Project: Country Reports 2010. Accessed May 03, 2018. http:// systemicpeace.org/polity/polity4.htm “UCDP.” UCDP - Department of Peace and Conflict Research - Uppsala University, Sweden. Accessed May 03, 2018. http://www.pcr.uu.se/ research/UCDP/. “Military Service.” Wikipedia. May 02, 2018. Accessed May 03, 2018. https://en.wikipedia.org/wiki/Militaryservice

112 | Columbia Economic Review

The Determinants Of Nationalism

Tenth Anniversary | 113

Dhivesh Dadlani

114 | Columbia Economic Review

The Determinants Of Nationalism

Tenth Anniversary | 115

THE EFFICACY OF BIOMEDICAL RESEARCH: A BIBLIOMETRIC ANALYSIS ACROSS LONGITUDINAL RESPIRATORY DISEASE DATA

Zoey Chopra Columbia University Abstract: I estimate the effects of innovation on tangible health outcomes. I investigate the relationship between cumulative research publication count and health outcomes for multiple respiratory diseases, measured by four indicators: (1) mortality rate, (2) length of hospitalization, (3) discharge rate, and (4) disability-adjusted life years (DALYs) rate. Annual incidence and expenditures for each disease are controlled using the framework of health production functions. Estimates indicate that mortality, hospitalization, and DALYs rates are weakly and inversely related to growth in unsupported publications, while discharge rate is positively related. This association strengthens as lagged effects of cumulative publication count are considered, especially in the time frame of 6 to 11 years post-publication. Counterintuitively, relations between growth in supported publications and outcome variables are directionally opposite, potentially a result of reverse causality. In the long-run, investigation may be expanded to include the effects of differences in regional production and medical innovation uptake.

mprovements in healthcare are generally attributed to biomedical research at a core level. From developments in our understanding of bodily molecular mechanisms to innovation in pharmaceutical products, biomedical research is often thought to form the backbone of healthcare improvements. The National Association for Biomedical Research (2015) notes that the very “questions of biological uncertainty... are necessary and critical to the advancement of both human and animal health.” Direct advancements stemming from wet lab research include immunizations for polio, diphtheria, mumps, rubella, and hepatitis, among others, to the benefit of countless people worldwide (Foundation for Biomedical Research 2016-17). Though not all research yields commercial application, the Federation of American Sciences for Experimental Biology (2017) notes that “clinical research... still greatly benefits society,” as in cases of innovation toward orphan disease treatments and of comparative best practices studies. As concluded by the National Center for Health Statistics (2015), research at the National Institute of Health (NIH) alone has contributed to reductions in the mortality rate of four of the ten leading causes of death in the United States: heart disease, cancer, stroke, and diabetes. Tenth Anniversary | 117

Zoey Chopra

Academic literature confirms the significance of the effects of biomedical research on population longevity and health. Historical analysis notes, “Since 1945, biomedical research has been viewed as the essential contributor to improving the health of individuals and populations, in both the developed and developing world” (Moses III and Martin 2011). Clinical insight also finds that “biomedical research provides the basis for progress in health and health care. Basic discoveries, translation to clinical medicine, and implementation into medical practice have been the story line of medical advances for decades” (Nabel 2009). A series of qualitative, case-oriented studies and surveys have revealed the effects of research-derived, long-term improvements for a range of health conditions (NIH 2016).

Health Production Function

Biomedical research is generally thought to be effective whilst operative in the following chain:

The majority of relevant topic literature, both in economics and in health, has focused on two intermediate linkages: (1) research to innovation, and (2) innovation to health. Previous research has found that “much of the developmental work and early research on various types of organ transplantation, diagnostic imaging technologies, prosthetic joints, and gene therapy have occurred in AHCs [Academic Health Centers]” (Anderson 1994, Matherlee 1995). Sampat and Lichtenberg’s (2011) analysis of patent and bibliometric data has confirmed that pharmaceutical innovation benefits from publicly-funded research, albeit more indirectly (drug development builds from previously-conducted relevant research) than directly (drug development results from funded research). Similar studies confirm the implications of the first linkage, research innovation. Two main categories exist within biomedical innovation: embodied growth (physically manifested) and disembodied growth (intangibly manifested).

118 | Columbia Economic Review

The Efficacy Of Biomedical Research

Research streams often investigate the tangible effects of biomedical innovation on metrics like mortality and quality of life. Recently, the Montreal Economic Institute found three major benefits of pharmaceutical innovation: longevity and health, cost savings, and increases in labor productivity (Lichtenberg 2016). Empirical economic research has shown similar relationships between product innovation and health outcomes in the United States, Switzerland, France, and Australia, using both government and individual patient data (Lichtenberg 2014a, Lichtenberg 2014b, Lichtenberg 2013a, Lichtenberg 2015). Improvement in this tangible vein, often concerning the development of physical products (i.e. drugs and medical devices), is commonly referred to as embodied growth. There is an emphasis on studying embodied growth for two reasons. First, analysis of embodied growth is more concrete. Since biomedical products yield definitive outcomes and require evaluative processes like clinical trials (short-term and long-term), assessment in controlled environments already occur, providing reliable and accessible data. Second, research towards embodied growth is more important to the largest contributors of privately-funded research, which include biotechnology, pharmaceutical, and medical equipment companies (IMC 2004). As illustrated below in Figure 1, these private research investments, largely directed towards short-term and commercially-driven product development, have surpassed public funding for biomedical research since the early 2000s (IMC 2004, WHO 2014).

Tenth Anniversary | 119

Zoey Chopra

As a result, economic literature largely has yet to explore the realm of disembodied growth, the innovation that is not physically manifested. And yet, disembodied growth is critical when evaluating the second linkage: (2) innovation health. Outside of research in the basic sciences, existing literature has shown that incentives that reward long-term success and do not necessitate pre-defined deliverables—both of which are characterizations of disembodied—often result in greater innovation (Azoulay et al. 2011). A large area of disembodied growth also stems from physician-created clinical practices (diagnostic procedures and surgical techniques) that do not manifest physically, yet are crucial for understanding improvements to clinical outcomes (Gelijns et al. 2001, McKinlay 1981, Coleman et al. 1957). For instance, the incorporation of palliative care into medical practice has had immense benefit toward patient health outcomes across ages and diseases (Smith et al. 2012, Hays et al. 2006, Gore et al. 2000). A last major consideration for disembodied growth is that of biomedical education, including health professionals and research practices, which can help secure long-reaching improvements to health outcomes caused by growth in both disembodied and embodied innovation. Rather than try to delineate the boundary between embodied and disembodied growth, since each area of innovation likely feeds the other, I instead aim to estimate the innovation factor (i.e. total factor productivity) as a whole, within the context of a health production function. As such, my research investigates the causal impact of ideas (proxied by publication count) on tangible health outcomes. This paper will assess this relationship econometrically by examining changes in long-run growth of research versus long-run changes in pivotal health indicators. Drawing from discussions of trifold endogenous growth—dependent upon ideas, labor, and capital (Romer 1990, Comin 2008, Comin and Mulani 2009)—and building from Lichtenberg’s (2013b) cancer-based, bibliometric model, I will estimate a health production function that focuses on the contribution of ideas towards improved health outcomes. In his paper outlining my theoretical model of choice, Lichtenberg (2013b) summarizes the efficacy of bibliometric analyses in sectors such as agriculture and manufacturing in order to indicate the positive relationship between ideas (which I take to be embodied by research, and proxy by longitudinal publication counts) and productivity. I take the success of these prior studies, as well as Lichtenberg’s (2013b) bibliometric analysis of cancer data in the United States, as my starting point. Yet, there are difficulties in evidence-based linkage between biomedical research and healthcare improvements, as Nabel (2009) notes: “It is difficult to draw a straight line between NIH-funded research and improvements in health care, many basic science advances take decades before being fully developed into therapeutics and fully implemented into medical practice, and reporting of 120 | Columbia Economic Review

The Efficacy Of Biomedical Research

of medical advances rarely includes attribution to NIH fundingâ&#x20AC;? (p. 2858). Most criticisms of the bibliometric approach focus on health impact: â&#x20AC;&#x153;A fundamental defect underlying studies of this kind, however, is the subjective nature of assigning relevance. A count of published studies of specific drugs cannot be used to assign values to relative public and private contributions to their development, because the relevance of the studies to the ultimate approval of the drugs often cannot be determinedâ&#x20AC;? (Reichert and Milne 2002). In other words, because pharmaceutical innovation generally benefits indirectly from scientific research, it appears unsubstantiated to claim that changes in health outcome are a direct product of research. Even Cockburn and Henderson (1998), who similarly use publication count to determine connectedness between privately and publicly-funded pharmaceutical research, acknowledge the difficulties of assigning impact to categories of research (i.e. publicly-funded research). My research circumvents these methodological concerns with a broader lens the innovation factor, encapsulating both embodied and disembodied growth.

Tenth Anniversary | 121

Zoey Chopra

The diseases my research focuses on are encapsulated within the broader category of Respiratory Diseases. As Figures 2 and 3 indicate, two of the leading causes of death in the United Statesâ&#x20AC;&#x201D;chronic lower respiratory diseases and influenza & pneumoniaâ&#x20AC;&#x201D;fall within this category. Between 1999-2013, these respiratory disease subsets alone accounted for nearly 8% of all deaths in the U.S. As Figure 3 more illustrates, there is variability in respiratory disease trends over time. In the same time-frame, where mortality rates for heart disease declined by 6.9%, mortality rates for chronic lower respiratory diseases declined by only 0.2%, and by only 0.5% for influenza & pneumonia. As we contrast these trends with those of the rising federal research and development (R&D) expenditures below in Figures 4 and 5, it is striking how drastically mortality rates for heart disease have decreased relative to these two categories of respiratory diseases, and, by extension, how unevenly R&D effects tangible health outcomes in different disease states. The scope of rising costs amidst changing demographics and aging populations raises a fundamental question: How effective is medically-relevant research towards tangibly beneficial health outcomes? Given the rapid increases in funded biomedical research for lung-associated conditions, my research hopes to explore this question within the specific context of respiratory diseases. Understanding the scope of the ideas behind medical innovation is critical when considering the individual lives at stake, R&D investments, and broader health policy decisions.

122 | Columbia Economic Review

The Efficacy Of Biomedical Research

Econometric Model Metrics for assessing tangible health outcomes for at-risk patient populations generally fall into two large categories: (1) survival rates and (2) mortality rates. Survival rates indicate the survival of a proportion of a diagnosed patient population within a certain time frame. Survival rates are therefore conditional mortality rates defined by diagnosis. Mortality rates, on the other hand, indicate the death of a proportion of a patient population in a certain time frame. Mortality rates are therefore unconditional, similarly defined by cause of death but without the requirement of pre-existing diagnosis. Both metrics are widely cited. While both variables appear to be viable indicators for determining changes in tangible health outcomes over time, we opt to use the unconditional mortality rate as our dependent variable. This is primarily due to three concerns about survival rates: (1) protopathic bias, (2) length time bias, and (3) lead time bias. Protopathic bias occurs when treatment is prescribed for symptoms of a condition before formal diagnosis. As a result, the symptoms themselves appear to cause the condition (Tamim et al. 2007). Changes in survival rates may be only be reflections of this bias, because people diagnosed at time t who remain surviving at time t+1 may be alive because of previous treatment at time t-1 rather than at time t. Length time bias identifies the notion that slower-paced conditions with longer asymptomatic periods are often overrepresented in diagnosed population samples (usually from screening programs), resulting in larger-than-accurate survival rates for such samples (Family Practice Notebook 2017). Tenth Anniversary | 123

Zoey Chopra

Lead time bias occurs when say two different tests for a condition are compared side-by-side, and the newer test leads to earlier diagnosis without any changes to health outcome. This result could indicate that the new test in fact prolonged survival rates, when in fact the only difference in the new test relative was an earlier diagnosis (Lichtenberg 2013b). Previous research studying survival rates for cancer patients, for example, has attested that comparing survival rates side-by-side across time and/or space can result in misleading conclusions, even if newer diagnostic procedures and technologies were wholly inconsequential (Welch et al. 2000). To avoid the various biases possible when considering survival rates, and to ensure that changing diagnostic standards do not affect results, I will primarily analyze unconditional mortality rate as my dependent variable. Endogeneity concerns, as discussed in a later note (see â&#x20AC;Ą), give reason for us to prefer unconditional mortality rate as our dependent variable. The Respiratory Disease (RD) mortality rate can be defined as follows: P(death from RD). Econometric Assumption via Law of Total Probability By the Law of Total Probability, we can extrapolate the following to simplify our model:

