__MAIN_TEXT__
feature-image

Page 1

BER

Berkeley Economic Review

IN THIS ISSUE ESSAY TOPIC The Economic Impact of Refugee Immigration

PAPERS Singapore’s Primary 1 Registration: Redesigning the School Choice Mechanism


THE BERKELEY ECONOMIC REVIEW

TABLE OF CONTENTS

ESSAY CONTEST America Is Failing Both Refugees and Itself...................................................................................................................................9 Daniel Cohen Investing in Refuge: The Opportunity Cost of Recluse..............................................................................................................10 Devesh Agarwal Help the World, Help Ourselves!.......................................................................................................................................................11 Dan Ma

RESEARCH PAPERS Noise Complaints and School Performance: Evidence from New York City.....................................................................12 Eric Qian (University of North Carolina, Chapel Hill) Singapore's Primary 1 Registration: Redesigning the School Choice Mechanism..........................................................46 Ruru Hoong (Stanford University) Human Capital Investment: An Examination of the Cyclicality of Bachelor's Degrees Conferred by Field of Study...........................................................................................................................................................................70 Mark Vandre (University of California, Berkeley)

2


VOLUME V

PEER REVIEW BOARD (IN ALPHABETICAL ORDER) KITAE AHN SUDESHNA BARMAN MARGARET CHEN DIVYA DHAR JOSEPH HERNANDEZ ANGEL HSU ASHWIN IYER TOMMY LAO KATHERINE SHENG GRACE WANG SIMON ZHU

SUBMISSIONS POLICY Format: Please format your submission as a Microsoft word document or Latex file. We would prefer submissions single-spaced in size 12, Times New Roman font. Length: Maximum five pages single-spaced for essays or op-eds, not including works cited page(s). No maximum for theses or original research papers. Plagiarism: We maintain a strict zero tolerance policy on plagiarism. According to the University of North Carolina, plagiarism is “the deliberate or reckless representation of another’s words, thoughts, or ideas as one’s own without attribution in connection with submission of academic work, whether graded or otherwise.” For more information about plagiarism – what it is, its consequences, and how to avoid it – please see the UNC’s website. Citations: Use MLA format for in-text citations and your works cited page. Include a separate works cited page at the end of your work. Send to: berkeleyeconreview@gmail.com

DISCLAIMER The views published in this journal are those of the individual writers or speakers and do not necessarily reflect the position or policy of The Undergraduate Berkeley Economic Review, the Undergraduate Economics Association, or the University of California at Berkeley.

COPYRIGHT POLICY All authors retain copyright over their original work. No part of our journal, whether text or image, may be used for any purpose other than personal use. For permission to reproduce, modify, or copy materials printed in this journal for anything other than personal use, kindly contact the respective authors.

3


THE BERKELEY ECONOMIC REVIEW

STAFF

Managing Editor Elena Stacy, '18 Economics

Content Edit Lead, Peer Reviewer, Recruitment Manager Margaret Chen, '19 Economics, Applied Mathematics

Managing Editor, Peer Review Lead Ashwin Iyer , '18 Economics, Political Science

Marketing Lead Joseph Hernandez, '19 Economics, History

Social Media Manager Marketing Lead Charles McMurry, '21 Juliana Zhao, '19 Economics, Economics, Business Administration Applied Mathematics 4

Layout Lead, Peer Reviewer, Recruitment Manager Katherine Sheng, '19 Economics, Statistics

Peer Reviewer Divya Dhar, '18 Economics


VOLUME V

STAFF

Content Editor, Layout Staff Anastasia Pyrinis, '20 Economics, Political Science, Media Studies

Marketing Staff Adam Andrews, '18 Bioengineering

Marketing Staff Ammar Inayalati, '19 Economics

Content Editor Avir Waxman, '18 Economics, History

Staff Writer Matthew Forbes, '20 Economics, Statistics

Peer Reviewer Angel Hsu, '19 Economics, Business Administration

Marketing Staff Abby Li, '21 Molecular and Cell Biology

Content Editor Rachel Hobbs, '19 Applied Mathematics

Marketing Staff Reilly Olson, '20 Economics

5


THE BERKELEY ECONOMIC REVIEW

STAFF

Marketing Staff Selena Zhang, '21 Economics, Rhetoric

Peer Reviewer Simon Zhu, '19 Economics, Applied Mathematics, Statistics

Peer Reviewer Tommy Lao, '18 Economics, Applied Mathematics

Staff Writer Vatsal Bajaj '21 Economics, Computer Science

Content Editor Sofia Guo, '19 Economics

Staff Writer, Content Editor Vinay Maruri, '20 Economics, Copmputer Science

NOT PICTURED: Layout Editor Shannon Balan, '21 Political Economy, Business Administration

Layout Editor Staff Writer Denesh Chandrahasan, '19 Jacob Fajnor, '21 Political Economy Economics, Applied Mathematics

6

Content Editor Renuka Garg, '20 Economics, Statistics Layout Editor Dana Wu, '20 Economics


VOLUME V

[THIS PAGE INTENTIONALLY LEFT BLANK]

7


THE BERKELEY ECONOMIC REVIEW

SPRING 2018 ESSAY CONTEST Would allowing more immigration of refugees be good both for refugees and for the economies they come to?

2


VOLUME V

America is Failing Both Refugees and Itself 1st Place Winner Daniel Cohen

While prosperity attracts many asylum-seekers, refugees benefit from emigration in other ways. In many cases, expatriation relieves the burden of brutal cultural cleansing and allows communities to be rebuilt.3 For Middle Eastern refugees, who comprise 40% of the world's displaced population,4 migration removes the threat of forcible recruitment by ISIL5 and allows strained familial relationships to be restored. Regardless of potential upside, the US is in no position to facilitate these benefits. America's job training programs, a necessity for refugee integration, are already sparse, and planned cuts will prevent refugees from accessing whatever resources remain.6 Arguably the country's most fundamental "career preparation program," its public primary and secondary education system, cannot even provide for its own citizens.7 America has no bandwidth to accept and educate refugees, but it also lacks a desire to remedy that issue. Recent Gallup and Quinnipiac polls found that a majority of Americans oppose admitting even 10,000 refugees,8 an unnoticeably small fraction of the 65.3 million9 that seek asylum worldwide. Nearly 60% of voters supported President Trump's wildly controversial Muslim travel ban.10 Even as major corporate employers like Walmart and

Class of 2021 Economics, Data Science

America is proudly dubbed the country of immigrants: founded by British settlers and filled with countless other foreign populations, this nation once epitomized inclusion. Yet as that storied immigrant heritage loses priority, some Americans have reverted to xenophobia and cultural protectionism, ignoring millions of refugees in the Middle East who anxiously await any escape from violence, food insecurity, and religious persecution. Although widespread integration could simultaneously strengthen the economy and provide stability for millions of displaced families, the altruism required to facilitate this change is embraced by too few Americans on too narrow a scale. This is an unfortunate truth, as refugees generally enhance economies that accept them. Asylum-seekers are nearly 50% more entrepreneurial than the average American, and as a result their earnings rapidly outpace initial government assistance. After two decades, median refugee income is 25% higher than median US income,1 and the American government earns over $20,000 in net tax revenue per refugee.2 If a nation has the resources to integrate refugees into the workforce, these economic benefits would scale indefinitely.

Starbucks hire increasing numbers of asylumseeking immigrants,11 this widespread political animosity precludes refugee integration and prevents Americans from accessing the economic benefits that accompany it. In the right economy at the right time, refugees bring prosperity and receive it in return. But as a nation of immigrants alienates the group that once allowed it to flourish, America's refugee-backed economic prosperity is in jeopardy. _____________________________ 1 New American Economy, From Struggle to Resilience: The Economic Impact of Refugees in America (June 19, 2017) 2 William N. Evans, Daniel Fitzgerald, The Economic and Social Outcomes of Refugees in the United States: Evidence from the ACS (NBER, June 2017) 3 Patrick Sisson, Syrian Refugees' Miniature Models Celebrate Threatened and Demolished Landmarks (Curbed, Jan 21, 2016) 4 Eric Reguly, The great displacement: Middle East now has more migrants than Spain has people (The Globe and Mail, November 12, 2017) 5 Musa al-Gharbi, To Defeat ISIS, Embrace Refugees (Middle East Policy Council, 2017) 6 Lydia DePillis, The U.S. needs to do a better job training its workers. Here's how (CNN Money, March 1, 2018) 7 Alana Semuels, Good School, Rich School; Bad School, Poor School (The Atlantic, Aug 25, 2016) 8 Jeremy L. Neufeld, What Americans Think About Refugees is More Complicated Than You Think (Niskanen Center, March 13, 2017) 9 Camila Domonoske, Refugees, Displaced People Surpass 60 Million For First Time, UNHCR Says (NPR, June 20, 2016) 10 Steven Shepard, Poll: Majority of voters back Trump travel ban (Politico, May 7, 2017) 11 Brad Tuttle, These American Companies Are Hiring Refugees — Even When It's Not Very Popular (Time Money, February 8, 2018)

9


THE BERKELEY ECONOMIC REVIEW

Investing in Refuge: The Opportunity Cost of Refuse Whether or not building a wall is a good idea is a moral debate and is better left to the political spectrum on CNN. But whether the perpetrators jumping the wall would benefit the land that they land on is far more quantifiable. The global political sphere is dominated by vague, ignorant deliberation about the loss of jobs and imminent annihilation of economies caused by an increasing influx of refugees. Deeper analysis reveals less intuitive and more encouraging repercussions of the same. First, the benefit of this bilateral relationship on refugees is undeniable by virtue of the very definition of the term. Refugees are those displaced from conflicted areas; where there is either economic incapability or conditions like war or natural disaster that are conducive to economic incapability their economic benefit from migrating to safe and stabler economies is hardly contentious. Second, contrary to popular belief, if we look at raw numbers, host countries significantly gain from a refugee influx. Notre Dame economist William Evans' study finds an approximate $107,000 spent in resettlement of working-age refugees on behalf of the US Government over a period of 20 years. This number is eclipsed by numbers north of $130,000 that refugees pay as taxes over the same period. Addressing, once and for all, the idea of everyone losing their jobs to refugees is the notion of immigrantnative complementarity. A 2015 study conducted by Mette Foged and Giovanni Peri observed the Danish 10

refugee intake and its effects on the labor market. A key finding was that the little 2nd Place Winner displacement of native Devesh Agrawal workers from jobs was Class of 2021 simply into other jobs of Business, Data Science more complex natures that demand proficiency It is computed to take eight or skill exclusive to these natives. This allows natives and refugees to allocate years for a refugee to become a net contributor to the US economy their respective human resource and even shorter periods in several more efficiently and maximize total countries in Europe. Despite the productivity for themselves as well moral dissension, one thing is as with the host economy. This is for certain: refugees are a longsubstantiated by the fact that the term investment which slowly, but most-affected natives in Denmark in surely reap mutual benefits for host this study had a median 3% increase economies. in income. _____________________________ Another pivotal argument of wallbuilders is the raw surge in refugee Works Cited count due to geopolitical turmoil. Casteel, Kathryn, and Michelle Cheng. The Hungarian Revolution in 1956 "Refugees May Be Good For The Economy." saw over 200,000 refugees enter FiveThirtyEight, FiveThirtyEight, 14 June Austria, an alarming 3% of Austria's 2017. own population. It is irrefutable that Clemens, Michael. “The Real Economic if such a large percentage of the host Cost of Accepting Refugees." Refugees, News country's population in refugees Deeply, 8 Aug. 2017. enters the country, it will be a huge economic burden and all marginal Evans, William N., and Daniel Fitzgerald. The Economic and Social Outcomes of Refugees benefits afore-discussed would be in the United States: Evidence from the ACS. , rendered void. However, we live, June 2017 today, in a globalized world where such a burden need not be born by Foged, Mette, and Giovanni Peri. How any nation alone. Just like the 200,000 Immigrants and Job Mobility Help LowHungarian refugees were rehabilitated Skilled Workers. 19 Apr. 2015. among 37 allied nations in 1956, the United Nations Commission High Commission on Refugees (UNHCR) and other bodies perform in an institutionalized manner to share responsibility of refuge today. This allows member states to economically gain from taking in numbers of refugees marginal to their population.


VOLUME V

Help the World, Help Ourselves! 3rd Place Winner Dan Ma Class of 2021 Environmental Economics & Policy

I am the son of refugees. My mother grew up in rural Vietnam where she did not have access to running water and faced the violence of war. In the U.S., my family settled in a diasporic community called Little Saigon in Westminster, CA. Vietnamese refugees who settle there not only provided skilled labor to companies such as Medtronic, where Vietnamese women continue to apply their sewing skills to handmake artificial heart valves, but also establish new businesses that create jobs and add to the diverse culture of the community. Through our foods, such as pho, as well as out traditions, such as Lunar New Year festivals, we develop a richer community experience for all. Refugees are forced to flee their country because of persecution, war, or violence for reasons such as race, religion, nationality, political opinion, or membership in a particular social group.1 Recent waves of refugees are coming from Syria, Afghanistan, Lake Chad Basin, South Sudan, and Somalia, where hunger and violence are forcing people to leave their homes.2 Though these refugees face language and educational barriers in the countries they resettle in, their ability to adapt and gain employment demonstrate their ability to improve their lives given the opportunity to

resettle. In addition, these refugees become an asset to the country they settle in by contributing taxes, adding skills, and creating jobs that benefit the society they come to as a whole. Estimates by William N. Evans and Daniel Fitzgerald for a National Bureau of Economic Research report show that refugees contribute "$21,000 more in taxes than they receive in benefits over their first 20 years in the U.S." On the socioeconomic outcomes of refugees, the report also shows that "refugees that enter the U.S. before 14 graduate high school and enter college at the same rate as natives" even though they have initially poor outcomes with low employment and high welfare use.3 Analysis by Giovanni Peri of the University of California, Davis also suggest that "an increase in the supply of refugee-country immigrants pushed less educated native workers (especially the young and lowtenured ones) to pursue less manualintensive occupations. As a result immigration had positive effects on native unskilled wages, employment and occupational mobility."4 The data hows the economic benefits that these immigrants can create for themselves and their economies, but they do not show the moral as well as ethical imperative of accepting refugees. Creating a paradigm that promotes cross-

cultural tolerance and understanding enables peace as well as prosperity. Policies that enable world citizens in need to seek refuge from poverty, violence, and discrimination are necessary to creating a society that values compassion and trust. _____________________________ 1 "What Is a Refugee?" Accessed April 24, 2018. https://www.unrefugees.org/refugee-facts/whatis-a- refugee/. 2 Huber, Chris. "Forced to Flee: Top 5 Countries Refugees Are Coming from." World Vision. February 27, 2018. Accessed April 15, 2018. https://www.worldvision.org/refugees­news­stories/ forced­flee­top­5­countries­refugees 3 Evans, William, and Daniel Fitzgerald. "The Economic and Social Outcomes of Refugees in the United States: Evidence from the ACS." National Bureau of Economic Research, March 2015. doi:10.3386/w23498. 4 Foged, Mette, and Giovanni Peri. "Immigrants and Native Workers: New Analysis on Longitudinal Data." Institute of Labor Economics, June 2017. doi:10.3386/w19315.

11


THE BERKELEY ECONOMIC REVIEW

Noise Complaints and School Performance: Evidence from New York City

by Eric Qian (University of North Carolina, Chapel Hill)


VOLUME V


THE BERKELEY ECONOMIC REVIEW

Noise Complaints and School Performance: Evidence from New York City by Eric Qian University of Nirth Carolina, Chapel Hill

14


VOLUME V

Abstract

A students’ learning environment is an important component for educational performance. Environmental factors may hinder student learning, especially in disadvantaged neighborhoods. This paper uses New York City’s 20112015 311 complaints dataset to examine the effect of noise on elementary school academic performance. Using multivariate ordinary least squares, yearly fixed effects, propensity score matching, and spatial regression models to control for neighborhood characteristics, I show that an increase in noise complaints during sleeping hours leads to lower test scores. This suggests that street-level noise reduction in neighborhoods may lead to improved learning. Human capital is an important determinant for economic growth, and one major driver is educaiton attainment (Cobb-Clark et al., 2012). Places with higher quality educational resources create more productive workers, so an increase in education quality would benefit regional and national growth. In particular, early level education in impoverished areas is a key pathway for decreasing intergenerational income inequality (Black and Devereux, 2010). In sum, increasing schooling quality is beneficial for general welfare due to significant positive exernalities on social welfare. Past research has focused on examining the schools themselves. The benefits of teacher quality, racial composition, school location, and lunch programs are well documented (Reiling, 2016; Lleras, 2008). However, students spend much of their time at home, so studying neighborhood effects is critical. Traditionally, in studying environmental determinants for educational quality, researchers have been constrained by data. Geographic and neighborhood-quality variables are notoriously difficult to measure. Objective measures of noise disturbances have required the use of sound meters, which can be costly if implemented across large areas. As a result, analysis connecting noise-level to educational performance is largely limited to the school setting (Gruber et al., 2014; Dewald et al., 2010; Shield Dockrell, 2008; Basner et al., 2011).

15


THE BERKELEY ECONOMIC REVIEW

In this paper, I shed light on to neighborhood-level environmental factors. I examine whether nighttime noise events play a meaningful role in student learning by using New York City’s 311 complaints data.1 The results suggest that noise during early sleeping hours strongly affects students’ academic performance. 311 complaints data can help broaden our understanding of enviromental disturbances. First, complaints data allow researchers to distinguish between noise types (e.g. traffic, railroad). This is key in isolating different acoustical properties. Second, complaints are generally reported for specific incidences. While constant noise is bothersome, noise events (e.g. car horn) are more influential in disrupting sleep. Third, 311 complaints are ultimately a subjective measure; people file complaints when noise is intrusive. To these ends, the measure is advantageous because complaints are directly related to general annoyance. Fourth, the dataset provides a detailed account for each disturbance. Information about incident date, time, exact location, and location type help inform the analysis. The remainder of the paper proceeds as follows. Section 1 discusses past literature on sleep and focus. Section 2 describes the datasets. “Noise” is measured through New York City’s 311 municipal complaints dataset. Additional variables are obtained from New York’s Department of Education, the 2010-2014 American Community Survey, and New York City’s open data portal. Section 3 gives the results of the analysis. After controlling for neighborhood and school characteristics, noise compaints still have a significant relation based on ordinary least squares regression analysis. The effect also holds after using yearly fixed-effects, propensity score matching, and spatial regression models. The paper concludes with Section 4, which provides discussion of future research areas.

311 is a phone number used in many North American municipalities to provide non-emergency services. In New York City’s case, the service is often used to report noise and other complaints. 1

16


VOLUME V

1. Literature Review

Noise does not connect to test results directly; it influences scores through behavioral changes. Based on literature in sleep and psychology, I focus on how noise affects student behavior indirectly through sleep. In Appendix A, I discuss using noise complaints to measure school-hour effects. For the analysis, I include only complaints during school hours, and find that noise complaints are a poor measure for daytime complaints. In a school environment, other distractions are likely more influential. Quality sleep is generally accepted as an important factor for daytime function. It plays a role in physical and psychological health. Sleep aids in memory consolidation, which converts initial memory into more stable forms (Stickgold and Walker, 2007). This is especially important for retaining knowledge. In addition, sleep is a factor for maintaining attention, and deprivation can lead to shortened attention spans (Chua et al., 2014). Through reduced sleep, students face greater difficulty in processing and retaining knowledge. Although sleep is important for learning, many students are not getting enough. Children and adolescents on average require around nine hours of sleep a night, yet 45% sleep eight hours or less (Mercer et al., 1998; Gibson et al., 2006). A 2010 meta-analysis found that quality and general sleepiness are significantly related to school performance, especially for younger children (Dewald et al., 2010). These results suggest that analysis should not solely be focused on duration; quality also plays a large role. While general noise can disrupt sleep patterns, specific noise events are especially influential. One approach to examining disturbance events is to study nighttime air traffic. A 2012 review of 12 studies suggests a link between aircraft noise and sleep disturbances (Perron et al., 2012). Subjects were more likely to be awake, have shorter slow-wave sleep, and to use medication. The results suggest that while adaptation to noise may occur, overall sleep quality still decreases. Another study examined the effect of different disturbance types on sleep, and found road noise to be especially problematic (Basner et al., 2011). Compared to aircraft and railroad noise, traffic noise was more likely to be awake, 17


THE BERKELEY ECONOMIC REVIEW

have shorter slow-wave sleep, and to use medication. The results suggest that while adaptation to noise may occur, overall sleep quality still decreases. Another study examined the effect of different disturbance types on sleep, and found road noise to be especially problematic (Basner et al., 2011). Compared to aircraft and railroad noise, traffic noise was more likely to disrupt sleep structure and decrease duration. The difference was attributed to traffic’s acoustical properties. At the populationlevel, road noise, particularly for events above 55 decibels, have been found to be a factor for insomnia symptoms (Halonen et al., 2012). The findings are particularly relevant for urban areas, where noise level and disturbance frequency are generally higher. The link between sleep and educational performance is confounded by an assortment of neighborhood and socioeconomic characteristics. This section will detail some of these mechanisms. Neighborhoods form the communities where students spend much of their non-school hours. Parental education level strongly influences student academic achievement. Parents with higher levels of education attainment are more involved with their child’s academic lives, resulting in better learning (Jeynes, 2005). In addition to educational effects within neighborhoods, economic factors tell a similar story. Parental employment is another strong factor for inf luencing educational performance. Joblessness decreases parents’ ability to provide a supportive learning environment, resulting in lower performance (Reynolds et al., 2004). Besides direct parental effects, unemployment captures general neighborhood characteristics. Neighborhoods with higher unemployment generally have lower tax bases, providing less money for city services and infrastructure. Although New York City’s municipal government pulls resources from economically diverse areas, expenditure doesn’t necessarily provide equal benefit. Wealthy areas often disproportionately benefit from larger tax bases (Hwang and Sampson, 2014). Aditionally, the effect of crime on neighborhood and educational performance is well documented. At the family-level, convicted parents often have convicted children. Without a stable living and neighborhood environments, criminal behavior is often transmitted intergenerationally (Farrington et al., 2009). 18