(1) P(death from RD) = [P(death from RD | RD diagnosis) * P(RD diagnosis)]+ [P(death from RD | no RD diagnosis) * (1 - P(RD diagnosis))] We can assume that P(death from RD | no RD diagnosis) ~ 0. Thus, Eq. (1) becomes: (2) P(death from RD) â&#x2030;&#x2C6; [P(death from RD | RD diagnosis) * P(RD diagnosis)]

Taking natural logs of Eq. (2), we obtain:

(3) ln[P(death from RD)] â&#x2030;&#x2C6; ln[P(death from RD | RD diagnosis)] + ln[P(RD diagnosis)]

124 | Columbia Economic Review

The Efficacy Of Biomedical Research

To examine Eq. (3) econometrically, via bibliometric analysis, I hypothesize similarly to Lichtenberg’s (2013b) bibliometric cancer model, and propose that the former term—RD mortality rate conditional upon RD diagnosis—maintains an inverse relationship with research relevant to RD. To proxy the extent of RD-relevant research, I assume that scientific publication counts can be used as indicators for the broader swath of RD-relevant research. Applying this assumption, we achieve the following proxy for conditional mortality rate: (4)

ln[P(death from RD | RD diagnosis)] = (3 ln(publicationt-k)

(5)

ln[P(RD diagnosis)] = y ln(incidencet)

Substituting our proxy, Eq. (4), into Eq. (3), we can derive:

(6)

ln[P(death from RD)] ≈ (β ln(publicationt-k) + ln[P(RD diagnosis)]

In order to estimate the effects of RD-relevant research on RD-conditional mortality rate, I propose to estimate Eq. (5) via bibliometric regression analyses. I will assess my mortality model across longitudinal data for a swath of respiratory diseases, controlling for annual hospital expenditure: (r1) ln(mortalitydt) = β ln(publication d, t-k) + γ ln(incidence dt) + ζ ln(expenditure dt) + αd + δt + εdt My variables are defined as follows: mortalitydt = age-adjusted mortality rate for specific RD (d) in given year (t) publicationd,t-k = cumulative (d)-relevant research articles published until end of year (t-k) incidencedt = age-adjusted incidence rate for specific RD (d) in given year (t) expendituredt = expenditure for specific RD (d) in given year (t) αd = disease-fixed effect for specific RD (d) δt = time-fixed effect for given year (t) εdt = residual perturbation of regression model, Eq. (r) Based on Eq. (r1), we can estimate models at different levels of time lag (k) across specific respiratory diseases (d), and can analyze differences in effect by publication type, from government-funded to privately-funded to unsupported research publications. Tenth Anniversary | 125

Zoey Chopra

The largest concern of my proposed framework is that it cannot account for exogenous factors influencing mortality rate. Changes in lifestyle may affect the mortality rates of respiratory diseases. Smoking, for instance, has been shown to exacerbate flare-ups in the symptoms of Chronic Obstructive Pulmonary Disease (Martin 2015). At the same time, however, extensive research has shown that exogenous lifestyle behaviors like smoking do not result in significant difference in symptomatic recovery, even for the most severe chronic respiratory diseases (Seemungal et al. 2000). To control for the effects of behavioral factors in the outcome variable (mortality rate), I have included fixed effects in my regression model across time and disease. Another potential concern of my regression framework is the inclusion of expenditure. While expenditure may be considered endogenous to the dependent variable in that the amount spent on treatment/time is correlated with probability of death, evidence suggests little correlation between increased expenditure and reduced mortality rates (Rothberg et al. 2010, Chen et al. 2010). Given that endogeneity appears to be of little concern, I include expenditure in my regression framework as a viable and necessary input within the health production function. Alternatives to Mortality Rate While unconditional age-adjusted mortality rates offer an absolute, empirical way of quantifying longitudinal differences in health outcomes, they too paint a skewed picture of the national population (CDC MMWR 1986). Particularly for conditions that do not contribute directly toward mortality, including many respiratory diseases, mortality rates generally depend on disease progression in elderly populations within a given year (CDC MMWR 1986). As such, it is not always clear whether the underlying cause of death is the condition (in this case, a specific respiratory disease), or the circumstance of age itself. The Center for Disease Control and Prevention (CDC) has given tabulated consideration to alternate premature mortality measures for this very reason. Years of Potential Life Lost (YPLL) is perhaps one of the more common measures of premature mortality in health economics, especially for conditions concerning occupational medicine (Romeder and McWhinnie 1977, Yoshida et al. 1997). However, given concerns for data accessibility and reliability, I instead turn to differences in longitudinal hospitalization and discharge rates, and disability-adjusted life years (DALYs). Hospitalization Rate Hospitalization rate is defined as the mean length of stay in-hospital for a patient of a given condition. This measure provides a reliable, alternate approach to assessing changes in tangible health outcomes, and previously has 126 | Columbia Economic Review

The Efficacy Of Biomedical Research

been shown to be more effective than mortality rate when considering respiratory diseases. Studying instances of pneumonia in Denmark between 1997 and 2011, Sogaard et al. (2014) found that although average 30-day mortality rates remained near constant across age categories (with slight increase in the population > 80 years), hospitalization rates were drastically different. Studying instances of pneumonia in the US from 2004 to 2006, Drye et al. (2013) concluded that in-hospital mortality statistics vary extensively across hospitals, dependent upon variance in hospitalization rate. These findings reflect the extent to which advancements made towards conditions that are not primary contributors to death (like many respiratory diseases I study in this paper) may be better studied with measures like hospitalization rate, in parallel with the more traditional measure of unconditional, age-adjusted mortality rate. An important benefit to using hospitalization rate relative to mortality rate is the magnitude of observed longitudinal difference. Because mortality is incredibly dependent on circumstance (as discussed above), degrees of difference over spans of time tend to be small, unless substantial technology/ treatment changes occur in a given year. Assessing national trends in mortality versus hospitalization from 1999 to 2013 in the US, Krumholz et al. (2015) found that for all Medicare beneficiaries, the all-cause mortality rate declined by about 0.85%. For Medicare fee-for-service beneficiaries, the hospitalization rate declined by about 23.65%. While there may be discrepancy in comparing all Medicare beneficiaries with Medicare fee-for-service beneficiaries, there are clear statistical benefits to exploring hospitalization rate in tandem with mortality rate. Similar to the approximation used previously, by the Law of Total Probability, we can simplify our model accordingly: ln[P(hospitalization from RD)] ≈ β ln(publication t-k) + γ ln(incidencet)

I control for annual incidence in order to estimate the innovation factor of the health production function on hospitalization rate without the influence of variance in the number of cases per year. This control was performed because the total number of cases may affect the length of stay any hospital can devote to a given patient. Similarly, in order to estimate the effects of RD-relevant research on RD-conditional hospitalization rates, I will assess my model across longitudinal data for a swath of respiratory diseases, controlling for annual hospital expenditure: (r2) ln(hospitalization dt) = β ln(publication d, t-k) + γ ln(incidence dt) + ζ ln(expenditure dt) + αd + δt + εdt Tenth Anniversary | 127

Zoey Chopra

My variables are defined as follows: hospitalizationdt = hospitalization rate for specific RD (d) in given year (t) publicationd,t-k = cumulative (d)-relevant research articles published until end of year (t-k) incidencedt = age-adjusted incidence rate for specific RD (d) in given year (t) expendituredt = expenditure for specific RD (d) in given year (t) Îąd = disease-fixed effect for specific RD (d) Î´t = time-fixed effect for given year (t) Îľdt = residual perturbation of regression model, Eq. (r) Similar to mortality rate, a potential concern of my regression framework is the inclusion of expenditure. A measure of expenditure may be considered endogenous to our dependent variable of hospitalization rate, given that the amount spent on treatment/time for a given patient is correlated with the length of stay. While surveys do not appear to have been conducted for respiratory diseases in particular, evidence does suggest a positive correlation between increased expenditure and increased hospitalization rates, which may be problematic for the regression model employed (Rahmqvist et al. 2016). Because expenditure is a viable and necessary input within the context of a health production function, I include expenditure in my regression framework. I also compare this model to another model excluding expenditure, in case of endogeneity concerns. Discharge Rate Discharge rate is defined as the proportion of in-hospital discharges for a given condition relative to all in-hospital discharges in a given year. I include this metric in order to provide a reliably- sourced means to assess changes in treatment possibilities for the respiratory diseases in question. Controlling for incidence, an increase in discharge rate from year t to year t+n could mean that treatment has become more effective in those n years, enabling institutions to discharge RD-afflicted patients more readily. A decrease in discharge rate, on the other hand, could also indicate that previous treatments no longer remain effective. In this way, discharge rate may be able to capture the effects of the innovation factor more readily than other measures, since discharge decisions are often directly influenced by treatment response. Perhaps more than measures of mortality rate and hospitalization rate, discharge rate can especially account for embodied and disembodied growth. This is plausible given the direct links between treatment (embodied or disembodied), health outcome, and discharge.

128 | Columbia Economic Review

The Efficacy Of Biomedical Research

However, confounding possibilities do exist when interpreting changes in discharge rate. Namely, discharge rate cannot proxy for readmission rate, defined as the rate at which a patient is readmitted for a specific condition, post-discharge. Discharge rate only reveals how often patients are discharged, and does not provide insight into whether or not these same patients must return at a later date.

Consider two anecdotal examples:

First, a patient comes in for a check-up for a certain respiratory disease. The hospital admits the patient, but sends the patient away due to a lack of resources, and asks that the patient return at a later date. The patient returns, is treated, and is sent away again. In this scenario, discharge rate would ostensibly increase because the same patient has been discharged twice, even though no treatment was received in the first instance. Second, a patient arrives for a check-up for a certain respiratory disease. The hospital determines the patient’s symptoms to be benign yet idiopathic, and, rather than discharge the patient, attempts to treat the symptoms further with new techniques. In this case, discharge rate would appear to decrease because for that certain respiratory disease—one fewer patient has been discharged—even though innovative treatment is being employed. In such cases, the discharge rate may not be reflective of positive influences of the health production function’s innovation factor. Similar to the approximation used previously, by the Law of Total Probability, we can simplify our model accordingly:

ln[P(discharge from RD)] ≈ β ln(publication t-k) + γ ln(incidence t)

I continue to control for annual incidence in order to remove the influence of variance in number of cases of respiratory diseases per year, a number which could affect: (1) frequency of discharge, and (2) interpretation of discharge rate. This is important since the same number of discharges in a year with fewer overall cases would present a higher discharge rate than a year with more cases. Similarly, in order to estimate the effects of RD-relevant research on RD-conditional discharge rates, I will assess my model across longitudinal data for a swath of respiratory diseases, controlling for annual hospital expenditure: (r3) ln(discharge dt) = β ln(publication d, t-k) + γ ln(incidence dt) + ζ ln(expenditure dt) + αd + δt + εdt My variables are defined as follows: dischargedt =

discharge rate for specific RD (d) in given year (t) Tenth Anniversary | 129

Zoey Chopra

publicationd,t-k = incidencedt = expendituredt = αd = δt = εdt =

cumulative (d)-relevant research articles published until end of year (t-k) age-adjusted incidence rate for specific RD (d) in given year (t) expenditure for specific RD (d) in given year (t) disease-fixed effect for specific RD (d) time-fixed effect for given year (t) residual perturbation of regression model, Eq. (r)

Similar to mortality rate and hospitalization rate, a potential concern of my regression framework is the inclusion of expenditure. A measure of expenditure may be considered endogenous to the dependent variable of discharge rate, in that the amount spent on treatment/time for a given patient is correlated with rate of discharge. While surveys tend to focus more on readmission rates than discharge rates, a positive correlation is evident between increased expenditure and post-discharge care, perhaps indicating higher discharge rates at higher cost institutions (Stukel et al. 2012). Because expenditure is a viable and necessary input within the context of a health production function, I include expenditure in my regression framework. I also compare this model to another model excluding expenditure, in case of endogeneity concerns. Disability-Adjusted Life Years (DALYs) Rate Disability-adjusted life years (DALYs) are defined by the Institute for Health Metrics and Evaluation (2018) as “the sum of years lost due to premature death (YLLs) and years lived with disability (YLDs), [i.e.] years of healthy life lost.” The metric aims to “measure the gap between a population’s health and a hypothetical ideal for health achievement” (Gold et al. 2002). DALYs are calculated from a combination of mortality and morbidity estimates, first by estimating years of potential life lost (YPLL), then by weighting YPLL with a disability factor decided by health experts on a disease-by-disease basis, and finally by adjusting these disability-adjusted YPLL with an age-oriented societal value factor. All this is conducted in order to measure the extent to which disabilities wrought from disease at a given age affect societal life (Parks 2014, Gold et al. 2012). There is contention over the ethical implications (since deaths at different ages are treated differently, individuals’ lives at different ages appear unequal), political implications (the metric was created to assess global burden of a given disease as well as direct resource allocation), and standardization coefficients (baselines calculated with data from wealthier countries without factoring in resource accessibility or regional differences) of DALYs (Parks 2014). However, the measure still provides a more nuanced understanding 130 | Columbia Economic Review