VOLUME V

Previous literature also suggests that crime negatively affects childrens’ educational attainment. Growing up in an environment with a convicted parent contributes to a familial instability, thus negatively impacting cognitive development (Rud et al., 2014; Geller et al., 2009). Negative peer effects can also yield poor learning in school (Nesmith and Ruhland, 2008). Neighborhoods aren’t the only environments that affect learning. The schools themselves are an important component of student life. Next, I’ll detail several school-level confounders. The student-teacher ratio is a proxy for the classroom environment. Teachers are better able to control classroom environment in lower student-teacher ratio classrooms, and would likely be more conducive for learning (Graue and Oen, 2008). In practice, the relationship between class size and academic outcomes isn’t entirely clear. Principals have leeway in determining class size, and frequently, more effective teachers are assigned to teach more students (Barrett and Toma, 2013). Conversely, schools with higher proportion of children with learning disabilities use resources more intensely. Yet these schools have lower average test scores. Controlling for class size and schools with higher proportions of students with learning disabilities would help model classroom environment. Additionally, elementary school cohort size has been found to be influential through positive peer effects (students learn better when surrounded by those similar to themselves) and positive returns to scale (Reiling, 2016). Accounting for average daily attendance rate serves two major purposes. First, student absenteeism results in less classroom time and learning. Attendance helps capture studentteacher and parent-school relationship dynamics. Parents with better relationships with the school are stricter about attendance, and students with better teacher relationships are more apt to want to attend class (Klem and Connell, 2004). Besides their school environment, schools provide rich information about student background. For example, schools offer low-income students free and reduced lunch. Thus, free and reduced-price meals are a measure of neighborhood poverty. Poverty is a clear constraint on a household. The families' immediate needs reduce the amount of attention spent on education, resulting in lower scores. Racial characteristics are also considered. In the United States, historic inequities in educational attainment still persist (Lleras, 2008). I use school racial 19


THE BERKELEY ECONOMIC REVIEW

composition because they better capture dynamics of social interactions; schools are a major driver for student social groups. Moreover, racial dummy variables help capture other sources of omitted variable bias, as racial characteristics are highly related with other environmental outcomes.

20


VOLUME V

2. Data

2.1 Model After aggregating by school zone geography, I use per capita complaints to estimate noise level. I use this measure for two main reasons. First, high population areas have more callers, so standardizing by population makes rates comparable between school zones. Second, aggregating by school zone captures the incident reporting mechanism. People make complaint calls based on noise incidents near them, rather than those farther away. Aggregating by zones creates a summary statistic for overall noise for a particular area. To test the two mechanisms described in the previous section, I use the following model. Indexed for each school zone i over time t, observe that:

where SCORES represents school test scores and NOISE gives the noise rate. X represents school-level structural control variables with associated vector β2it and Y represents neighborhood-level socioeconomic controls with vector β3it. The data come from three sources: the New York City Department of Education, the 2010-2014 5-year American Community Survey, and New York City 311 noise complaints. Table 1 gives descriptive statistics at the school-zone level. The following three sections will give details on theory and data collection. 2.2 Test Scores I use test scores to measure academic performance. Since 2006, the New York State Education Department has administered standardized tests in English Language Arts (ELA) and Mathematics.

21


THE BERKELEY ECONOMIC REVIEW

22


VOLUME V

These tests aim to track progress in achieving career and post-secondary preparedness standards.2 For this study, they provide useful information about making cross-school proficiency comparisons. To account for different test forms and differences in difficulty, scaled scores are reported for ELA and Mathematics. I construct a combined z-score to give a combined-subject proficiency indicator:

where 0 corresponds to a mean score for all students in a particular grade and school, and 1 corresponds to a score a standard deviation above the mean. There has been some recent criticism of normalizing scaled test data in measuring student ability (Jacob and Rothstein, 2016) Often, just as in this case, reported scores are scaled versions of raw scores, which makes an additional score transformation arbitrary. This may mask the associated meaning of the original score. However, in this paper, the main goal is to make a cross-school comparison, not to evaluate students’ absolute knowledge. In this sense, a z-score is appropriate. For the analyses, I use the 2015 testing results for OLS and construct a five year panel dataset (2011-2015) for the fixed effects model. Since New York State has changed procedure and content over the time period, direct score comparison is not useful; a 2011 raw score is not directly comparable to a 2015 score. The z-score transformation constructed above gives information about about rank-order performance within each year. Using primary school scores is a useful measure for predicting life course events. First, prior research suggests that early age academic performance is a strong predictor for later performance. Among cognitive, social, and economic factors, early math scores were the greatest predictor for later school performance (Duncan et al., 2007). Moreover, early academic performance relate to later-life economic wellbeing. Through changes in attitudes, high-quality early education programs can boost future earnings (Heckman et al., 2013). Second, focusing on early education provides a clearer signal for understanding neighborhood charac2 See http://schools.nyc.gov/Accountability/data/TestResults/ELAandMathTestResults for more information

23


THE BERKELEY ECONOMIC REVIEW

teristics compared to studying high school data. While older students are closer to employment age, test scores are partially a function of cumulative knowledge. Results not only reflect current living and educational environment, but also student background. In short, focusing on primary education captures neighborhood information and acts as a predictor for future academic outcomes.

Figure 1 shows standardized test scores mapped by school zone. Unsurprisingly, geography sems to matter. In particular, scores in Lower Manhattan and Northeastern Queens appear to be higher than other areas in the city. Upon inspection, bordering school zones. The score grouping suggests that neighborhood general demographic characteristics play some role in academic performance. These variables require further investigation.

24


VOLUME V

2.3 Noise complaints I use New York City's 311 complaints data to measure noise disturbances. Made publically available on New York City's Open Data portal, the dataset begins in 2010 and contains 14,128,288 entries as of 12/05/2016.3 The city updates the dataset every day. Each complaint entry contains

detailed information about type, general description, coordinates, location description, date, and time. Since the analysis focuses on school year noise complaints, I only include complaints with the term “noise” in their description. The City of New York places its own categorization for each complaint type, so all complaints in the “noise” category begin with the word “Noise:” followed by a subcategory.

3

See https://data.cityofnewyork.us/Social-Services/311-Service-Requests/fvrb-kbbt

25


THE BERKELEY ECONOMIC REVIEW

To create the noise measure, I aggregate by elementary school zone geographies. The school zoning data are available as a GIS shapefile.4 While data for secondary-age students are also available, primary school data are advantageous for a couple reasons. First, at the elementary level, New York City Public Schools gives preference to students living within each zone. These students are more likely to attend their zoned school, and the scores are therefore more reflective of the neighborhood. In contrast, for secondary school, students have the option to apply for �zone choice,� which allows students to attend out-of-zone schools. As a result, middle and high schools have less geographic homogeneity and are not as representative of neighborhood characteristics

(New York City Department of Education, 2016). Second, there are more school zones at the elementary school level than at either middle or high school levels. This allows for a finer granularity analysis. 4

See https://data.cityofnewyork.us/Education/2014-2015-School-Zones/mshx-yvwq

26


VOLUME V

However, using number of complaints per school zone is not useful by itself. To make complaints comparable, I divide by the total population within each school zone. The rate is reported as noise complaints per 1000 people. I apply an alternate noise complaints schemes in the appendix. In one scheme, I weight noise complaints by intensity using complaint follow-up information (i.e. if complaints were forwarded to the New York Police Department). Only those that received police action were counted. However, under this specification, noise complaint rate is not significant–likely since noise need not be illegal to affect behavior. The primary advantage to using complaints data is that these capture bothersome noise events. Complaint-reporting is an active process. Individuals in a neighborhood report when noise is a nuisance, not when it is unnoticed. The noise complaints measure, however, is not designed to capture all noise. While responsive to individual noise events, the measure is not designed to capture lowdecibel persistent noise. For example, the effects of a noisy air-conditioning units may not be completely captured in the data. However, these events are often not the ones that are bothersome. Over time, people largely habituate to these sound types (Halonen et al., 2012; Waye et al., 2003). Figures 2 and 3 give insight to the complaints’ spatial structure. Figure 2 shows complaint rate by elementary school zone. Upper Manhattan appears to have a higher rate compared to the rest of the city. Similarly, Staten Island and outer Queens appear to have lower rates. Figure 3 gives the average street-level complaint rate by borough. Manhattan has the highest average complaint rate. The Bronx, Brooklyn, and Queens are less than half of Manhattan’s. These two descriptives suggest that controlling for spatial factors is important–complaint location matters.

27


THE BERKELEY ECONOMIC REVIEW

2.4 School and neighborhood controls

In addition to test performance data, the New York City Department of Education provides rich school-level variables. These school structural variables iinclude grade size, racial composition, average attendance, percent with learning disabilities, and student-teacher ratio. Figures 4 and 5 show race and poverty by school zone. Unsurprisingly, similar neighborhoods border each other. Clumping is especially clear in the race plot, where southern Staten Island is mostly white (the purple region) and northeastern Queens is mostly Asian.

28


VOLUME V

In addition to using data directly from the New York City Department of Education, I use additional demographic information. From schools’ boundary shape files, I impute neighborhood characteristics from the 2010-2014 block grouplevel American Community Survey. These include income, age, and neighborhood characteristic variables. They help provide further neighborhood demographic information. For details, see table 1. Each measure is aggregated by school zone.