The Efficacy Of Biomedical Research

of health outcome relative to mortality rate, adjusting for age based upon societal value and adjusting for extent of disability. Furthermore, DALYs provide an important estimate of disease burden for conditions that are not necessarily fatal (a category within which many respiratory diseases I study in this paper fall). When assessing changes in health outcome, we must be attuned not only to life versus death situations, but also to effects on health in everyday life. Similar to the approximation used previously, by the Law of Total Probability, we can simplify our model accordingly:

ln[P(DALYs from RD)] ≈ β ln(publication t-k) + γ ln(incidence t)

I continue to control for annual incidence in order to estimate the innovation factor on DALYs without the influence of variance in number of cases per year, a value which could affect the disability weighting of a given respiratory disease. Similarly, in order to estimate the effects of RD-relevant research on RD-conditional DALYs, I will assess my model across longitudinal data for a swath of respiratory diseases, controlling for annual hospital expenditure: (r4) ln(daly dt) = β ln(publication d, t-k) + γ ln(incidence dt) + ζ ln(expenditure dt) + αd + δt + εdt My variables are defined as follows: dalydt = rate of DALYs for specific RD (d) in given year (t) publicationd,t-k = cumulative (d)-relevant research articles published until end of year (t-k) incidencedt = age-adjusted incidence rate for specific RD (d) in given year (t) expendituredt = expenditure for specific RD (d) in given year (t) αd = disease-fixed effect for specific RD (d) δt = time-fixed effect for given year (t) εdt = residual perturbation of regression model, Eq. (r) Similar to mortality rate, hospitalization rate, and discharge rate, a potential concern of the regression framework is the inclusion of expenditure. A measure of expenditure may be considered endogenous to our dependent variable of DALYs rate, in that the amount spent on treatment/time for a given patient is correlated with their DALYs: although there is little correlation between expenditure and remaining years of life (another perspective on mortality), higher expenditure on treatment either could Tenth Anniversary | 131

Zoey Chopra

decrease the impact of a given respiratory disease on an individual, and therefore decrease the disability weight for that given respiratory disease. Higher expenditure may also signify greater expected disability due to greater procedural complexity. While surveys tend to focus on conditions other than respiratory diseases, a positive correlation seems evident between expenditure and DALYs (Cortez-Pinto et al. 2010). Estimation Strategies Extrapolating from (r1), (r2), (r3), and (r4) above, we estimate the following regression (Later referred to as “Lag-Dependent OLS”): ln(Ydt) = P ln(publication d, t-k) + γ ln(incidencedt) + ζ ln(expendituredt) + αd + δt + εdt My variables are defined as follows: Ydt = mortality/hospitalization/discharge/DALYs rate for specific RD (d) in given year (t) publicationd,t-k = cumulative (d)-relevant research articles published until end of year (t-k) incidencedt = age-adjusted incidence rate for specific RD (d) in given year (t) expendituredt = expenditure for specific RD (d) in given year (t) αd = disease-fixed effect for specific RD (d) δt = time-fixed effect for given year (t) εdt = residual perturbation of regression model, Eq. (r) The causal implications of the innovation factor, β (proxied by publication count), of our health production function can be estimated in two ways: (1) panel data and (2) dynamic panel data. Since I use a linear model to estimate the innovation factor of the health production function, the best unbiased estimation strategy by which to estimate β is via ordinary least squares (OLS), as per the Gauss-Markov Theorem. I include disease-fixed and time-fixed effects to control for unobserved heterogeneity across disease states and time in years. Thus, we arrive at our first estimation possibility: (1) estimate panel data via OLS. However, there is cause for concern: Our dependent variables may be serially correlated with their lagged values, meaning that there is correlation between Yt and Yt-i. In many ways, this is not so unexpected, the mortality rate from last year does have a bearing on the mortality rate this year, and likewise for hospitalization rate, discharge rate, and DALYs rate. In order to account for this probability of serial correlation, our dependent variables must be dynamic, 132 | Columbia Economic Review

The Efficacy Of Biomedical Research

dependent upon past realizations of itself. Here, we arrive at our second estimation possibility: (2) estimate dynamic panel data via the Arellano-Bond test (Arellano and Bond 1991). As Roodman (2009) explains, Arellano-Bond dynamic panel estimation is designed for situations satisfying the following requirements: (1) few time periods, (2) many panel states, (3) linear model, (4)dynamic dependent variable, (5) independent variables that may or may not be strictly exogenous, (6) fixed effects, (7) heteroskedasticity within but not across panel states, and (8) autocorrelation within but not across panel states. If these conditions are satisfied, regressors are differenced, and estimates are made via Hansen’s (1982) generalized method of moments (GMM). For this reason, the Arellano-Bond test is often referred to more generally as “difference GMM” (Roodman 2009, Holtz-Eakin et al. 1988). To estimate using difference GMM, rather than simply regress via Lag-Dependent OLS, i.e. regress the dependent variable on the first-differenced dependent variable (regress Yt on Yt-1), the Arellano-Bond test instead instruments the first-differenced dependent variable with the second-differenced dependent variable (which can be instrumented by the third-differenced dependent variable, and so on and so forth). To explain, I adapt an example similar to that presented by Roodman (2009). Consider the lag-dependent model: Yit = α Y i,t-1 + β Xit + vit, where vit = δit+ μi Regressing via Lag-Dependent OLS could create “dynamic panel bias” if Yi,t-1 is correlated with the fixed effects, μi, within the error term, vit, thereby affecting the consistency of the OLS estimate (Bai 2013, Roodman 2009, Nickell 1981). To account for dynamic panel bias, the Arellano-Bond test first-transforms the model.

Consider the first-differenced model: ΔYit = α ΔYi,t-1 + β ΔXit + Δε it

Even at a glance, autocorrelation seems possible between ΔYi,t-1 and Δεit = εit ε i,t-1, since both terms exhibit at time t-1. To correct for this possibility, the Arellano-Bond test instruments for ΔYi,t-1 with ΔYi,t-2 (and additional lagged values if desired). Since the advent of the Arellano-Bond test, other dynamic panel estimators have been formulated, namely the Arellano-Bover/Blundell-Bond estimator (Arellano and Bover 1995, Blundell and Bond 1998). This estimator makes an additional assumption that first differences of variables instrumented (in order to avoid serial correlation) are uncorrelated with fixed effects, allowing for further instrumentation and therefore further efficiency (Arellano and Bover 1995, Blundell and Bond 1998, Roodman 2009). Rather than only transform its regressors, the Arellano- Bover/Blundell-Bond estimator instead makes a dual system of regressors, including both the original regressors and the transformed regressors, before estimating via Hansen’s (1982) GMM.

Tenth Anniversary | 133

Zoey Chopra

As a result, the Arellano-Bover/Blundell-Bond estimator is more generally referred to as “system GMM” (Roodman 2009). The Arellano-Bover/Blundell-Bond estimator is preferred to the Arellano-Bond estimator in two instances: (1) if Yit approaches a random walk, and (2) if the panel is unbalanced, because differencing further increases gaps in the panel (Roodman 2009). If neither of these criteria is fulfilled, the Arellano-Bond estimator is preferred, because it yields the least variance in its estimates (Roodman 2009). Furthermore, the Arellano-Bover/Blundel-Bond estimator requires a stationarity assumption where the Arellano-Bond estimator does not (Roodman 2009). Because my Yit neither nears a random walk nor is my panel unbalanced, I estimate via the Arellano-Bond estimator. Extrapolating from above, we estimate the following lag-dependent regression: ln(Ydt) = β ln(publication d, t-k) + η ln(Y d, t-i) + γ ln(incidencedt) + ζ ln(expendituredt) + αd + δt + εdt

My variables are defined as follows: Ydt = mortality/hospitalization/discharge/DALYs rate for specific RD (d) in given year (t) publicationd,t-k = cumulative (d)-relevant research articles published until end of year (t-k) incidencedt = age-adjusted incidence rate for specific RD (d) in given year (t) expendituredt = expenditure for specific RD (d) in given year (t) αd = disease-fixed effect for specific RD (d) δt = time-fixed effect for given year (t) εdt = residual perturbation of regression model, Eq. (r) To confirm robustness of the dynamic panel data estimation, I include panel data estimates via the Arellano-Bond test (difference GMM) in Appendix A, via lag-dependent OLS in Appendix B, and via lag-independent OLS in Appendix C. Limitations Outside of data constraints, three large limitations remain. First, my regression models are only estimated on U.S.-localized dependent variables, even though estimating the innovation factor today requires insight into global affairs. Especially within the realm of biomedical publications, international research constantly influences domestic research, as scientists and healthcare professionals often travel globally to attend conferences and help disperse information. While the majority of filters used to extract publication data are U.S.-specific to help address this limitation, there is an 134 | Columbia Economic Review

The Efficacy Of Biomedical Research

obvious framework-level issue in that my innovation factor estimated is influenced by the international sphere, whereas my outcome variables are all confined to the domestic sphere. Future renditions of this paper will aim to address the broader research environment. Second, my regression framework cannot differentiate different respiratory diseases by severity. While it is possible to weight each regression by mean mortality rate over the years I estimate (1990 to 2014) as a proxy, this method fails to capture the vast differences in mortality rate on a year-by-year basis for certain diseases, and in fact could reduce the estimated impact of innovation for a given disease in a given year. Furthermore, the syntax of the Arellano-Bond test does not allow regression weighting. Future renditions of this paper will explore weighting disease severity by mean research funding granted to each disease over the years I estimate as a proxy—similar to controlling for expenditure within the health production function—as well as investigate new dynamic panel estimation syntax, in the form of “xtabond2,” developed by Roodman (2009). Third, my regression framework assumes function linearity. The linear model of innovation has received immense criticism in recent years, given its relative simplicity (Balconi et al. 2010). While there are benefits to a linear regression model, including ease of interpretation, fewer assumptions made, and policy neutrality, criticism of the model’s simplicity may be warranted (Balconi et al. 2010). Future renditions of this paper will seek to adapt other models of production functions to the realms of health and biomedical research and compare findings with those put forth here. Data For the purposes of this paper, given current and reliable data accessibility, I will estimate from 1990 to 2014 across a limited spread of respiratory diseases: Chronic Obstructive Pulmonary Disease (COPD), Pneumoconiosis, Asthma, Interstitial Lung Disease & Pulmonary Sarcoidosis, Lower Respiratory Infections, Upper Respiratory Infections, Respiratory Tract Neoplasms, Nose Neoplasms, Laryngeal Neoplasms, Lung & Bronchial Neoplasms, and Tracheal & Other Respiratory Neoplasms. Mortality Data The age-adjusted, unconditional mortality data for the entire range of respiratory diseases considered in this paper come from the Center for Disease Control and Prevention (CDC)’s WONDER Database’s Compressed Mortality Files (CMFs). Spanning 1968 to 2015, CMFs are updated annually to reflect data sourced from the National Center for Health Statistics (NCHS). Data analysis uses age-adjusted death rates originally extracted from CMFs. The NCHS calculates age adjustment as follows (Anderson and Tenth Anniversary | 135