29


THE BERKELEY ECONOMIC REVIEW

3. Results

For the analyses, I use four main methods: Ordinary least squares, yearly fixed effects, propensity score matching, and spatial regression. Ordinary least squares on all observations allows for control of some observable extraneous factors. Holding neighborhood and school-level variation constant, the method helps isolate the relationship between noise complaints and score. However, the model implicitly assumes that the covariates are not correlated with the error term. From the previous discussion on geographic descriptives, this is likely not the case. Families select into certain neighborhoods and geographies are not independent. The two other techniques help address concerns about omitted variable bias. The yearly fixed effect models help account for time invariant extraneous factors. This specification gives a conservative estimate. Compared to ordinary least squares, fixed effects has less statistical power, and is more prone to type II error. The method, however, is advantageous in that it accounts for unobservable time-invariant extraneous factors. Hence, not all controls need to be specified. Propensity-score matching helps account for selection bias by restricting observations to observationally similar schools. Families do not randomly choose their location, so treating school zones as independent units can be incomplete. Implementation consists of several steps. First, I treat complaint rate as a dummy variable by defining values in the upper third quantile as “high noise.” Second, using R’s “matchit” package, I create propensity scores and match the two groups according to student-teacher ratio, percent on free/reduced lunch, age, percent with at least some college, sex, crime rate, and school attendance rate. Third, after implementing the nearest observations matching algorithm, I rerun the full OLS specification. Since some observations are unmatched, they are discarded in the propensity-score sample. In effect, the method removes observably dissimilar schools, which helps account for omitted variable bias. Ultimately, this makes estimation more conservative than OLS. In total, 434 out the total 551 observations were matched. The control group (low noise) had 217 observations compared to 217 in the treatment group (high noise).

30


VOLUME V

Finally, Spatial regression models help account for spatial autocorrelation. Noise complaint clustering in the spatial descriptives make this scenario likely. Beyond the empirics, noise complaints near neighborhood borders can affect multiple regions; sound can travel across neighborhood boundaries. To control for these, I use two specifications. First, the spatial-error model helps account for spatial autocorrelation in the regression error term–this happens when the independent variables possess a spatial structure. Second, I use a spatial-lag model. In addition to controlling for regressor spatial structure, this specification accounts for spatially correlation in the dependent variable (i.e. test scores). To setup the model, I first take the centroid of each school zone, which helps orient the data in space. Then, using R’s ”spdep” package, I create a binary spatial weight matrix for the two nearest neighbors. I then use Moran’s I to formally test for spatial dependence. Finally, to test whether spatial regression is appropriate, Lagrange multiplier diagnostics for spatial dependence are used. The results support the theory; under normal and robust specifications, the spatial lag model is significant. As described in the literature section, confounders influence the relationship between noise and academic performance. All specifications rely on control variables. They fall into two primary categories: neighborhood-level and schoollevel characteristics. At the neighborhood level, the percentage of individuals who are 26 and older with some college helps control for educational attainment. The unemployment rate controls for economic variability. I include median age to account for age structure; since they spend more time at home, neighborhoods with a larger proportion of older residents are potentially more likely to file noise complaints. In a similar vein, I include population density to control for complaint reporting–the descriptives in the previous section suggested that population dense areas had a higher complaint rate. At the school level, student teacher ratio, cohort size, and daily attendance rate account for some structural differences across schools. Including percentage with learning disabilities, percentage in assisted lunch, and racial composition dummy variables helps control for differences in student composition. For the race variables, the analysis uses the "schools' most prevalent race" (e.g. "Asian" would be the most prevalent rance in a school with 50

31


THE BERKELEY ECONOMIC REVIEW

students and 20 white students). Compared to using raw percentages, these are a better reflection of the school/neighborhood environment- they give insight to general social dynamics.

3.1 Noise during sleep hours Tables 2 and 3 give results for the sleeping hours hypothesis. Results 1-4 in Table 2 show the output for the OLS and fixed effects models. Table 3 shows the output for propensity score matching and spatial regression models. Both show that even after neighborhood and school controls, noise complaint rate is still significant. This holds at ι = .001 in the OLS and fixed effects model, and holds at ι = .05 for propensity score matching and spatial lag. Moreover, the magnitude of the coefficient is roughly the same. All else equal, an increase in one noise complaint per 1000 people causes approximately a .07-.10 standard deviation decrease in composite score. Equivalently, a one standard deviation increase in noise complaints per 1000 causes approximately a .06 standard deviation in composite score. Model 1 gives OLS results without controls. The low R2 value (.04) show that much of the variation in scores is explained by other factors. More control variables are needed. Model 2 includes neighborhood-level and school-level controls. These variables help capture some of noise complaint rate’s effect. The free/reduced lunch variable shows the relevance of economic factors. All else equal, high-poverty schools (where the proportion of students on free/reduced lunch is greater than 75%) score .82 standard deviations lower than the control group (schools with less than 25% of students on free/reduced lunch). The negative coefficients with felony rate and unemployment rate tell a similar story, namely that high-poverty areas have lower scores. Consistent with prior literature, racial characteristics appear to matter. Compared to schools with mostly white students, schools with mostly Asian students do slightly better. Schools with mostly black and Hispanic students have lower scores. Also, attendance rate appears to be significant. A 1 % increase in 32


VOLUME V

attendance is related to a .081 standard deviation increase in standardized score. The next two models display the results for yearly fixed effects. Model 3 gives the results without noise complaints, and model 4 gives the full specification. Since neighborhood characteristics come from the 5-year ACS, these variables are time invariant and therefore are not included in the analyses. I include structural school-level characteristics; these can change in the short run.

Despite being a more conservative estimation technique, the noise complaints term is still highly significant (see Model 4). Furthermore, the results for the school-level control variables are similar to the OLS specifications, indicating robustness. Consistent with the previous results, student-teacher ratio has a positive effect on composite score. The coefficient’s magnitude (.02) is similar to the OLS specifications, and percent with learning disabilities results in lower scores. Also, notice that the magnitude of the learning disabilities, school enrollment, 33


THE BERKELEY ECONOMIC REVIEW

and student teacher ratio coefficients are the same with and without noise complaints. However, the racial control variables are slightly different. In particular, after introducing noise complaints, the coefficients for mostly Hispanic and black schools slightly decrease (by .03 and .02 respectively). The decrease implies that adding noise complaints helps capture and account for some omitted variable bias. In Table 3, the results for street-level noise complaint rate are similar. The first two specifications are the result of two spatial regression models, and the last model uses propensity score matching. Spatial error is helpful in addressing spatial correlation in the regression error term. In this case, the measurement unit is individual school zones. Using Moranâ&#x20AC;&#x2122;s I, I find that characteristics across observations are spatially correlated. Spatial lag helps to address spatial dependence in the dependent variable. In models 1 and 2, I also include neighborhood and school-level control variables. Qualitatively and quantitatively, the results are similar. The regression coefficients are within -.05 to -.07 street-level noise complaints per thousand. The control variables are also similar. Schools with higher proportion of students on free or reduced lunch do significantly worse than the control group (< 25%). Similarly, schools with mostly black and Hispanic students do significantly worse than the white control group. Model 3 uses propensity score matching. The method helps controls for observable selfselection bias: people are not randomly distributed across neighborhoods. Similar to the spatial regression models, the regression coefficients are qualitatively similar. The coefficient for the noise complaints term is -.07. Again, the neighborhood-level and school-level control variables are significant. Even controlling for observable self-selection, the results are robust. In sum, the spatial regression specifications and propensity score matching results show that the yearly fixed effects specification is highly robust, even considering self-selection and spatial structure.

34


VOLUME V

35


THE BERKELEY ECONOMIC REVIEW

4. Discussion

In this paper, I use New York Cityâ&#x20AC;&#x2122;s 311 complaints dataset to proxy for neighborhood noise disturbances. In conjunction with data from New York Cityâ&#x20AC;&#x2122;s Department of Education and the American Community Survey, the findings suggest that environmental noise is inf luential in academic performance for elementary school students. During bedtime hours, external noise disrupts normal sleeping patterns for some children, resulting in reduced learning ability. The results are consistent cross-sectionally with neighborhood and school control variables. They also hold under a fixed effects model with school-level controls. In contrast, noise is not significantly related to composite score during school hours. Due to self-selection bias and habituation, these results are likely understated. Since 311 complaints are self-reported, self-selection bias could be a relevant factor. Communities with older, wealthier residents likely over-report. Conversely, communities with younger, poorer residents under-report. These two factors combine to reduce variation in the noise complaint measure. Similarly, complaints data does not take into account habituation. Residents in noisy areas may become accustomed to noise events, and may not see value in reporting; such efforts could be seen as being futile. In short, despite being an imperfect imperfect proxy for noise, the results are still robust. Robust results at the aggregate level are a start. While this analysis uses school-zones as the unit of study, further investigation at the individual level would help with generalizing conclusions. Beyond its humanistic value, providing quality education is a pathway for reducing economic inequality in the United States. Accounting for environmental factors is important for countering geographic disparities. Results suggest that creating a suitable living environment is consistent with this goal, especially during formative years.

36


VOLUME V

5. References

Barrett, N. and E. F. Toma (2013). Reward or punishment? Class size and teacher quality. Economics of Education Review 35, 41–52. Basner, M., U. Muller, and E.-M. Elmenhorst (2011). Single and combined effects of air, road, and rail traffic noise on sleep and recuperation. Sleep 34, 11–23. Black, S. E. and P. J. Devereux (2010). Recent developments in intergenerational mobility. NBER Working Paper No. 15889, 1–91. Chua, E. C.-P., S.-C. Yeo, I. T.-G. Lee, L.-C. Tan, P. Lau, S. Cai, X. Zhang, K. Puvanendran, and J. J. Gooley (2014). Sustained attention performance during sleep deprivation associates with instability in behavior and physiologic measures at baseline. Sleep 37(1), 27–39. Cobb-Clark, D. A., M. Sinning, and S. Stillman (2012). Migrant Youths’ Educational Achievement: The Role of Institutions. The ANNALS of the American Academy of Political and Social Science 643(1), 18–45. Dewald, J. F., A. M. Meijer, F. J. Oort, G. A. Kerkhof, and S. M. Bogels (2010). The influence of sleep quality, sleep duration and sleepiness on school performance in children and adolescents: A meta-analytic review. Sleep Medicine Reviews 14(3), 179–189. Duncan, G. J., A. Claessens, A. C. Huston, L. S. Paganini, M. Engel, H. Sexton, C. J. Dowsett, K. Magnuson, P. Klebanov, L. Feinstein, J. Brooks-Gunn, and K. Duckworth (2007). Supplemental Material for School Readiness and Later Achievement. Developmental Psychology 43(6).109–124.