Zoey Chopra

Rosenberg 1998): R R = age-adjusted death rates Psi = standard population for age group, i Ps = total U.S. standard population (Psi/Ps) = standard population weight Ri = age-specific death rates Age-adjusted mortality data for RD-associated neoplasms (cancers) from 1969 to 2014 is obtained from the National Cancer Institute’s (NCI) Surveillance, Epidemiology, and End Results (SEER) Program. Mortality rates from both the CDC and the NCI are calculated as the number of deaths per 100,000 individuals within a given population. Publication Data The time-series data for publication count per year for the span of respiratory diseases considered in this paper come from the National Center for Biotechnology Information (NCBI)’s PubMed database (PubMed Help 2005). More broadly, the database is managed by the U.S. National Library of Medicine (NLM), a branch of the National Institutes of Health (NIH). The PubMed database consists of more than 26 million citations (and abstracts) for articles published in biomedical journals. The majority of these references (over 24 million) are sourced from the MEDLINE database, the online counterpart to the MEDLARS (Medical Literature Analysis and Retrieval System) that began in 1964 under the National Library of Medicine (NLM 2017). MEDLINE is referred to as the NLM’s “premier bibliographic database [for] journal articles in life sciences with a concentration on biomedicine (“Fact Sheet: MEDLINE®”). In terms of coverage, citations are extracted from over 5,600 journals worldwide, 93% of which are published in English, and, most importantly, indexed with standardized NLM Medical Subject Headings (MeSH terms). MeSH descriptors are both alphabetical and hierarchical, permitting condition-relevant searches at varying levels of specificity (MeSH Tree Structure publicly available online). Currently, there are over 28,000 unique MeSH terms with more than 90,000 corresponding entry terms to ensure that varying user-inputted search terms for similar conditions/compounds—like “Vitamin C” and “Ascorbic Acid”—yield the same correct MeSH terms (“Fact Sheet: Medical Subject Headings®”). MeSH terms also are not statically defined, but rather updated annually in accordance with changes in biomedical terminology (“Fact Sheet: MEDLINE®”). Though PubMed citations are derived from 136 | Columbia Economic Review

The Efficacy Of Biomedical Research

MEDLINE, PubMed Central, and NCBI Bookshelf databases, searching using MeSH terms limits the results to MEDLINE-sourced citations ((“Fact Sheet: MEDLINE®”). To acquire publication data, I use the NCBI’s Entrez Programming Utilities (E-utilities), public APIs that allow programmable data extraction from the PubMed database. I narrow citation searches using two pre-existing PubMed attribute types (search fields): [MeSH Major Topic] and [Publication Type]. Using Python script, I extract the number of publications per year, restricted by these two attribute fields. Although previous bibliometric analyses on cancer data by Lichtenberg (2013b, 2001) have indicated insignificant impact from non-funded research publications without financial support, correlational data within the realm of Respiratory Diseases suggests otherwise, evidenced by Table 1 below. Therefore, I estimate publication data as the cumulative of citations that have indicated some form of research support, public and private, and those that have indicated no form of research support, public and private.

While the PubMed script I wrote enables my collection of publication count per year for designated search terms, the NCBI’s public APIs do not permit extraction of funding timelines specific to each research article published in a given year. As a broad estimate, we can look to the NIH Research Portfolio Online Reporting Tools (RePORT) to assess the distribution in lag-time between start date and publication date for NIH-supported research projects. Using similar PubMed-derived data, Lichtenberg (2013b) finds that lag-time distribution from start to publication peaks unimodally at ~6 years, a lower-bound estimate for the length of time necessary for biomedical research to manifest tangibly, a similar lower bound was confirmed by Stern and Simes (1997) for publications stemming from clinical trials. Existing literature estimates an upper-bound estimate of ~17 years for biomedical research to induce real-world changes (Morris et al. 2011, Grant et al. 2003). Thus, we may assume that determining the effects of respiratory disease-relevant biomedical research on respiratory-disease-relevant outcomes will be most applicable in the range of 6 to 17 years after publication. Tenth Anniversary | 137

Zoey Chopra

Incidence Data Age-adjusted incidence data for Chronic Obstructive Pulmonary Disease (COPD), Pneumoconiosis, Asthma, Interstitial Lung Disease and Pulmonary Sarcoidosis, Lower Respiratory Infections, and Upper Respiratory Infections come from the Institute for Health Metrics and Evaluation (IHME)’s Global Health Data Exchange (GHDx)’s Global Burden of Disease Study 2016 (Global Burden of Disease Collaborative Network 2017). Incidence rates are calculated as the number of cases per 100,000 individuals within a given population. The GHDx extracts the majority of its data from the Centers for Disease Control and Prevention’s (CDC) National Center for Health Statistics’ (NCHS) annual United States Vital Statistics Reports and National Vital Statistics System (NVSS). Age-adjusted incidence data for RD-associated neoplasms (cancers), including Laryngeal Neoplasms, Lung & Bronchial Neoplasms, and Tracheal & Other Respiratory Neoplasms from 1973 to 2014 is obtained from the National Cancer Institute’s Surveillance, Epidemiology, and End Results (SEER) Program (NCI 2016). While age-adjusted incidence data is not readily available for a larger swath of respiratory diseases, public health literature has provided ways of estimating incidence data by pooling National Ambulatory Medical Care Survey (NAMCS)/National Hospital Ambulatory Medical Care Survey (NHAMCS) outpatient data and estimating yearly averages of the number of visits treated in doctors’ offices per specified condition (Finkelstein et al. 2006). These estimates can be cross- referenced with data from the Medical Expenditure Panel Survey (MEPS), which maintains unique records for every medical condition reported by survey participants per year (Finkelstein et al. 2006). Fortunately, these three sources all classify medical conditions with the same standardized International Classification of Diseases (ICD) codes. The one caveat is that MEPS does not tabulate the year in which a survey participant first contracts a specific medical condition and thus provides a more accurate sample estimate of yearly prevalence rates rather than yearly incidence rates. Incidence data can also be estimated for a range of specific respiratory diseases if administrative data is available, as systemic lupus erythematosus (SLE) studies have shown (Ward 2013). While future versions of this paper will expand to estimations of incidence in order to access data across a broader range of respiratory diseases, this current version is limited to disease states where incidence data exists, is publicly available, and is reliable. Hospitalization Data Time-series data for lengths of hospitalization for Chronic Obstructive Pulmonary Disease (COPD), Asthma, Lower Respiratory Infections, Upper 138 | Columbia Economic Review

The Efficacy Of Biomedical Research

Respiratory Infections, Influenza, Pneumonia, Lung & Bronchial Neoplasms, and Tracheal & Other Respiratory Neoplasms come from the U.S. Department of Health & Human Services’ Agency for Healthcare Research and Quality’s Healthcare Cost and Utilization Project (HCUPnet). Spanning 1997 to 2014, HCUPnet compiles annual in-hospital data, collected statewide from community hospitals, and remains the largest collection of longitudinal care-related data within the U.S. Beginning in 2012, HCUPnet reformulated data collected from the National Inpatient Sample (NIS)—notable for hospitalization and discharge data especially—in order to optimize national estimates from state-level data compilations. Estimates provided prior to 2012 were recalculated with newer trend weights to permit reliable longitudinal analysis. Slight differences remain between estimates provided pre-2012 and post-2012. While hospitalization data is currently limited to only a few respiratory disease categories, future renditions of this paper will expand to include data from the Center for Disease Control and Prevention (CDC)’s National Center for Health Statistics (NCHS)’s National Hospital Discharge Survey, with reliable estimates of hospitalization rates on a disease basis (coded by ICD-9CM codes) from 2000 to 2010 (Hall et al. 2013). Discharge Data Time-series data for discharge rates for Chronic Obstructive Pulmonary Disease (COPD), Asthma, Lower Respiratory Infections, Upper Respiratory Infections, Influenza, Pneumonia, Lung & Bronchial Neoplasms, and Tracheal & Other Respiratory Neoplasms come from the U.S. Department of Health & Human Services’ Agency for Healthcare Research and Quality’s Healthcare Cost and Utilization Project (HCUPnet). Spanning 1997 to 2014, HCUP compiles annual in-hospital data, collected statewide from community hospitals, and remains the largest collection of longitudinal care-related data within the U.S. Beginning in 2012, HCUPnet reformulated data collected from the National Inpatient Sample (NIS)—notable for hospitalization and discharge data especially—in order to optimize national estimates from state-level data compilations. Estimates provided prior to 2012 were recalculated with newer trend weights to permit reliable, longitudinal analysis. Slight differences remain between estimates provided pre-2012 and post-2012. While discharge data is currently limited to only few respiratory disease categories, future renditions of this paper will expand to include data from the Center for Disease Control and Prevention (CDC)’s National Center for Health Statistics (NCHS)’s National Hospital Discharge Survey, with reliable estimates of discharge rates on a disease basis (coded by ICD-9-CM codes) from 2000 to 2010 (Hall et al. 2013).

Tenth Anniversary | 139

Zoey Chopra

DALYs Data Data for DALYs rates for Chronic Obstructive Pulmonary Disease (COPD), Pneumoconiosis, Asthma, Interstitial Lung Disease and Pulmonary Sarcoidosis, Lower Respiratory Infections, Upper Respiratory Infections, and Laryngeal Neoplasms comes from the Institute for Health Metrics and Evaluation (IHME)’s Global Health Data Exchange (GHDx)’s Global Burden of Disease Study 2016 (Global Burden of Disease Collaborative Network 2017). DALYs rates are calculated as the number of cases per 100,000 individuals within a given population. The GHDx extracts the majority of its data from the Centers for Disease Control and Prevention’s (CDC) National Center for Health Statistics’ (NCHS) annual United States Vital Statistics Reports and National Vital Statistics System (NVSS). As previously explained, the disease-specific disability factors used to calculate DALYs are calculated by summing the years of life lost (YLL) from premature mortality and years lived but constrained by disease-yielding disability (Global Burden of Disease Collaborative Network 2017). The years lived with disability (YLD) metric are determined with a disease-specific disability factor, bounded between 0 and 1, assessed by a range of health experts across nations and cultures (Parks 2014, Gold et al. 2012, Global Burden of Disease Collaborative Network 2017). DALYs are also weighted by age, similarly to age-adjustment in incidence and mortality data but based upon expert-determined societal value by age rather than upon age itself (Parks 2014, Gold et al. 2012). Expenditure Data Time-series data for hospital expenditure for Chronic Obstructive Pulmonary Disease (COPD), Asthma, Lower Respiratory Infections, Upper Respiratory Infections, Influenza, Pneumonia, Lung & Bronchial Neoplasms, and Tracheal & Other Respiratory Neoplasms derive from the U.S. Department of Health & Human Services’ Agency for Healthcare Research and Quality’s Healthcare Cost and Utilization Project (HCUPnet). Spanning 1997 to 2014, HCUP compiles annual in-hospital data, collected statewide from community hospitals, and remains the largest collection of longitudinal care-related data within the U.S. To gauge expenditure, I analyze two measures: (1) Median Charge and (2) Median Cost. Median Charge reflects the median amount billed by healthcare providers (community hospitals in the case of HCUPnet data) to patients during their length of stay. Median Cost is so-called to represent the median cost of production of healthcare providers (community hospitals in the case of HCUPnet data) for patient care and treatment. HCUPnet gathers charge data from annual in-hospital data and converts to cost data via cost-to-charge ratios 140 | Columbia Economic Review

The Efficacy Of Biomedical Research

on a per-hospital level, calculated from accounting reports of the Centers for Medicare and Medicaid Services (CMS). Though Median Cost perhaps better reflects the expenditure input of an estimated health production function, cost data is less preferred to charge data for three reasons. Firstly, Median Cost data is only available from 2001 to 2014, while Median Charge data is available from 1997 to 2014. Secondly, Median Cost data is estimated on a hospital-wide level, which could prove unreliable for certain conditions, whereas Median Charge data is collected directly. Thirdly, Median Cost data is converted from Median Charge data by CMS reports, which contain accounting data only for certain population subsets that qualify for Medicare/Medicaid, limiting the reliability of Median Cost data.

The high correlations between measures of Median Charge and Median Cost, as observed above in Table 2, serve as further confirmation that controlling for expenditure via Median Charge alone is unlikely to skew or bias regression estimates any differently than would controlling for expenditure via Median Cost. The high correlations present in Table 2 also rule out the possibility of controlling for expenditure with both Median Charge and Median Cost, due to endogeneity between the two variables. This finding is definitive since, as discussed above, Median Cost data is derived from Median Charge data. Therefore, since the data is more directly sourced, more reliable, and more longitudinal, Median Charge rather than Median Cost is used to control for expenditure. Data Mapping CDC-derived mortality data is classified via International Classification of Diseases (ICD) codes, each of which provides a unique alphabetical and numerical combination to reference a specific medical condition. Data pre-1999 is classified by ICD-9 codes, whilst data 1999 and onwards is classified by ICD-10 codes. Fortunately, reliable mapping algorithms exist between ICD-9 and ICD-10 codes to ensure consistency in mortality rates before and after 1999. NCI-derived mortality data is classified by cancer site, which can easily be mapped at high levels to ICD-9 and ICD-10 codes. Otherwise, mortality data from both the CDC and NCI is initially extracted as age-adjusted, and per 100,000 population. Tenth Anniversary | 141

Zoey Chopra

PubMed-derived publication data is classified via National Library of Medicine Medical Subject Headings (MeSH terms). Though precise one-to-one mappings of MeSH terms to ICD codes are hard to determine for the most specific conditions, broader disease states are analyzed as they align accordingly. The MeSH term of “Laryngeal Neoplasms,” for instance, corresponds to the ICD descriptor of “Malignant Neoplasm of Larynx,” which corresponds to the NCI’s SEER cancer site descriptor of “Larynx.” IHME-derived incidence data is classified specific to the aforementioned categories. Fortunately, these categories directly correspond with both ICD codes and MeSH terms, given their relative non-specificity. NCI-derived incidence data is classified by the cancer site, which can easily be mapped at high levels to ICD-9 and ICD-10 codes. Incidence data from both the IHME and NCI is initially extracted as age-adjusted, and per 100,000 population. IHME-derived DALYs data is similarly classified to incidence data and maps readily to both ICD codes and MeSH terms. DALYs data is initially extracted as per 100,000 population. HCUPnet-derived hospitalization, discharge, and expenditure data is classified by Diagnoses— Clinical Classification Software (CCS), which groups and incorporates ICD-9 codes similarly to the larger disease categories specified by PubMed’s MeSH terms and the IHME. Fortunately, mapping each disparate regression variable across data sets is reliable. Empirical Estimates Given that optimal research efficacy operates within the bounds of 6 to 17 years, I estimate: My variables are defined as follows: Ydt = mortality/hospitalization/discharge/DALYs rate for specific RD (d) in given year (t) publicationd,t-k = cumulative (d)-relevant research articles published until end of year (t-k) Yd,t-i = mortality/hospitalization/discharge/DALYs rate for specific RD (d) in year (t-1) incidencedt = age-adjusted incidence rate for specific RD (d) in given year (t) expendituredt = expenditure for specific RD (d) in given year (t) αd = disease-fixed effect for specific RD (d) δt = time-fixed effect for given year (t) εdt = residual perturbation of regression model, Eq. (r) 142 | Columbia Economic Review

The Efficacy Of Biomedical Research

for the following time lags in publication: (1) k = 0 (2) k = 6 (3) k = 11 (4) k = 17 Rather than assume homoskedasticity, I account for heteroskedasticity-exhibiting data, and so estimate via robust standard errors. Unlike standard lagged panel-series regressions, a separate regression model is required for varying lag values (k) because publication-associated variables, lagged or not, are highly serially correlated with one another, as indicated by Table 6. Assume a certain research article effects difference in mortality one year from publication, it is then likely that the same research article will create differences in mortality two years from publication as well. Because regressing lagged publication-associated variables together would yield high multicollinearity, I refrain, and test the following models at the four above lagged values (k=0, k=6, k= 11, k= 17) independently:

For the purposes of this paper, I will estimate across a limited spread of respiratory diseases given current data accessibility: COPD, Pneumoconiosis, Asthma, Interstitial Lung Disease & Pulmonary Sarcoidosis, Lower Respiratory Infections, Upper Respiratory Infections, Respiratory Tract Neoplasms, Nose Neoplasms, Laryngeal Neoplasms, Lung & Bronchus Neoplasms, and Tracheal Neoplasms. Y = Mortality Rate Table A1 outlines Arellano-Bond estimates of Models 1.0, 2.0, and 3.0 from 1990 to 2014. Table A2 outlines Arellano-Bond estimates of Models 1.6, 2.6, and 3.6 from 1990 to 2014. Table A3 outlines Arellano-Bond estimates of Models 1.11, 2.11, and 3.11 from 1990 to 2014. Table A4 outlines Arellano-Bond estimates of Models 1.17, 2.17, and 3.17 from 1990 to 2014. These results will be displayed in Appendix A. Table B1 outlines Least Squares estimates of Models 1.0, 2.0, and 3.0 from 1990 to 2014. Table B2 outlines Least Squares estimates of Models 1.6, 2.6, and 3.6 from 1990 to 2014. Table B3 outlines Least Squares estimates of Models 1.11, 2.11, and 3.11 from 1990 to 2014. Table B4 outlines Least Squares estimates of Models 1.17, 2.17, and 3.17 from 1990 to 2014. These results will be displayed in Appendix B.

Tenth Anniversary | 143

Zoey Chopra

For comparisonâ&#x20AC;&#x2122;s sake, I include Least Squares estimates without lag-dependence: Table Cl outlines Least Squares estimates of Models 1.0, 2.0, and 3.0 from 1990 to 2014. Table C2 outlines Least Squares estimates of Models 1.6, 2.6, and 3.6 from 1990 to 2014. Table C3 outlines Least Squares estimates of Models 1.11, 2.11, and 3.11 from 1990 to 2014. Table C4 outlines Least Squares estimates of Models 1.17, 2.17, and 3.17 from 1990 to 2014. These results will be displayed in Appendix C. Y = Hospitalization Rate Table A5 outlines Arellano-Bond estimates of Models 1.0, 2.0, and 3.0 from 1997 to 2014. Table A6 outlines Arellano-Bond estimates of Models 1.6, 2.6, and 3.6 from 1997 to 2014. Table A7 outlines Arellano-Bond estimates of Models 1.11, 2.11, and 3.11 from 1997 to 2014. Table A8 outlines Arellano-Bond estimates of Models 1.17, 2.17, and 3.17 from 1997 to 2014. These results will be displayed in Appendix A. Table B5 outlines Least Squares estimates of Models 1.0, 2.0, and 3.0 from 1997 to 2014. Table B6 outlines Least Squares estimates of Models 1.6, 2.6, and 3.6 from 1997 to 2014. Table B7 outlines Least Squares estimates of Models 1.11, 2.11, and 3.11 from 1997 to 2014. Table B8 outlines Least Squares estimates of Models 1.17, 2.17, and 3.17 from 1997 to 2014. These results will be displayed in Appendix B. For comparisonâ&#x20AC;&#x2122;s sake, I include Least Squares estimates without lag-dependence: Table C5 outlines Least Squares estimates of Models 1.0, 2.0, and 3.0 from 1990 to 2014. Table C6 outlines Least Squares estimates of Models 1.6, 2.6, and 3.6 from 1990 to 2014. Table C7 outlines Least Squares estimates of Models 1.11, 2.11, and 3.11 from 1990 to 2014. Table C8 outlines Least Squares estimates of Models 1.17, 2.17, and 3.17 from 1990 to 2014. These results will be displayed in Appendix C. Y = Discharge Rate Table A9 outlines Arellano-Bond estimates of Models 1.0, 2.0, and 3.0 from 1997 to 2014. Table A10 outlines Arellano-Bond estimates of Models 1.6, 2.6, and 3.6 from 1997 to 2014. Table A11 outlines Arellano-Bond estimates of Models 1.11, 2.11, and 3.11 from 1997 to 2014. Table A12 outlines Arellano-Bond estimates of Models 1.17, 2.17, and 3.17 from 1997 to 2014. These results will be displayed in Appendix A. Table B9 outlines Least Squares estimates of Models 1.0, 2.0, and 3.0 from 1997 to 2014. Table B10 outlines Least Squares estimates of Models 1.6, 2.6, and 3.6 from 1997 to 2014. 144 | Columbia Economic Review

The Efficacy Of Biomedical Research

Table B11 outlines Least Squares estimates of Models 1.11, 2.11, and 3.11 from 1997 to 2014. Table B12 outlines Least Squares estimates of Models 1.17, 2.17, and 3.17 from 1997 to 2014. These results will be displayed in Appendix B. For comparisonâ&#x20AC;&#x2122;s sake, I include Least-Squares estimates without lag-dependence:

Table C9 outlinesLeast Squares estimatesof Models 1.0, 2.0, and 3.0 from 1997 to 2014.

Table C10 outlines Least Squares estimates of Models 1.6, 2.6, and 3.6 from 1997 to 2014.

Table C11 outlines Least Squares estimates of Models 1.11, 2.11, and 3.11 from 1997 to 2014. Table C12 outlines Least Squares estimates of Models 1.17, 2.17, and 3.17 from 1997 to 2014. These results will be displayed in Appendix C.

Y = DALYs Rate Table A13 outlines Arellano-Bond estimates of Models 1.0, 2.0, and 3.0 from 1997 to 2014. Table A14 outlines Arellano-Bond estimates of Models 1.6, 2.6, and 3.6 from 1997 to 2014. Table A15 outlines Arellano-Bond estimates of Models 1.11, 2.11, and 3.11 from 1997 to 2014. Table A16 outlines Arellano-Bond estimates of Models 1.17, 2.17, and 3.17 from 1997 to 2014. These results will be displayed in Appendix A. Table B13 outlines Least Squares estimates of Models 1.0, 2.0, and 3.0 from 1997 to 2014. Table B14 outlines Least Squares estimates of Models 1.6, 2.6, and 3.6 from 1997 to 2014. Table B15 outlines Least Squares estimates of Models 1.11, 2.11, and 3.11 from 1997 to 2014. Table B16 outlines Least Squares estimates of Models 1.17, 2.17, and 3.17 from 1997 to 2014. These results will be displayed in Appendix B. For comparisonâ&#x20AC;&#x2122;s sake, I include Least Squares estimates without lag-dependence: Table C13 outlines Least Squares estimates of Models 1.0, 2.0, and 3.0 from 1997 to 2014. Table C14 outlines Least Squares estimates of Models 1.6, 2.6, and 3.6 from 1997 to 2014. Table C15 outlines Least Squares estimates of Models 1.11, 2.11, and 3.11 from 1997 to 2014. Table C16 outlines Least Squares estimates of Models 1.17, 2.17, and 3.17 from 1997 to 2014. These results will be displayed in Appendix C. Discussion Comparing within tables, it is notable that for almost every lag in publication tested, whether the estimation occurs via Arellano-Bond, Lag-Dependent OLS, or Lag-Independent OLS, the coefficients estimating incidence (y) and median charge (Q are statistically significant. Equally consequential is that, Tenth Anniversary | 145

Zoey Chopra

across estimation strategies, controlling for incidence appears necessary to generate notable and consistent statistical significance in either the (3 coefficient for supported research or the P coefficient for unsupported research. This implies that controlling for incidence is necessary and controlling for expenditure is important. Otherwise, as indicated by Model 1 estimates in Tables Al, A5, A13, B6, and Cl5, estimates for the coefficient of supported publication growth (P) may be biased towards zero, or even in the positive direction. Thus, my discussion will primarily focus upon estimates derived from Model 3, which controlled for both annual incidence and annual expenditure (in the form of median charge). Similarly, since the majority of estimates produced with lag k = 17 years did not result in statistical significance, discussion will primarily focus upon estimates with lag k = 0, 6, and 11 years. I originally hypothesized that publication growth is inversely related to mortality rate for respiratory diseases, because innovation allows treatments, techniques, and technologies to improve, keeping patients to live longer. Lichtenberg (2013b) arrived at a similar conclusion with cancer-associated data. I similarly hypothesized that publication growth is inversely related to hospitalization rate and DALYs rate for respiratory diseases, since patients would spend less time in the hospital when there is technological breakthrough. I hypothesized that discharge rate, however, is in a direct relationship with publication growth, because treatment improves with more innovation, rendering less need for hospitalization. The data, on the other hand, presents interesting and rather counterintuitive results.