37


THE BERKELEY ECONOMIC REVIEW

Farrington, D. P., J. W. Coid, and J. Murray (2009). Family factors in the intergenerational transmission of offending. Criminal Behaviour and Mental Health 19(2), 109–124. Geller, A., I. Garfinkel, C. E. Cooper, and R. B. Mincy (2009). Parental Incarceration and Child Well-Being: Implications for Urban Families. 90(5). Gibson, E. S., a. C. P. Powles, L. Thabane, S. O’Brien, D. S. Molnar, N. Trajanovic, R. Ogilvie, C. Shapiro, M. Yan, and L. Chilcott-Tanser (2006). ’Sleepiness’ is serious in adolescence: Two surveys of 3235 Canadian students. BMC public health 6, 116. Graue, M. and D. Oen (2008). You Just Feed Them With a Long-Handled Spoon Families Evaluate Their Experiences in a Class Size Reduction Reform. Educational Policy 23(5), 685. Gruber, R., G. Somerville, P. Enros, S. Paquin, M. Kestler, and E. Gillies-Poitras (2014). Sleep efficiency (but not sleep duration) of healthy school-age children is associated with grades in math and languages. Sleep Medicine 15(12), 1517–1525. Halonen, J. I., J. Vahtera, S. Stansfeld, T. Yli-tuomi, P. Salo, I. Halonen, P. Salo, P. Jaana, J. Pentti, and M. Kivimaki (2012). Associations between Nighttime Traffic Noise and Sleep: The Finnish Public Sector Study. The National Institute of Environmental Health Sciences 120(10), 1391–1396. Heckman, J., R. Pinto, and P. Savelyev (2013). Understanding the Mechanisms Through Which an Influential Early Childhood Program Boosted Adult Outcomes. American Economic Review 103(6), 2052–2086. Hwang, J. and R. J. Sampson (2014). Divergent Pathways of Gentrification: Racial Inequality and the Social Order of Renewal in Chicago Neighborhoods. American Sociological Review 79(4), 726–751.

38


VOLUME V

Jacob, B. and J. Rothstein (2016). The Measurement of Student Ability in Modern Assessment Systems. 30(3), 85–108. Jeynes, W. H. (2005). A Meta-Analysis of the Relation of Parental Involvement to Urban Elementary School Student Academic Achievement. Urban Education 40(3), 237–269. Klem, A. M. and J. P. Connell (2004). Linking Teacher Support to Student Engagement and Achievement. 74(7), 262–274. Lleras, C. (2008). Race, Racial Concentration, and the Dynamics of Educational Inequality Across Urban and Suburban Schools. American Educational Research Journal 45(4), 886–912. Mercer, P. W., S. L. Merritt, and J. M. Cowell (1998). Differences in reported sleep need among adolescents. Journal of Adolescent Health 23(5), 259–263. Nesmith, A. and E. Ruhland (2008). Children of incarcerated parents: Challenges and resiliency, in their own words. Children and Youth Services Review 30(10), 1119– 1130. New York City Department of Education (2016). Achieve NYC A Complete Guide to New York City Public Schools PK-12. Perron, S., L.-F. Tetreault, N. King, C. Plante, and A. Smargiassi (2012). Review of the effect of aircraft noise on sleep disturbance in adults. Noise & health 14(57), 58–67. Reiling, R. B. (2016). Does Size Matter? Educational Attainment and Cohort Size. Journal of Urban Economics 94, 73–89.

39


THE BERKELEY ECONOMIC REVIEW

Reynolds, A. J., S.-R. Ou, and J. W. Topitzes (2004). Paths of Effects of Early Childhood Intervention on Educational Attainment and Delinquency: A Confirmatory Analysis of the Chicago Child-Parent Centers. Child Development 75(5), 1299–1328. Rud, I., C. Van Klaveren, W. Groot, and H. Maassen van den Brink (2014). The externalities of crime: The effect of criminal involvement of parents on the educational attainment of their children. Economics of Education Review 38, 89– 103. Shield, B. M. and J. E. Dockrell (2008). The effects of environmental and classroom noise on the academic attainments of primary school children. J Acoust Soc Am 123(1), 133–144. Stickgold, R. and M. P. Walker (2007). Sleep-dependent memory consolidation and reconsolidation. Sleep Medicine 8(4), 331–343. Waye, K. P., J. Bengtsson, A. Agge, and M. Bjorkman (2003). A Descriptive CrossSectional Study ¨ of Annoyance from Low Frequency Noise Installations in an Urban Environment. (1999), 35–46.

40


VOLUME V

A. Noise complaints during school hours In contrast to the results from noise during sleeping hours, complaints during school hours tell a different story. Tables 4 and 5 show the results from the OLS, propensity score matching, and fixed effects specifications. Noise is not significant under any of the specifications. The coefficients for the control variables are qualitatively similar to those of sleep and school specification, which gives evidence of robustness. Shown in the free-reduced lunch dummy variables, poverty still plays a role in testing. Several factors can explain the result. First, consistent with the London elementary school sound study, internal noise could be more influential than external noise (Shield and Dockrell, 2008). Being adjacent to a noisy classroom could have a larger effect on teaching quality than outside traffic noise. Second, noise and teaching could be endogenous. In response to frequent noise events, instructor can adapt teaching to be less dependent on auditory cues, reducing the negative effects of noise. While external noise could play a role, other factors may be more relevant.

B. Noise complaints requiring police action

In this section, I use â&#x20AC;&#x153;police actionâ&#x20AC;? as a measure for complaint intensity. In prior analysis, I treat all complaints the same, regardless of outcome. In addition to providing information on complaint location and type, data on follow-up action is given. In general, complaints can be grouped into two categories; they are either ignored or referred to the police. I see whether complaints that are referred to the police correspond to more intense noise violations. I find that these reported complaints have little influence on school test scores. Using ordinary least squares and yearly fixed effects, notice that the coefficientsâ&#x20AC;&#x2122; magnitude is small (-.005 to -.014) relative to the standard error, suggesting a weak relationship. This is likely the case because complaints do not need to be illegal to be influential.

41


THE BERKELEY ECONOMIC REVIEW

More concretely, consider general road traffic noise. Cars passing through a street can be a nuisance living in the area. However, there is little law enforcement can do to reduce the general noise level. This type of annoyance is likely not captured using this measure.

42


VOLUME V

43


THE BERKELEY ECONOMIC REVIEW

C. Results by test subject This section gives ordinary least squares and fixed effects results by testing subject. Each year, the New York State Department of Education administers an ELA (English Language Arts) and Mathematics test for grades 3-8. For the analyses in the paper’s main body, I use a combined standardized score (called “Standardized composite”) to measure academic performance. In effect, I combine these two scores together. To see if results differ based on subject, I use identical ordinary least square and fixed effect models as in the main body text, with the exception of different dependent variables (I use standardized scores by subject). In particular, in Table 7, Models 1 and 3 give mathematics while Models 2 and 4 give ELA results. Examining the OLS specifications (models 1 and 2), the coefficients for noise complaint rate are roughly the same (a difference of .01). Additionally, nonrace control variables have roughly the same coefficient magnitudes. However, the results differ slightly for the racial dummy variables. Compared to the reference group (mostly white schools), mostly Asian schools do better on on both tests, but there is a larger difference for mathematics (.25 versus .09). In contrast, black and Hispanic schools do worse on both tests. Black schools have lower scores for the mathematics test than on ELA (-.65 vs. -.53), while Hispanic schools have similar scores for both exams (-.39 vs. -.40). Similar results hold for the fixed effects specifications (models 3 and 4). The coefficients for noise complaint rate are both .1. Again, non-race control variables have roughly the same coefficient magnitudes, while the race dummy variables differ in the same direction as in the OLS specification.

44


VOLUME V

45


THE BERKELEY ECONOMIC REVIEW


VOLUME V

Singapore's Primary 1 Registration: Redesigning the School Choice Mechanism

by Ruru Hoong (Stanford University)


THE BERKELEY ECONOMIC REVIEW

Singapore's Primary 1 Registration: Redesigning the School Choice Mechanism by Ruru Hoong Stanford University

48


VOLUME V

Abstract

A prominent issue in school choice is the design of a student assignment mechanism. Prior mechanism design approaches to resolving school choice problems have been met with much success in implementation; a notable example is the reform of the Boston Mechanism in Boston Public Schools in 2005. This paper seeks to build on previous literature on school choice and present an altered DA (deferred acceptance) algorithm to suit the Singapore context. I show that this altered mechanism provides a practical solution to some of the critical issues faced in the current Singapore P1 Registration allocation system, resulting in a (i) strategyproof, (ii) fair, and (iii) nonwasteful algorithm that is preferable to the existing one.

49


THE BERKELEY ECONOMIC REVIEW

I. Introduction Every year, over 40,000 Singaporean prospective Primary 1 students and their parents undergo the undeniably stressful Singapore Primary 1 Registration Exercise. The current exercise has been in place since the 1980s; over time, more and more rules have been added to its structure, resulting in an increasing convoluted and long-drawn system. This paper seeks to propose a centralised school choice system that incorporates parent preferences and a strategy-proof mechanism that will not only provide a more stable allocation of students to schools, but also significantly reduce the time and resources spent on the registration exercise. Examples of successful school allocation mechanism reforms proliferate; in 2005, the Boston School Committee voted to reform the existing Boston student assignment mechanism with a deferred-algorithm mechanism based on the GaleShapley algorithm that is strategy-proof. Two algorithms were proposed: the deferred acceptance (DA) mechanism as well as a top-trading cycles mechanism. This paper seeks to apply the findings of the Boston school choice mechanism to the Singaporean context, and modify the proposed DA algorithm to suit the Singapore Primary 1 (Grade 1) registration exercise. The current Singapore mechanism, like the Boston mechanism, is a priority matching mechanism. [1] However, unlike the Boston mechanism, where students submit a list of preferences and are matched to schools according to a strict priority order, the Singapore mechanism is carried out through separate demarcated phases, and the onus is on the parents to submit an application for their child in their desired school and corresponding phase(s). Furthermore, in the Boston mechanism, all students undergo the first step, as all applicants have a first choice by default. Contrastingly, under the Singapore mechanism, not all students qualify for all phases, and only students that submit applications to a specific school will be considered for the phase in question.

50


VOLUME V

Idiosyncrasies of the Singapore Mechanism

Taking into consideration the complexities of the Singapore mechanism as described above, it would be instructive to outline a few of its unique features: 1. Multi-phase Eligibility. One consequence of the phased system specified by the Ministry is that a student may be eligible for multiple slots under different priorities in any one school. 2. Minimum Quotas. In 2014, a rule was introduced requiring all primary schools to set aside 40 places for children in the later stages. The 40 places will be split equally between children registering in Phases 2B and 2C.1 Given the idiosyncrasies outlined above, a simple student-proposing deferred acceptance algorithm or top-trading cycle will not be sufficient to account for the priority-orders, multi-phase eligibility, and the quota capacities of the Singapore school choice system whilst satisfying other important desiderata. This paper therefore proposes an alternative mechanism, and examines its suitability for application in Singaporeâ&#x20AC;&#x2122;s P1 registration process.

Desiderata

In the process of designing a mechanism for the Singapore context, several desiderata will be kept in mind: 1. Timeliness. The current system is a long-drawn process that takes place annually from July - November, and each school handles its own balloting and admissions despite standardised rules specified by the Ministry. Furthermore, parents are required to take time off work to register their child in person at each individual school (with the exception of a few 1 It is important to note that these are soft quotas. If, for instance, only 35 students applied for Phase 2B and 2C for a specific school s, there would be no need to force 5 more students to attend school s in order to fulfill the quota.