146 | Columbia Economic Review

The Efficacy Of Biomedical Research

Tenth Anniversary | 147

Zoey Chopra

For supported publications, the takeaways can be derived from the above: significant differences are not wrought amongst various estimation strategies. While Lag-Independent OLS generally appears less significant as an estimate, especially given the breadth of confidence intervals in Figures 6 and 9, nonetheless the three estimation strategies generally correspond to one another, at least in terms of directional trend. This provides directional confidence with regards to the estimates produced. Across dependent variables and time lags, we observe the following with statistical significance: (1) (2) (3) (4) (5)

positive relation between supported publications and mortality rate at k=6 & k=l 1 (Fig. 6) positive relation between supported publications and hosp. rate at k=0 & k=6 (Fig. 7) inverse relation between supported publications and discharge rate at k=6 & k=l 1 (Fig. 8) inverse relation between supported publications and DALYs rate at k=0 (Fig. 9) positive relation between supported publications and DALYs rate at k=6 (Fig. 9)

Each observed result points contrary to my initial hypotheses, save Observation (4). It is difficult to make much of Observation (4), however, given its model has lag k=0, meaning the effects observed are unlikely a product of biomedical innovation. 148 | Columbia Economic Review

The Efficacy Of Biomedical Research

Tenth Anniversary | 149

Zoey Chopra

150 | Columbia Economic Review

The Efficacy Of Biomedical Research

For unsupported publications, a similar takeaway can be gathered: the three estimation strategies appear to mimic each other, at least in terms of directional trend. While Lag-Independent OLS similarly appears less significant as an estimate for unsupported publications, especially given the breadth of confidence intervals in Figures 10 and 13, nonetheless there is marked similitude across dependent variables and estimation strategies. This provides directional confidence to the estimates produced. Across dependent variables and time lags, we observe the following with statistical significance: (1) (2) (3) (4) (5)

inverse relation between unsupported publications and mortality rate at k=0 & k=6 (Fig. 10) inverse relation between unsupported publications and hosp. rate at k=0 & k=6 (Fig. 11) positive relation between unsupported publications and discharge rate at k=6 & k=l 1 (Fig. 12) positive relation between unsupported publications and DALYs rate at k=0 (Fig. 13) inverse relation between unsupported publications and DALYs rate at k=6 (Figure 13)

Each observed result supports my initial hypotheses, save Observation (4). It is difficult to make much of Observation (4), however, given its model has lag k=0, meaning the effects observed are unlikely a product of biomedical innovation. It is possible that these effects observed are more likely relevant for unsupported publications rather than supported publications because unsupported publications may be, for instance, a summary product of years of research from supported publications. However, given the other trends in Observations (1), (2), (3), and (5), it is also possible that the interpretation I assumed for the effect of biomedical innovation on DALYs could not, in fact, be assumed. Comparing between the effects of supported and unsupported publications on the outcome variables above, it is worth noting that in every observation of statistical significance, estimates for supported and unsupported publications are directionally opposed. This implies that research without funding, public or private, yields more beneficial health outcomes than does research with funding. While this conclusion appears problematic, there is perhaps a warranted explanation. We must take into account the nature of, and diffusion of unfunded research. It is important to note that unfunded research most likely results in entirely disembodied innovation, given the high capital expenditure necessary for production, approval, and advertisement of embodied innovation in the health space. As such, when we envision unsupported research, a mainstream article may come to mind, say a publication from TIME magazine, or a more

Tenth Anniversary | 151

Zoey Chopra

scientifically rigorous publication—secondary literature. This type of unsupported research inevitably has greater reach and larger audience, relative to supported publications. An argument could be made that the process of innovation diffusion appears insensible, since highly skilled biomedical researchers would be unlikely to read and/or learn from such articles, however, there is still high likelihood that medical practitioners and healthcare professionals would uptake and perhaps even regularly follow such kinds of unsupported research to remain up-to-date in their respective fields, resulting in the transition from innovation without financial support to improved health outcomes. It is important to keep in mind that tangible differences in health outcome, especially with regards to disembodied innovation, stem from healthcare workers rather than from biochemical researchers. With that said, an even larger puzzle comes to light upon considering short-term versus long-term effects in lag-dependent regression models. In our above regression model, because of the specification of lag-dependent endogeneity of our dependent variable, there exists a difference between short-term and long-term effects of publications on dependent variable, Yt (Abbassi and Linzert 2011). As expected, the short-term effect of publications on Yt is captured by the coefficient, β. Because Yt-1 is included as a regressor, however, publications do not only have an effect on Yt , but also on Yt+1 , through the lagged variable. The size of this effect is β * η. Similarly, publications will have an effect on Yt+2, equivalent to β * η2. Ad infinitum, we can approximate and surmise that the long-run effect of publications on Yt is captured by the coefficient β/(1- η). Outside of the exceptions observed in Tables A13 and B13, η < 1, meaning that the effects observed from supported publications only further exacerbate without shifts in directionality (since denominator remains positive). Reconciling these results, it appears that respiratory disease-relevant, funding-supported biomedical research growth contributes to increased mortality, hospitalization, and DALYs rates and decreased discharge rates, 6 to 11 years after publication. Counterintuitively, respiratory disease-relevant, funding-absent biomedical research contributes to decreased mortality, hospitalization, and DALYs rates and increased discharge rates, 6 to 11 years after publication. Conclusion In this paper, I utilized a bibliometric analysis to examine growth in biomedical research—the relationship between the growth of supported and unsupported, disease-specific research accumulation and changes in diseasespecific rates of mortality, hospitalization, discharge and DALYs. In particular, I 152 | Columbia Economic Review

The Efficacy Of Biomedical Research

examine the effects of research accumulation across a swath of respiratory diseases (RD). RD-associated deaths have accounted for nearly 8% of all deaths in the U.S from 1999 to 2013 (Figures 1 and 2, Tejada-Vera et al. 2015). Understanding the drivers behind changes in RD mortality, hospitalization, discharge, and DALYs rates holds vast significance for the broader US population and for future RD-associated biomedical research and development. My time-fixed and disease-fixed estimates, confirmed via three different estimation strategies, indicate that growth in unsupported publications in a given year are weakly inversely related to mortality rate, hospitalization rate, and DALYs rate, and weakly positively related to discharge rate. This relationship appears to strengthen as the lagged effects of cumulative publication count are considered, especially in the time frame 6 to 11 years after publication. Given the statistical estimates for the coefficient of the lagged dependent variable (p), it appears that this relationship continues to strengthen in the long-run as well. However, growth of supported publications in a given year appear to have the opposite effect, in every way. As discussed, however, these results may be a product of reverse causality. Given limited incidence and expenditure data for a portion of respiratory diseases tested, my small sample size might affect the regression estimates untowardly. Future research will develop this paper in four veins: outcome variables, publication types, research proxies, and data expansions. While I only considered the effects of publication growth on rates of mortality, hospitalization, discharge, and DALYs, I will expand my analysis to include outcome variables like years of potential life lost (YPLL), changes to health adjusted life years (HALY), and changes to quality adjusted life years (QALY), which are sometimes preferred in health economics literature because they capture different aspects of health (Romeder and McWhinnie 1977, Yoshida et al. 1997). Since a large percentage of respiratory diseases are not fatal, these alternate outcome variables may offer more nuanced insight into the tangible effects of biomedical research. I will also test whether or not publication types affect outcome variables differently. Though my current analysis considers publication data to articles with versus without research support (i.e. funding), this data set is limited. Fortunately, since 1975, the PubMed database has distinguished between articles that did and did not receive funding support at the level of funding source (i.e. government-funded vs. privately-funded). I am currently writing a script using the NCBIâ&#x20AC;&#x2122;s public APIs to extract time-series publication data by category of funding source. Though research in other health domains has found this separation fruitless, the endeavor remains yet to be seen within the realm of respiratory diseases (Cockburn and Henderson 1998, Lichtenberg 2016).

Tenth Anniversary | 153

Zoey Chopra

Another possibility involves determining publication data at a different level of MeSH designation. While in this paper I filtered by [MeSH Major Topic], it is possible that other research publications have tangentially affected the research space for the respiratory disease under examination even though their scope may be broader (i.e. respiratory disease is not MeSH Major Topic. This possibility has not been widely explored, outside of my consideration of certain respiratory-related neoplasms, including lung, bronchial, tracheal, and respiratory tract cancers. Given remaining concerns for endogeneity and serial correlation when considering lagged publication growth, I will explore other quantifiable proxies for biomedical research as well, like grant counts (Lichtenberg 2001), which may assess grant supply, or grant submission counts, which may assess grant demand. In large part, my paper has been limited by data accessibility. Unlike conditions such as cancer which have national, data-accumulating registries, the majority of respiratory diseases lack sources for reliable incidence data. As mentioned before, controlling for incidence rate is necessary for determining reliable estimates of publication growthâ&#x20AC;&#x2122;s effects on mortality rate. To account for the lack of data, I will explore various methods of estimating incidence rates, for which pre-existing research exists (Fowlkes et al. 2013, Finkelstein et al. 2006, Corso et al. 2006). Given that data is available, including rates of prevalence, hospitalization, outpatient visits, and ambulatory trips, reliably estimating incidence seems plausible, albeit with large standard errors. Though extensive further analysis is required, my findings presented in this paper finally conclude that estimating publication count as a proxy for the innovation factor of health production functions is plausible, even with stated limitations. So far, my research confirms that support (though not necessarily monetary) towards biomedical research and development will tend to positively affect health and longevity in the United States, conversely, reduction in support may have adverse effects, even in the short-run. Positive effects may not be realized until the long-run, however, because in the vein of respiration, research must breathe. REFERENCES Abbassi, Puriya, and Tobias Linzert. 2011. â&#x20AC;&#x153;The Effectiveness of Monetary Policy in Steering Money Market Rates during the Recent Financial Crisis.â&#x20AC;? European Central Bank: Eurosystem, Working Paper Series, no. 1328 (April): 1-33, Footnote 23.

154 | Columbia Economic Review

The Efficacy Of Biomedical Research

Achen, Christopher H. 2000. “Why Lagged Dependent Variables Can Suppress the Explanatory Power of Other Independent Variables.” Political Methodology Section, American Political Science Association, UCLA Annual Meeting (July): 1-42. Anderson, Gerard, Earl Steinberg, and Robert Heyssel. 1994. “The Pivotal Role of the Academic Health Center.” Health Affairs 13 (3): 146-58. https:// d0i.0rg/l0.1377/hlthaff. 13.3.146. Anderson, Robert N., and Rosenberg, Harry M. 1998. “Age Standardization of Death Rates: Implementation of the Year 2000 Standard.” National Center for Health Statistics: National Vital Statistics Report 47 (3): 1-20. https://doi.org/10.4135/9781412952484.n432. Arellano, Manuel, and Stephen Bond. 1991. “Some Tests of Specification for Panel Data: Monte Carlo Evidence and an Application to Employment Equations.” The Review of Economic Studies 58 (2): 277. https://doi.org/10.2307/2297968. Arellano, Manuel, and Olympia Bover. 1995. “Another Look at the Instrumental Variable Estimation of Error-Components Models.” Journal of Econometrics 68 (1): 29-51. https://d0i.0rg/l 0.1016/0304 4076(94101642-D. Azoulay, Pierre, Joshua S. Graff Zivin, and Gustavo Manso. 2011. “Incentives and Creativity: Evidence from the Academic Life Sciences.” The RAND Journal of Economics 42 (3): 527- 54. https://doi.0rg/lO.l 11 l/i.1756-2171.2011.00140.x. Bai, Jushan. 2013. “Fixed-Effects Dynamic Panel Models, a Factor Analytical Method.” Econometrica 81 (1): 285-314. https://doi.org/10.3982/ ECTA9409. Balconi, Margherita, Stefano Brusoni, and Luigi Orsenigo. 2010. “In Defence of the Linear Model: An Essay.” Research Policy 39 (1): 1-13. https:// doi.org/10.1016/j .respol.2009.09.013. “Bibliometrics Basics.” 2018. University College London Library Services. 2018. http://www.ucl.ac.uk/librarv/research-support/bibliometrics/basics. Blundell, Richard, and Stephen Bond. 1998. “Initial Conditions and Moment Restrictions in Dynamic Panel Data Models.” Journal of Econometrics 87 (1): 115-43. https://d0i.0rg/l 0.1016/S0304-4076(98t00009-8. Chen, Lena M., Ashish K. Jha, Stuart Guterman, Abigail B. Ridgway, E. John Orav, and Arnold M. Epstein. 2010. “Hospital Cost of Care, Quality of Care, and Readmission Rates: Penny Wise and Pound Foolish?” JAMA Internal Medicine 170 (4): 340-46. https://doi.org/10.1001/ archinternmed.20Q9.511.