51


THE BERKELEY ECONOMIC REVIEW

phases), resulting in significant time wastage. With the introduction of a centralised matching mechanism, parents will only need to submit their preferences once, eliminating the need for in-person registration. This will significantly reduce the time needed to be set aside for the registration exercise and save a lot of time for parents, schools, and also the Ministry.

2. Transparency. The complexity of the current Singapore mechanism has made the process opaque and sometimes indecipherable to the average Singaporean parent. In designing the mechanism, an important consideration would be the transparency of its algorithmic description and how comprehensible it is to the average Singaporean parent. 3. Strategyproof-ness. A mechanism is considered strategyproof if no student ever has the incentive to misreport her preferences, no matter what the other students report. The current structure of the Singapore mechanism gives parents the strategic incentive to misrepresent their true preferences. As with the Boston mechanism, it can be costly to list a first- choice that you do not succeed in getting because once other students are assigned to their places, they cannot be displaced even by a student with a higher priority. [1] Unsurprisingly, there are entire websites dedicated to strategising the registration exercise - one such website, KiasuParents.com, [7] even goes so far as to compute oversubscription risks for each school. Compounded with the lack of transparency, this predisposition for “gaming” the system heavily disadvantages families that do not strategise, or that strategise inadequately. 4. Fairness. Fairness (also called “stability” or “no justified envy”) requires that there is no student u that prefers school s to her assigned school when some other student with lower priority is assigned to s. Inadequate strategising inherent to the Singapore mechanism has rendered fairness unlikely; it is very common for students to apply to a less desired school in order to avoid having a ballot for splace in their more preferred school, 52


VOLUME V

only to eventually ency another student that gets in to their preferred school with a lower priority.

5. Nonwastefulness. A matching is considered nonwasteful if for any school s that has empty seats, no student u would prefer school s to her existing assignment. Given that only a very small handful of school have remaining empty seats at the end of the exercise (and these empty seats are usually then filled up by non-Singaporeans), it would not be unreasonable to state that the current Singapore mechanism is nonwasteful. There is the possibility, however, of parents strategising inadequately and applying to a less preferred school even if their preferred school has remaining capacity. Parents could misjudge the demand for a school that will have empty seats at the end of the registration cycle, and apply to a less preferred school to be â&#x20AC;&#x153;safe.â&#x20AC;? Another concern that arises with the Sinapore mechanism is its current priority hierarchy: the Singaporean public has been embroiled in much debate over whether enrolment criteria should be altered.[10] This paper only seeks to propose a student assignment mechanism under the current priority hierarchy specified by the Ministry of Education; it does not seek to assert normative judgments on the existing priority system.2 For instance, one question commonly raised by the public asks whether local schools should give priority to Primary 1 students who live nearby, despite indications that this perpetuates existing economic inequality and implicitly prioritises students from wealthy backgrounds. Due to the complications and vast scope of such an examination, this paper will refrain from attempting to resolve such overarching ethical questions. Instead, I focus on designing a mechanism that is simultaneously strategy-proof, fair, and nonwasteful without altering the current hierarchy of priorities specified by the Ministry.

Granted, strategyproofness, fairness, and nonwastefulness can be considered normative axioms as well. However, ethical questions that arise from the determination of priority hierarchy require other (arguably more contentious) normative criteria; thus, I will treat priority hierarchy as requiring a different class of normative judgment, and allow school priorities to be exogenously subscribed in accordance to the current priority system. 2

53


THE BERKELEY ECONOMIC REVIEW

2. Related Literature School choice is a well-trodden area of research; the Boston mechanism in particular has been studied extensively since Abdulkadiroglu and Sonmez’s [3] seminal paper on school choice mechanisms. Since then, many variations on the standard Gale-Shapley deferred acceptance algorithm have been explored in the context of school matching. Adjustments for multi-phase eligibility, and accomodations for minimum quotas certainly aren’t untrodden ground: even the original Boston Public Schools (BPS) implementation of DA had a walk-zone priority feature that bears resemblance to the multi-phase eligibility I will accommodate for in our Singapore model. Interestingly, the decision of how to deal with the “multi-phase eligibility” of students that qualified for both the reserved walk-zone priority slots (50% of total slots) as well as the remaining open slots was viewed as an inconsequential detail and left to BPS software support. Research on how the precedence order, i.e. the order in which these different types of seats were filled by students, later emerged in Dur et al.’s [4] paper on the demise of walk zones in Boson. Minimum quotas have been a somewhat recent extension to standard matching models, and the theory of matching has been more extensively developed for markets with maximum quotas. Most research into maximum quotas has been in context of affirmative action policies; in their seminal paper, Abdulkadiroglu and Sonmez extend their analysis of the DA algorithm to accommodate a simple affirmative action policy with type-specific maximum quotas. More recently, Kojima [8] investigates the welfare effects of affirmative action policies with quotas for majority students. Hafalir et al. [6] then offer a different interpretation of affirmative action policies based on minority reserves, where schools give higher priority to minority studensts up to the point that the minorities fill the reserves. Westkamp [11] then studies the German university admissions system in a “matching with complex constraints” problem, accommodating transferable quotas on different subpopulations. This paper beas closest comparison to Fragiadakis et al. [5]. who introduce two new classes of strategyproof mechanisms that allow for minimum quotas as an 54


VOLUME V

explicit input. Their first new mechanism, extended-seat deferred acceptance (ESDA), divides seats into two classes: “regular” seats equal to its minimum quota, and “extended” seats equal to the difference between the minimum and maximum quota. According to their individual preferences, students first apply to “regular” seats, and then to the “extended” seats. The assignment procedure proposed in this paper differs from the ESDA in that it does not deal with hard minimum quotas; that is, there is no need to fill up all the “regular” seats. Therefore, instead of filling “regular” seats first, the proposed algorithm fills up all the “extended” seats first, treating the minimum quota as an “inverse” maximum quota. I present a modified DA algorithm that combines both multi-phased eligibility and soft minimum quotas, and that is simultaneously strategyproof, fair, and nonwasteful. To the best of my knowledge, this paper is the first study of the Singapore P1 Registration Exercise from a mechanism design approach. The proposed alternative holds practical relevance for the reform of the current Singapore mechanism; therefore, it is the hope that this study will spark a reexamination of the existing assignment system and encourage consideration of a new assignment mechanism.

55


THE BERKELEY ECONOMIC REVIEW

3. The Current Singapore Mechanism In Singapore, students are allocated seats at local schools through a complex process laid out by the Ministry of Education. The details of the order of priority are specified on the Ministryâ&#x20AC;&#x2122;s website, and comprehensive videos have been released to aid parents in understanding the registration process. [9] The exercise is conducted in consecutive phases, with buffer time of about a week between each phase for parents to hear back about their application result and to decide whether they need to submit an application for the next phase. The chronology of the phases is summarised by the diagram on the following page. For phases that are oversubscribed (i.e. the number of applications to a school exceeds the number of vacancies available in that particular phase), balloting will occur. Within any oversubscribed phase (barring Phase 3), Singaporean Citizens are given absolute priority over Singapore Permanent Residents, before home-school distance is considered. A random lottery number is then used to break the ties in each category and determine a strict priority order, as with the Boston Mechanism. [2] Consider the following example for illustration:

56


VOLUME V

In the example above, the total number of SC applications exceeds the number of vacancies. The vacancies are allocated based on home-school distance in the following order of priority: (1) within 1 km, (2) between 1 km and 2 km and (3) outside 2 km. The 38 SC children living within 1 km from the school are admitted first, followed by the 10 SC children living between 1 km and 2 km from the school. After the 48 vacancies are taken up, the remaining 2 vacancies are balloted by random lottery amongst the 8 SC children living more than 2 km from the school.

57


THE BERKELEY ECONOMIC REVIEW

The Singapore mechanism assigns students as follows:

Step 1. For each school, consider the students that have applied to it through the first phase and assign places to these students in priority order until either no places remain (setting aside reserved seats for Phase 2B and 2C) or all students have been placed.

Step 2. For each school, consider the students that have applied to i through the kth phase and assign places to these students in priority order until either no places remain (setting aside reserved seats for Phase 2B and 2C, if these phases have not already occurred) or all students have been placed. The procedure terminates when all students are assigned a seat, or when all places at schools are allocated. Admission is not guaranteed for Phase 3 applicants as there are limited vacancies remaining for non-SC or non-PR children.

58


VOLUME V

4. The Model There is a finite set U of students and a finite set S of schools. Each student u has a strict preference relation Pu over S. For each school s, there are two types of spaces students can qualify for: Phase A (encompassing Phases 1 through to 2A(2)), and Phase B (Phase 2B onwards).3 The demarcation of these phases is to account for the minimum quota of 40 students that must be reserved for Phase B.4 In practice, there would be three types of spaces – Phase 2B and 2C would be considered two separate categories, each with 20 reserved places – but for simplicity’s sake, our model will only have two categories. This will be sufficient for our illustrative purposes. For each school , we use Ss to represent the total capacity of the school. A matching µ : U -> S is a function which assigns a school to each student such that no school s is assigned to more than Ss students. Let µu denote the assignment of student u, and let µs denote the set of students assigned to school s. Define Ik,s as the set of students consisting of the applicants applying to school s in round k that qualify for Phase A, and Jk,s as the set of students consisting of the applicants applying to school s in round k that qualify for Phase B. For a school , Phase A slots have a linear priority order πi,s over students in U, and Phase B slots have a linear priority order πj,s over students in U. Given a school , a list of priority orders πi,s, the maximum Phase A capacity Ss - 40, and a set of students Ik,s , the choice of school s from the set of students Jk,s , denoted by Cs(Jk,s,x), is obtained as follows: the top x students in Jk,s are tentatively accepted under order πj,s. Based on preferences, priorities, and school capacities, student assignments are determined with the following algorithm:

For the sake of simplicity, we omit Phase 3 from our considerations as non-SC and non-PR children are not guaranteed admission. 4 All students are eligible for Phase B. 3

59


THE BERKELEY ECONOMIC REVIEW

Step 1. -Each student u applies to her top choice of school. Each school s considers the set I1,s of students consisting of the applicants qualifying for Phase A and the set J1,s of students consisting of the applicants qualifying for Phase B. Not that it is possible for a student to be both I1,s and J1,s. Consider the two following cases: Case 1: Phase A is oversubscribed, meaning that I1,s > Ss - 4o. Each school s tentatively holds the applicants in Cs(I1,s,Ss-40) and Cs(J1,s,40) and rejects the rest. Note that applicants rejected from I1,s may be eligi ble to be tentatively accepted within Cs(J2,s,40). Case2: Phase A is undersubscribed, meaning I1,s < Ss - 40. Each school s tentatively holds all qualifying applicants I1,s and Cs(J1,s,Ss-I1,s) and rejects the rest. If an applicant can be tentatively accepted in both Cs(I1,s,Ss-40) and Cs(J1,s,x), she is included in Cs(I1,s,Ss-40); in other words, she is considered to be accepted under Phase A and excluded from Cs(J1,s,40). Step k. -Each student u rejected in step k - 1 applies to her most preferred school that has not yet rejected her. If a student has been rejected in step k - 1 after being tentatively accepted in Phase A during som previous step (i.e. s?Cs(Ia,s) | a < k - 1), consider their eligibility for Phase B at the same school before proposing to their next most preferred school. Each school s considers the set Ik,s consisting of the new applicants to s and the students held by s at the end of step k - 1 under Phase A, and the set Jk,s consisting under Phase B. Note that it is possible for a student to be in both Ik,s and Jk,s. Consider the following two cases: Case 1: Phase A is oversubscribed, meaning that Ik,s > Ss - 4o. Each school s tentatively holds the applicants in Cs(Ik,s,Ss-40) and Cs(Jk,s,40) and rejects the rest. Note that applicants rejected from Ik,s may be eligi ble to be tentatively accepted within Cs(Jk+1,s,40). 60


VOLUME V

Case2: Phase A is undersubscribed, meaning Ik,s < Ss - 40. Each school s tentatively holds all qualifying applicants Ik,s and Cs(Jk,s,Ss-Ik,s) and rejects the rest. The procedure terminates when all students are assigned a seat, or when all places at school are allocated.