Tenth Anniversary | 155

Zoey Chopra

“Chronic Obstructive Pulmonary Disease (COPD).” 2017. Centers for Disease Control and Prevention (CDC). 2017. https://www.cdc.gov/copd/index. html. Cockburn, Iain M., and Henderson, Rebecca M. 1998. “Absorptive Capacity, Coauthoring Behavior, and the Organization of Research in Drug Discovery.” The Journal of Industrial Economics 46 (2): 157-82. https://d0i.0rg/l0.1111/1467-6451.00067. Coleman, James, Elihu Katz, and Herbert Menzel. 1957. “The Diffusion of an Innovation Among Physicians.” Sociometry 20 (4): 253-70. https://doi.org/10.2307/2785979. Comin, Diego. 2008. “Total Factor Productivity.” Edited by Steven Derlauf and Larry Blume. The New Palgrave Dictionary of Economics. Hampshire, U.K.: Palgrave Macmillan. https://www.hbs.edu/facultv/Pages/item. aspx?num=30762. Comin, Diego, and Sunil Mulani. 2009. “A Theory of Growth and Volatility at the Aggregate and Firm Level.” Journal of Monetary Economics 56 (8): 1023-42. https://doi.Org/10.1016/i.imoneco.2009.10.004. Centers for Disease Control and Prevention, National Center for Health Statistics. Compressed Mortality File 1968-1978. CDC WONDER Online Database, compiled from Compressed Mortality File CMF 1968-1988, Series 20, No. 2A, 2000. http://wonder.cdc.gov/cmf- icd8. html. Centers for Disease Control and Prevention, National Center for Health Statistics. Compressed Mortality File 1979-1998. CDC WONDER Online Database, compiled from Compressed Mortality File CMF 1968-1988, Series 20, No. 2A, 2000 and CMF 1989-1998, Series 20, No. 2E, 2003. http://wonder.cdc.gov/cmf-icd9.html. Centers for Disease Control and Prevention, National Center for Health Statistics. Compressed Mortality File 1999-2015 on CDC WONDER Online Database, released December 2016. Data are from the Compressed Mortality File 1999-2015 Series 20 No. 2U, 2016, as compiled from data provided by the 57 vital statistics jurisdictions through the Vital Statistics Cooperative Program. http://wonder.cdc. gov/cmf-icdlO.html. Comroe, J. H., and R. D. Dripps. 1976. “Scientific Basis for the Support of Biomedical Science.” Science 192 (4235): 105-11. https://doi.Org/10.l 126/science.769161. Corso, Phaedra S., Eric A. Finkelstein, Ted R. Miller, I Fiebelkom, and E Zaloshnja. 2006. “Incidence and Lifetime Costs of Injuries in the United States.” Injury Prevention 12 (4): 212-18. https://doi. org/10.1136/ip.2005.01Q983.

156 | Columbia Economic Review

The Efficacy Of Biomedical Research

Cortez-Pinto, Helena, Miguel Gouveia, Luis dos Santos Pinheiro, Joao Costa, Margarida Borges, and Antonio Vaz Cameiro. 2010. “The Burden of Disease and the Cost of Illness Attributable to Alcohol Drinking— Results of a National Study.” Alcoholism: Clinical and Experimental Research 34 (8): 1442-49. https://doi.Org/10.l 111/i.1530 0277.2010.01229.x. Dharmarajan, Kumar, Yongfei Wang, Zhenqiu Lin, Sharon-Lise T. Normand, Joseph S. Ross, Leora I. Horwitz, Nihar R. Desai, et al. 2017. “Association of Changing Hospital Readmission Rates With Mortality Rates After Hospital Discharge.” JAMA 318 (3): 270- 78. hhps://doi.org/10.1001/iama.2017.8444. Drye, Elizabeth, Sharon-Lise T. Normand, Yun Wang, Joseph S. Ross, Geoffrey C. Schreiner, Lein Han, Michael Rapp, and Harlan M. Krumholz. 2012. “Comparison of Hospital Risk- Standardized Mortality Rates Using In-Hospital and 30-Day Models: Implications for Hospital Profiling.” Annals of Internal Medicine 156 (1 Pt 1): 19-26. https://doi.org/! 0.1059/0003-4819-156-1-201201030-00004. Egger, Peter, and Michael Pfaffermayr. 2005. “Estimating Long and Short Run Effects in Static Panel Models.” Econometric Reviews 23 (3): 199-214. https://d0i.0rg/l 0.1081 /ETC- 200028201. “Fact Sheet: Medical Subject Headings (MeSH®).” 2015. Fact Sheets. U.S. National Library of Medicine. 2015. https://www.nlm.nih.gov/pubs/ factsheets/mesh.html. “Fact Sheet: MEDLINE®.” 2017. Fact Sheets. U.S. National Library of Medicine. 2017. https://www.nlm.nih.gov/pubs/factsheets/medline. html. “Fact Sheet: MEDLINE, PubMed, and PMC (PubMed Central): How Are They Different?” 2017. Fact Sheets. U.S. National Library of Medicine. 2017. https://www.nlm.nih.gov/pubs/factsheets/dif med_pub.html. Finkelstein, Eric A., Corso, Phaedra S., and Miller, Ted R. 2006. Incidence and Economic Burden of Injuries in the United States. Oxford University Press. “Fixed-Effects Dynamic Panel Models, a Factor Analytical Method - Bai - 2013 - Econometrica - Wiley Online Library.” n.d. Accessed April 20, 2018. https://onlinelibrarv.wilev.com/doi/abs/10.3982/ECTA9409. Fowlkes, Ashley, Sharoda Dasgupta, Edward Chao, Jennifer Lemmings, Kate Goodin, Meghan Harris, Karen Martin, et al. 2013. “Estimating Influenza Incidence and Rates of Influenza- like Illness in the Outpatient Setting.” Influenza and Other Respiratory Viruses 7 (5): 694- 700. https://doi.Org/10.l 11 l/irv.12014.

Tenth Anniversary | 157

Zoey Chopra

Gelijns, Annetine C., Joshua Graff Zivin, and Richard R. Nelson. 2001. “Uncertainty and Technological Change in Medicine.” Journal of Health Politics, Policy and Law 26 (5): 913-24. Global Burden of Disease Collaborative Network, 2017. Global Burden of Disease Study 2016 (GBD 2016) Results. Seattle, United States: Institute for Health Metrics and Evaluation (IHME). http://ghdx. healthdata.org/gbd-results-tool. “Global Health Expenditure Database.” 2014. World Health Organization (WHO). 2014. http://apps.who.int/nha/database. Gold, Marthe R., David Stevenson, and Dennis G. Fryback. 2002. “HALYS and QALYS and DALYS, Oh My: Similarities and Differences in Summary Measures of Population Health.” Annual Review of Public Health 23: 115-34. https://doi.org/10.1146/annurev. publhealth.23.100901.140513. Gore, J. M., C. J. Brophy, and M. A. Greenstone. 2000. “How Well Do We Care for Patients with End Stage Chronic Obstructive Pulmonary Disease (COPD)? A Comparison of Palliative Care and Quality of Life in COPD and Lung Cancer.” Thorax 55 (12): 1000- 1006. https:// doi.org/10.1136/thorax.55.12.1000. Grant, Jonathan, Liz Green, and Barbara Mason. 2003. “Basic Research and Health: A Reassessment of the Scientific Basis for the Support of Biomedical Science.” Research Evaluation 12 (3): 217-24. https:// doi.org/10.3152/1471544Q3781776618. Green, Lawrence W., Judith M. Ottoson, Cesar Garcia, and Robert A. Hiatt. 2009. “Diffusion Theory and Knowledge Dissemination, Utilization, and Integration in Public Health.” Annual Review of Public Health 30 (1): 151-74. https://doi.org/10.1146/annurev. ublhealth.031308.100Q49. Hall, Margaret Jean, Shaleah Levant, and Carol J. DeFrances. 2013. “Trends in Inpatient Hospital Deaths: National Hospital Discharge Survey, 2000-2010.” NCHS Data Brief No. 118. National Center for Health Statistics (NCHS). March 2013. https://www.cdc.gov/nchs/products/ databriefs/dbl 18.htm. Hammer, Jeffrey S., and Peter Berman. 1995. “Ends and Means in Public Health Policy in Developing Countries.” Health Policy 32 (1): 29-45. Hansen, Lars Peter. 1982. “Large Sample Properties of Generalized Method of Moments Estimators.” Econometrica 50 (4): 1029-54. https://doi.org/10.2307/1912775.

158 | Columbia Economic Review

The Efficacy Of Biomedical Research

Hays, Ross M., Jeanette Valentine, Gerri Haynes, J. Russel Geyer, Nanci Villareale, Beth Mckinstry, James W. Varni, and Shervin S. Churchill. 2006. “The Seattle Pediatric Palliative Care Project: Effects on Family Satisfaction and Health-Related Quality of Life.” Journal of Palliative Medicine 9 (3): 716-28. https://doi.org/10.1089/ipm.2006.9.716. “Healthcare Cost and Utilization Project (HCUPnet): Free Health Care Statistics.” 1997. U.S. Department of Health & Human Services Agency for Healthcare Research and Quality (AHRQ). 2014 1997. https://hcupnet.ahrq.gov. Holtz-Eakin, Douglas, Whitney Newey, and Harvey S. Rosen. 1988. “Estimating Vector Autoregressions with Panel Data.” Econometrica 56 (6): 1371-95. https://doi.org/l 0.2307/1913103. “Impact of NIH Research.” 2014. National Institutes of Health. November 21, 2014. https://www.nih.gov/about-nih/what-we-do/impact-nih-research/ our-health. Jann, Ben. 2014. “Plotting Regression Coefficients and Other Estimates.” The STATA Journal 14 (4): 708-37. Khoury, Muin J., Marta Gwinn, and John P. A. Ioannidis. 2010. “The Emergence of Translational Epidemiology: From Scientific Discovery to Population Health Impact.” American Journal of Epidemiology 172 (5): 517-24. https://d0i.0rg/l 0.1093/aie/kwq211. KC>9, Cagatay. 2004. “The Productivity of Health Care and Health Production Functions.” Health Economics 13 (8): 739-47. https://doi. org/10.1002/hec.855. Krumholz, Harlan M., Sudhakar V. Nuti, Nicholas S. Downing, Sharon-Lise T. Normand, and Yun Wang. 2015. “Mortality, Hospitalizations, and Expenditures for the Medicare Population Aged 65 Years or Older, 1999-2013.” JAMA 314 (4): 355-65. https://doi.org/10.1001/ iama.2015.8Q35. “Length Bias.” 2017. Family Practice Notebook, LLC. 2017. https://fpnotebook.com/Prevent/Epi/LngthBs.htm. Lichtenberg, Frank R. 2001. “The Allocation of Publicly-Funded Biomedical Research.” In Medical Care Output and Productivity, edited by Cutler, David M. and Berndt, Ernst R., 565-90. University of Chicago Press. http://www.ssrn.com/abstract=44640. 2013a. “The Impact of Therapeutic Procedure Innovation on Hospital Patient Longevity: Evidence from Western Australia, 2000-2007.” Social Science & Medicine 77 (January): 50-59. https://doi.Org/10.1016/j. socscimed.2012.l 1.004.

Tenth Anniversary | 159

Zoey Chopra

2013b. “The Impact of Biomedical Knowledge Accumulation on Mortality: A Bibliometric Analysis of Cancer Data.” NBER Working Paper 19593. National Bureau of Economic Research, https://doi.org/10.3386/ wl9593. 2014a. “The Impact of Pharmaceutical Innovation on Longevity and Medical Expenditure in France, 2000-2009.” Economics & Human Biology 13 (C): 107-27. 2014b. “The Impact of Pharmaceutical Innovation on Disability Days and the Use of Medical Services in the United States, 1997-2010.” Journal of Human Capital 8 (4): 432- 80. https://doi.org/10.1086/679110. 2015. “The Impact of Cardiovascular Drug Innovation on the Longevity of Elderly Residents of Switzerland, 2003-2012.” Nordic Journal of Health Economics 5 (1): 23-44. https://doi.org/10.5617/nihe.1291. Lichtenberg, Frank R. 2016. Benefits of Pharmaceutical Innovation: Health, Longevity, and Savings. Place of publication not identified: Montreal Economic Institute. http://deslibris.ca/ID/l 0065527. Martin, Laura J. n.d. “Smoking and COPD.” U.S. National Library of Medicine MedlinePlus. Accessed March 19, 2018. https://medlineplus. gov/encv/patientinstructions/000696.htm. Matherlee, K. R. 1995. “The Outlook for Clinical Research: Impacts of Federal Funding Restraint and Private Sector Reconfiguration.” Academic Medicine: Journal of the Association of American Medical Colleges 70 (12): 1065-72. McGeary, Michael, and Kathi E. Hanna, eds. 2004. “Sources of Funding for Biomedical Research.” In Strategies to Leverage Research Funding: Guiding DOD Peer Reviewed Medical Research Programs. Washington (DC): National Academies Press (US). https://www.ncbi.nlm.nih.gov/ books/NBK215472/. McKinlay, John B. 1981. “From ‘Promising Report’ to ‘Standard Procedure’: Seven Stages in the Career of a Medical Innovation.” The Milbank Memorial Fund Quarterly. Health and Society 59 (3): 374 411. https://doi.org/10.2307/3349685. “Medical Advances.” 2016. Foundation for Biomedical Research. 2017 2016. https://fbresearch.org/medical-advances/. Morris, Zoe Slote, Wooding, Steven, and Grant, Jonathan. 2011. “The Answer Is 17 Years, What Is the Question: Understanding Time Lags in Translational Research.” Journal of the Royal Society of Medicine 104 (12): 510-20. https://doi.org/10.1258/irsm.2011.110180. Moses, Hamilton III, and Joseph B. Martin. 2011. “Biomedical Research and Health Advances.” New England Journal of Medicine 364 (6): 567-71. https://doi.Org/l 0.1056/NEJMsb 1007634.