61


THE BERKELEY ECONOMIC REVIEW

5. Evaluation of the Two Mechanisms Unlike the Boston mechanism, the current Singapore system does not collect ranked preferences from students, so preferences are implied through the studentsâ&#x20AC;&#x2122; actual applications to schools. Furthermore, given the distorting incentives under Singaporeâ&#x20AC;&#x2122;s current school choice mechanism, the obtainable data that rely on implied preferences cannot be used to assess the effectiveness of the mechanism. As such, two situations that illustrate the differences between the current Singapore mechanism and the new altered DA mechanism will be presented. In doing so, this paper hopes to highlight the preferability of the new system as measured against the desiderata established in the introduction.

62


VOLUME V

Consider Phase A applications. Students u1, u2, u3, u5, u6, and u7 all apply to their top choice of school as they each qualify for Phase A in their most preferred schools. u4 does not qualify for Phase A at any school. u8 most prefers school h but decides to apply to school n under Phase A, because she fears she will not be able to get into h under Phase B.

The outcome of Phase A in this case is:

In the next phase, the rest of the students that have yet to be assigned a school apply to their next preferred choice. If they are rejected, they apply to the next preferred school that has available spaces. The final outcome of the current Singapore mechanism for this situation is as follows:

Notice how this mechanism is not strategy-proof. If u7 had refrained from entering Phase A and applied to n in Phase B instead, she would have been accepted into her top choice of school. Furthermore, this example illustrates how the Singapore mechanism is not devoid of justified-envy. For example, u6 justifiably envies u2 because u2 has a preferred match of l even though she has lower priority in Phase B than u6.

63


THE BERKELEY ECONOMIC REVIEW

64


VOLUME V

One consideration to be made is whether the new mechanism actually benefits the very people it is trying to help. Observe that six students are assigned to their most-preferred school under the new altered DA, as opposed to three students in the Singapore mechanism. Notice, furthermore, that every student is either equally well off or better off under the DA mechanism as compared to the Singapore mechanism, with the exception of u2, who gets her third choice as opposed to her second choice. However, recall from the previous example that u2 is the object of justified envy, so it was the failings of the previous system that had allowed u2 to receive a better school assignment than she should have gotten, given her position in the priority orders. The new model in question results in a stable matching. Note that the algorithm must end in a finite number of rounds. Suppose that a student is matched to a school s, but u prefers sâ&#x20AC;&#x2122;. At some point within the algorithm, u would have proposed to sâ&#x20AC;&#x2122; under Phase B (or both Phase A and Phase B), and been rejected by the school. Note that if that student has been rejected, it must be because that the school has been filled to its capacity with higherranked students in all of the phases u qualifies for. Since the ranking of students assigned to each phase in a school only improves as the algorithm continues, at the end of the algorithm, u must be lower ranked than every student assigned to s, and hence cannot form a blocking pair with any of these students.5 Therefore, the algorithm is fair. Fairness in arguably an important characteristic to have, because it prevents feelings of discontent from arising amongst the parents. Discussion forums about the registration exercise are often inundated with disgruntled parents, so having a system that is fair would relieve some of the tension as well as lift some burden off of the MoE to respond to parent queries. Furthermore, because the algorithm starts by considering studentsâ&#x20AC;&#x2122; highest ranked school, and only rejects a student if it is filled to its capacity with higher-ranked students, and each student has an opportunity to be considered non-wasteful. For any school that has empty seats, no student u would prefer s to This is given the condition that a student admitted under Phase A cannot form a blocking pair with someone who would have to be admitted through Phase B, due to the reservation of slots. 5

65


THE BERKELEY ECONOMIC REVIEW

to her existing assigned school.6 Another major advantage that arises from this model is that it is strategyproof, similar to the original DA algorithm. Fixing all the priority hierarchies of schools and preferences of all but one student u, the best option for u is to make a truthful report of her preferences. Suppose that the proposed DA algorithm will result in matching u to school s. Note that the matching can only be affected by the preferences that u lists before s. Misrepresenting her preferences by omitting some of her preferred schools or by listing them in a different order will not result in a change to her matching to school s. Listing a less preferred school sâ&#x20AC;&#x2122; above school s could possibly result in a less optimal match, as it is possible that u would be matched to sâ&#x20AC;&#x2122; instead of u. This would not be preferable for u. This analysis holds whether or not u is being considered under Phase A or Phase B. Strategyproofness is certainly a compelling argument for moving to a new algorithm. The acute need to strategise in the current mechanism provides an unfair advantage to families that have the resources and time to conduct the necessary research. Enabling families to to list their true choices of schools without jeopardizing their chances of being assigned to any school would reduce a lot of stress and time spent on the exercise. It would also increase clarity, allowing for straightforward advice to be provided to parents regarding how to rank schools.

In the presence of minimum quotas, it has been established that it is not necessarily the case that matchings are simultaneously fair and nonwasteful. [5] This boiled down to the nature of hard minimum quotas, as a student u might have to be assigned to a school s 0 with minimum quotas despite the possibility of being accepted into a more preferred school s. As a result, u could justifiably envy a student who has been accepted at school s despite being lower ranked than u. However, the algorithm proposed in this paper does not face this issue, because the minimum quotas are soft quotas. 6

66


VOLUME V

6. Conclusion Singapore’s Primary 1 Registration Exercise has long been a fixture in the young parents’ list of concerns – stories of parents starting to strategise about where to live or which school to volunteer at even before their child has been born are not uncommon. This paper builds upon years of research into school choice and mechanism design and draws upon the successful reform of the Boston mechanism as an impetus to prompt the reexamination of the Singapore mechanism. As expounded upon in the previous section, the proposed altered DA mechanism has the following properties: (1) strategyproofness, (2) fairness, and (3) nonwastefulness, making it a more preferable system to the existing one. It also has its advantages in practical implementation; the current system is a longdrawn process that takes place annually from July - November, whereas the new system would only need parents to submit a ranked list of preferences once, eliminating the need for in-person registration and phased admission. Theoretically, the exercise could be concluded in one afternoon. The transparency of the process would also be improved, and parents can be told to submit their true preferences rather than be urged to strategise and “game” the system. In closing, it is important to note that this paper does not tackle normative questions like whether or not it would be a good idea to remove distance-priority or phase quotas; research has shown that that quotas can actually be detrimental to the minority it tries to protect, because they potentially make competition for other schools higher. [4] These important questions warrant future investigation within the Singapore context, and are worth consideration should the Ministry decide to redesign the Primary 1 Registration Exercise.

67


THE BERKELEY ECONOMIC REVIEW

References Abdulkadiroglu, A., P. A. Pathak, A. E. Roth, and T. Sonmez 2005. The boston public school match. American Economic Review, 95(2):364–371. Abdulkadiroglu, A., P. A. Pathak, A. E. Roth, and T. Sonmez 2006. Changing the boston school choice mechanism. National Bureau of Economic Research, (w11865). Abdulkadiroglu, A. and T. Sonmez 2003. School choice: A mechanism design approach. American Economic Review, 93(3):729–747. Dur, U., S. D. Kominers, P. A. Pathak, and T. Sonmez 2014. The demise of walk zones in boston: Priorities vs. precedence in school choice. National Bureau of Economic Research, (w18981).

Fragiadakis, D., A. Iwasaki, P. Troyan, S. Ueda, and M. Yokoo 2015. Strategyproof matching with minimum quotas. ACM Transactions on Economics and Computation, 4(6). Hafalir, I. E., M. B. Yenmez, and M. A. Yildirim 2013. Effective affirmative action in school choice. Theoretical Economics, 8(2):325–363. Kiasu Parents 2017. 2017 p1 registration oversubscription risk. https://www.kiasuparents.com/kiasu/article/2017-p1-registration-oversubscription-risk/. Kojima, F. 2012. School choice: Impossibilities for affirmative action. Games and Economic Behavior, 75(2):685–693. Ministry of Education 2017. Registration phases and procedures. https://www.moe. gov.sg/admissions/primary-one-registration/phases.

68


VOLUME V

Teng, A. 2017. Pri 1: 12% admitted via 2nd, 3rd phases. The Straits Times. Westkamp, A. 2013. An analysis of the german university admissions system. Economic Theory, 53(3):561â&#x20AC;&#x201C;589.

69


THEBERKELEY BERKELE Y ECONOMIC ECONOMIC RE VIE W THE REVIEW

Human Capital Investment: An Examination of the Cyclicality of Bachelorâ&#x20AC;&#x2122;s Degrees Conferred by Field of Study

By Mark Vandre (University of California , Berkeley)


VOLUME VOLUMEIV V


THE BERKELEY ECONOMIC REVIEW

Human Capital Investment: An Examination of the Cyclicality of Bachelorâ&#x20AC;&#x2122;s Degrees Conferred by Field of Study by Mark Vandre University of California, Berkeley

72


VOLUME V

Abstract

Using data combined from the Integrated Postsecondary Data System (IPEDS) and the Bureau of Labor Statistics (BLS) covering 907 institutions from 1997 through 2015, I investigate the effect of business cycles on the distribution of degrees conferred in five fields. Employing a fixed effects regression model with entity and year effects, I show the cyclicality of the proportion of degrees conferred in business as well as science, technology, engineering, and mathematics (STEM). I find the proportion of STEM degrees conferred to be counter-cyclical, while the proportion of business degrees conferred is pro-cyclical.