160 | Columbia Economic Review

The Efficacy Of Biomedical Research

Murray, C. J. 1994. “Quantifying the Burden of Disease: The Technical Basis for Disability- Adjusted Life Years.” Bulletin of the World Health Organization 72 (3): 429-45. Nabel, Elizabeth G. 2009. “Linking Biomedical Research to Health Care.” The Journal of Clinical Investigation 119 (10): 2858. https://doi.0rg/ lO.l 172/JCI41035. National Cancer Institute. Surveillance, Epidemiology, and End Results (SEER) Program. SEER*Stat Database: Incidence - SEER 9 Regs Research Data, Nov 2016 Sub (1973-2014), National Cancer Institute, DCCPS, Surveillance Research Program, Surveillance Systems Branch, released April 2017, based on the November 2016 submission. Cancer Query System. Nickell, Stephen. 1981. “Biases in Dynamic Models with Fixed Effects.” Econometrica 49 (6): 1417-26. https://doi.org/10.2307/1911408. “NIH Research Portfolio Online Reporting Tools (RePORT): Estimates of Funding for Various Research, Condition, and Disease Categories (RCDC).” 2017. U.S. Department of Health & Human Services. 2017. https://report.nih.gov/categorical spending.aspx. “NIH Research Portfolio Online Reporting Tools (RePORT): EXPORTER.” 2017. U.S. Department of Health & Human Services. 2017. https:// exporter.nih.gov/ExPORTER Catalog.aspx. “NIH Research Portfolio Online Reporting Tools (RePORT): Frequently Requested Reports: Table #301: Federal Obligations for Health Research and Development by Federal Agency.” 2017. U.S. Department of Health & Human Services. 2017. https://report.nih.gov/ffrs/index.aspx. Parks, Rachel. 2014. “The Rise, Critique and Persistence of the DALY in Global Health.” The Journal of Global Health, April, //www.ghiournal.org/the rise-critique-and-persistence-of- the-dalv-in-global-health/. Pouw, Maurice E., L. M. Peelen, K. G. M. Moons, C. J. Kalkman, and H. F. Lingsma. 2013. “Including Post-Discharge Mortality in Calculation of Hospital Standardised Mortality Ratios: Retrospective Analysis of Hospital Episode Statistics.” 347 (October): f5913. https://doi. org/10.1136/bmi.f5913. “Premature Mortality in the United States: Public Health Issues in the Use of Years of Potential Life Lost.” 1986. Centers for Disease Control and Prevention Morbidity and Mortality Weekly Report. December 19, 1986. https://www.cdc.gov/mmwr/preview/mmwrhtml/000Q1773.htm. PubMed Help [Internet]. 2018. [Updated 2018 Feb 5]. Bethesda, MD: National Center for Biotechnology Information (US). https://www.ncbi. nlm.nih.gov/books/NBK3827/.

Tenth Anniversary | 161

Zoey Chopra

Rahmqvist, Mikael, Annika Samuelsson, Salumeh Bastami, and Hans Rutberg. 2016. “Direct Health Care Costs and Length of Hospital Stay Related to Health Care-Acquired Infections in Adult Patients Based on Point Prevalence Measurements.” American Journal of Infection Control 44 (5): 500-506. https://doi.Org/10.1016/i.aiic.2016.01.035. Reichert, Janice M., and Christopher-Paul Milne. 2002. “Public and Private Sector Contributions to the Discovery and Development of ‘Impact’ Drugs.” American Journal of Therapeutics 9(6): 543-55 Romeder, J. M., and J. R. McWhinnie. 1977. “Potential Years of Life Lost between Ages 1 and 70: An Indicator of Premature Mortality for Health Planning.” International Journal of Epidemiology6 (2): 143-51. Romer, Paul M. 1990. “Endogenous Technological Change.” Journal of Political Economy 98 (5, Part 2): S71-102. https://doi.org/10.1086/261725. Romley, John A., and Neeraj Sood. 2013. “Identifying the Health Production Function: The Case of Hospitals.” Working Paper 19490. National Bureau of Economic Research (NBER). https://doi.org/10.3386/ wl949Q. Roodman, David. 2009. “How to Do Xtabond2: An Introduction to Difference and System GMM in Stata.” Stata Journal 9(1): 88-136. Rothberg, Michael B., Joshua Cohen, Peter Lindenauer, Judith Maselli, and Andy Auerbach. 2010. “Little Evidence of Correlation between Growth in Health Care Spending and Reduced Mortality.” Health Affairs (Project Hope) 29 (8): 1523-31. hhps://doi.org/10.1377/ hlthaff.2009.0287. Sampat, Bhaven N., and Frank R. Lichtenberg. 2011. “What Are the Respective Roles of the Public and Private Sectors in Pharmaceutical Innovation?” Health Affairs (Project Hope) 30 (2): 332-39. https:// d0i.0rg/l0.1377/hlthaff.2009.0917. Seemungal, T. A., G. C. Donaldson, A. Bhowmik, D. J. Jeffries, and J. A. Wedzicha. 2000. “Time Course and Recovery of Exacerbations in Patients with Chronic Obstructive Pulmonary Disease.” American Journal of Respiratory and Critical Care Medicine 161 (5): 1608-13. https://doi.Org/10.l 164/airccm. 161.5.9908022. Smith, Thomas J., Sarah Temin, Erin R. Alesi, Amy P. Abernethy, Tracy A. Balboni, Ethan M. Basch, Betty R. Ferrell, et al. 2012. “American Society of Clinical Oncology Provisional Clinical Opinion: The Integration of Palliative Care Into Standard Oncology Care.” Journal of Clinical Oncology 30 (8): 880-87. https://doi.org/! 0.1200/ JCQ.2011.38.5161.

162 | Columbia Economic Review

The Efficacy Of Biomedical Research

Sogaard, Mette, Rikke B. Nielsen, Henrik C. Schonheyder, Mette Norgaard, and Reimar W. Thomsen. 2014. “Nationwide Trends in Pneumonia Hospitalization Rates and Mortality, Denmark 1997-2011.” Respiratory Medicine 108 (8): 1214-22. https://doi.org/10.1016/i .rmed.2014.05.004. Stern, Jerome M., and R. John Simes. 1997. “Publication Bias: Evidence of Delayed Publication in a Cohort Study of Clinical Research Projects.” BMJ: British Medical Journal 315 (7109): 640—45. Stukel, Therese A., Elliott S. Fisher, David A. Alter, Astrid Guttmann, Dennis T. Ko, Kinwah Fung, Walter P. Wodchis, Nancy N. Baxter, Craig C. Earle, and Douglas S. Lee. 2012. “Association of Hospital Spending Intensity With Mortality and Readmission Rates in Ontario Hospitals.” Jama307 (10): 1037—45. https://doi.org/10.1001/ia ma.2Q12.265. Tamim, H., A. A. Tahami Monfared, and J. LeLorier. 2007. “Application of Lag-Time into Exposure Definitions to Control for Protopathic Bias.” Pharmacoepidemiology and Drug Safety 16 (3): 250-58. https://doi. org/10.1002/pds.1360. Tejada-Vera, B., Chong, Y., and Lu, L. 2015. “Top 10 Leading Causes of Death: United States, 1999-2013.” National Center for Health Statistics. 2015. https://blogs.cdc.gov/nchs-data- visualization/leading-causes-of death/. “Terms Defined.” 2018. Institute for Health Metrics and Evaluation. 2018. http://www.healthdata.org/terms-defmed. “The Benefits: Biomedical Research.” 2015. National Association for Biomedical Research. 2015. http://www.nabr.org/biomedical-research/ the-benefits/. “The Value of Federally Funded Biomedical Research.” 2017. Federation of American Societies for Experimental Biology. 2017. http://faseb.org/ Science-Policv~Advocacv-and- Communications/Become-an Advocate/Benefits-of-Biomedical-Research.aspx. Tsugawa, Yusuke, Anupam B. Jena, Jose F. Figueroa, E. John Orav, Daniel M. Blumenthal, and Ashish K. Jha. 2017. “Comparison of Hospital Mortality and Readmission Rates for Medicare Patients Treated by Male vs Female Physicians.” JAMA Internal Medicine 177 (2): 206-13. https://doi.org/10.1001/iamainternmed.2016.7875. “UlS.Stat: Science, Technology, and Innovation Dataset.” n.d. UNESCO Institute for Statistics. Accessed March 20, 2018. http://data.uis. unesco.org/.

Tenth Anniversary | 163

Zoey Chopra

Ward, Michael M. 2013. “Estimating Disease Prevalence and Incidence Using Administrative Data: Some Assembly Required.” The Journal of Rheumatology 40 (8): 1241—43. https://doi.org/10.3899/ irheum.130675. Welch, H. Gilbert, Lisa M. Schwartz, and Steven Woloshin. 2000. “Are Increasing 5-Year Survival Rates Evidence of Success Against Cancer?” 283 (22): 2975-78. Wilbowo, D. H„ and C. A. Tisdell. 1992. “HEALTH PRODUCTION FUNCTION FOR PREVENTIVE HEALTH PROGRAMS.” In Proceedings of the Fourteenth Australian Conference of Health Economists, 106-33. Faculty of Economics, Commerce and Management Monash University and National Centre for Health Program Evaluation Fairfield Hospital. Wratschko, Katharina. 2009. “Empirical Setting: The Pharmaceutical Industry.” In Strategic Orientation and Alliance Portfolio Configuration: The Interdependence of Strategy and Alliance Portfolio Management, 87-96. Gabler. Yoshida, Katsumi, Hiroki Sugimori, Yoshihiro Yamada, Takashi Izuno, Michiko Miyakawa, Chieko Tanaka, and Eiko Takahashi. 1997. “Years of Potential Life Lost as the Indicator of Premature Mortality in Occupational Medicine.” Environmental Health and Preventive Medicine 2 (1): 40-44. https://doi.org/10.1007/BFQ2931228. Yves, Croissant, and Millo Giovanni. 2008. “Panel Data Econometrics in R: The Plm Package.” Journal of Statistical Software 27 (2): 1-43.

164 | Columbia Economic Review

The Efficacy Of Biomedical Research

Tenth Anniversary | 165

Zoey Chopra

166 | Columbia Economic Review

The Efficacy Of Biomedical Research

Tenth Anniversary | 167

Zoey Chopra

168 | Columbia Economic Review

The Efficacy Of Biomedical Research

Tenth Anniversary | 169

Zoey Chopra

170 | Columbia Economic Review

The Efficacy Of Biomedical Research

Tenth Anniversary | 171

Zoey Chopra

172 | Columbia Economic Review

The Efficacy Of Biomedical Research

Tenth Anniversary | 173

Zoey Chopra

174 | Columbia Economic Review

The Efficacy Of Biomedical Research

Tenth Anniversary | 175

Zoey Chopra

176 | Columbia Economic Review

The Efficacy Of Biomedical Research

Tenth Anniversary | 177

Zoey Chopra

178 | Columbia Economic Review

The Efficacy Of Biomedical Research

Tenth Anniversary | 179

Zoey Chopra

180 | Columbia Economic Review

The Efficacy Of Biomedical Research

Tenth Anniversary | 181

Zoey Chopra

182 | Columbia Economic Review

The Efficacy Of Biomedical Research

Tenth Anniversary | 183

Zoey Chopra

184 | Columbia Economic Review

The Efficacy Of Biomedical Research

Tenth Anniversary | 185

Zoey Chopra

186 | Columbia Economic Review

The Efficacy Of Biomedical Research

Tenth Anniversary | 187

Zoey Chopra

188 | Columbia Economic Review

Founded in 2009 as the first undergraduate economic journal in the United States, the Columbia Economic Review (CER) aims to promote discourse and research at the intersection of economics, business, politics, and society by publishing a rigorous selection of research papers in our print journal. We further strive to engage individuals on campus, locally, and globally through speaker series, symposia, competitions, and other events established to promote dialogue and encourage deeper insights on economic issues. CER is sponsored by the Program for Economic Research (PER) at Columbia University, and is entirely led, organized, and operated by undergraduate students at Columbia University across a range of academic disciplines.

econreview@columbia.edu columbiaeconreview.com

COLUMBIA ECONOMIC REVIEW | THE TENTH ANNIVERSARY ISSUE