Acknowledgements Many thanks to Kayleigh Barnes, Stephanie Bonds, and Ganesh Viswanath for their fielding of my many questions and for providing guidance and comments. I would also like to thank Zachary Bleemer for his general guidance and mentorship.

73


THE BERKELEY ECONOMIC REVIEW

I. Introduction The Great Recession proved to be one of the most devastating economic events in recent history. Farber (2011) finds approximately 16% of 20 to 64-yearolds lost a job, 20% of those who lost a job were only able to find part-time work to replace the lost job, and those who did find a new job earned on average 17.5% to 21.8% less compared to before they were laid off. Further, the persistence of jobloss was more severe than previous downturns, with fewer than 50% of job-losers finding employment two years after the recession hit (Farber 2011). Similar, though less severe, patterns occur during the normal ups and downs of the business cycle. Modern theory in labor economics predicts that when there is persistently high unemployment, the opportunity cost of foregoing earnings to invest in higher education decrease and more individuals will pursue such investments (Ehrenberg & Smith 2012, 285). Hillman and Orians (2013) find evidence supporting such a prediction by showing that a 1% increase in the unemployment rate is associated with an increase between 1.1% and 3.3% in new community college enrollment. Similarly, Long (2014) finds four-year universities experience an increase in new enrollment demand in response to rising unemployment. More specifically, recessions have a relatively significant impact on increasing enrollment in higher education as shown by Barrow and Davis (2012). Additionally, data from 1981 through 2004 shows that a 1% increase in unemployment increases the likelihood of program completion by between 4% and 12%, according to Reiling and Strøm, (2015). Messer and Wolter (2010) find not only are students more likely to graduate, but they also graduate more quickly during downturns by 12%, on average, for every 1% increase in unemployment. While it is important to demonstrate how patterns in higher education investment coincide with theoretical predictions, these patterns do not explain what students tend to do upon enrolling, or about what field students tend to study. If higher education is an investment decision and business cycles influence whether individuals invest, then it seems likely business cycles would also influence what field of study individuals invest in as a response. If unemployed individuals respond by pursuing degrees in high-return fields, this would support the notion that individuals treat higher education as an investment. If unemployed individuals instead respond by pursuing degrees in fields with relatively low

74


VOLUME V

returns, then there may be a violation of the rationality assumption, or possession of incomplete information on the part of newly enrolled students. I examine what effect the business c ycle, as measured by local unemployment, has on the proportions of degrees conferred by field. The results connect the segment of literature focused on the decision to invest in higher education with the segment of literature focused on the return on investment of specific fields of education. If the connection drawn shows unemployment driven increases in enrollment result in more degrees conferred in more lucrative fields, then the investment model of the decision to pursue higher education is further bolstered. My analysis uses balanced panel data of 907 four-year universities across the United States, which provides full data for the 19 years covered from 1997 to 2015. Using the entity-demeaned fixed effects regression model, the effect of business cycles on the variation in proportion of degrees awarded in each field can be measured through unemployment. By using entity and year fixed effects, the model is able to control for unobservable variables unique to each university across time and each year. By examining the years from 1997 through 2015, the variation in unemployment includes exogenous shocks from the bursting of the Dotcom Bubble and also the Great Recession. Exogeneity is assumed as it is unlikely that overvaluation of corporate stocks during the Dotcom Bubble or the overvaluation of mortgage-backed securities leading into the Great Recession held much sway in either the decision of individuals to study a specific field or of universities to supply degrees in a specific field prior to the respective downturns. The results of my analysis show significant effects for STEM and business degrees with the proportion of STEM degrees conferred being significantly counter-cyclical and the proportion of business degrees being significantly procyclical. The effects for the remaining fields of arts and design, humanities, and social sciences remain ambiguous with no statistical significance. The data shows a 1 percentage point increase in unemployment results in a 0.697 percentage point increase in the proportion of STEM degrees conferred, less 0.084 percentage points for each 1% increase in enrollment, holding all else constant. Conversely, the data shows a 1 percentage point increase in unemployment results in a 0.714 percentage point decrease in the proportion of business degrees conferred, less 0.096 percentage points for each 1% increase in enrollment, holding all else constant.

75


THE BERKELEY ECONOMIC REVIEW

To verify the robustness of the results, five regressions are estimated for each dependent variable, adding in a new independent variable each time. Additionally, several tests are employed. The Hausman Test is used to determine whether fixed or random effects are appropriate for the model, the Ramsey RESET test is used to test for misspecification and omitted variables, and the joint significance of year fixed effects are tested to determine if including them is appropriate. The rest of this paper is organized as follows. The next section reviews the relevant literature on the topic being studied. Section IV discusses the empirical strategy, model, and data. Section V presents the results and provides discussion on the hypotheses presented in this paper. Section VI provides a brief review and conclusion with special attention paid to the validity of the results.

76


VOLUME V

II. Literature Review Enrollment demand increases and strengthens when the economy weakens, which indicates individuals prioritize human capital accumulation, or respond to the lessened opportunity costs of pursuing education when times are tough as shown by Messer and Wolter (2010), Barrow and Davis (2012), Hillman and Orians (2013), Long (2014), and Reiling and Strøm (2015). Such a behavioral response is likely accompanied by similar changes in attitudes regarding which field to study. Theory would suggest these changes in attitude would push more individuals to choose higher return fields. With regard to earnings returns to specific majors, male/female engineering students earn 24%/39% more than the average male/ female graduate, and male/female students who major in health-related fields earn 12%/19% more than the average male/female graduate (respectively), according to Rumberger and Thomas (1993). Another estimate by Webber (2014) shows that STEM graduates earn lifetime earnings premiums of about $1.5 million over high school graduates, compared to an earnings premium of $700,000 for arts and humanities graduates. However, Hamermesh and Donald (2008) find the difference to be present, but less striking. When considering the earnings premium for STEM fields, in addition to the key role such premiums play in a studentâ&#x20AC;&#x2122;s choice of major in higher education (Montmarquett et al., 2002), it is not hard to imagine the role business cycle f luctuations may play in inf luencing the field students choose to study. Additionally, Arcidiacono et al. (2011) find 7.8% of students would switch majors if the majors only differed in the studentâ&#x20AC;&#x2122;s perceived comparative advantage across majors. This suggests the earnings students expect to receive play a larger role in the major decision than other considerations. These results, however, only indicate what students think about field choice, but do not demonstrate what students do when faced with choosing a major. Lian Niu (2017) shows that students of lower socioeconomic status are less likely to choose a STEM field as compared to higher socioeconomic status families. This shows earnings premiums may not be the only factor at play in major choice, and the influence business cycles have on said choice may differ by socioeconomic strata. In viewing education as an investment decision, it is not only important to take returns into account, but also to take risks into account. STEM fields are

77


THE BERKELEY ECONOMIC REVIEW

widely considered to be more difficult than other majors with lower average grade point averages (GPA). If students are faced with the uncertainty of a recession, they may be unwilling to take on the additional uncertainty of pursuing a STEM degree as compared to another field that would be a relatively safe choice in terms of perceived difficulty. Main and Ost (2014), however, provides evidence that letter grades do not significantly influence student’s course taking or major selection behavior. This would suggest that the differences in average GPA may not be taken into account during a recession. Some recessions are either the result or cause of the contraction of specific industries. In these instances, it may be the case that if a certain industry is in distress, the accompanying degrees relevant to the industry may also see a fall in student pursuit. Kinsler and Pavan (2015) find that students with a degree in STEM fields earn 30% less working in an unrelated field, business majors earn 3% less, and other majors earn an average of 11% less working in fields unrelated to the individual’s degree. Similar studies done by Robst (2006), Yakusheva and Nordin (2010), and Persson and Rooth (2010) estimate wage penalties around 20% for business, engineering, health, computer science, and law professionals. These results might imply that students will choose to major in the highest paying field having the most reasonable expectation of employment. The body of literature synthesizes nicely into a backdrop for my primary question: during fluctuations in the business cycle, what changes can be observed in the proportion of degrees conferred by field of study? During dips in the business cycle, individuals are more likely to enroll in college to invest in human capital. Individuals decide what field to study mostly based on expected returns on wages. STEM majors will be most favored, since those fields have the highest return on investment and the industries involving most STEM fields are less likely than another industry—such as manufacturing—to experience a relatively severe contraction during such a downturn. My paper complements the literature as I will focus on observing actual changes in field of degree composition in response to varying local unemployment rates using institution-level IPEDS panel data from 1997 through 2015. Similar to Hillman and Orians (2013), Reiling and Strøm (2015), Messer and Wolter (2010), and Webber (2014), the fixed-effects panel data model will be used to analyze the IPEDS data. This approach will extend the existing literature and provide a

78


VOLUME V

springboard for more detailed examination of this question in the future.

79


THE BERKELEY ECONOMIC REVIEW

III. Model, Methods, and Data The data I used is panel data taken from the Integrated Postsecondary Education Data System (IPEDS), as aggregated by the American Institutes for Researchâ&#x20AC;&#x2122;s Delta Cost Project and the Department of Educationâ&#x20AC;&#x2122;s College Scorecard Data. Additional data from the Bureau of Labor Statistics (BLS) is used to complement the IPEDS data. The IPEDS data provides institutionlevel information on colleges and universities, and the BLS data provides local economic conditions. All data covers the years from 1997 through 2015. The sample of institutions includes public and private non-profit colleges and universities that offer degrees in the fields of study examined, and are primarily bachelordegree granting with graduate degrees available. After eliminating institutions that are missing data for the years covered and balancing the data, the number of universities totals 907 over 19 years for a total sample of 17,233 observations. One of the strengths of this data set is that these years include the recession associated with the bursting of the Dotcom bubble and the Great Recession. This data will be used to test the hypotheses that degrees conferred will be counter-cyclical for the STEM field, and all other fields will be pro-cyclical. The dependent variables come from the IPEDS data and they include the proportion of degrees conferred in five broad fields of study: STEM (Stem it), business (Busit), the social sciences (Socialit), humanities (Humit), and arts and design (Artit). The IPEDS data includes fields of study that are already aggregated into more general fields of study that were then further aggregated into these five broad categories. The control variables that come from the IPEDS data include the natural logarithm of full-time fall enrollment (Feit), of in-state tuition and fees (Pit), of total revenue from state grants and appropriations (Sf it), and of total revenue from federal grants and appropriations (Ff it). In-state tuition and fees is the variable used in the literature to measure the price of attending a university and is colloquially referred to as the sticker-price by Martin (2003), Abel and Deitz (2014), Altringer and Summers (2015), Lucca et al. (2015), and Bleemer et al. (2017). For private universities that do not distinguish between residents and non-residents of their state, this number is simply the universityâ&#x20AC;&#x2122;s single tuition and fees number. Tuition and fees, state funding, and federal funding are expressed in 2015 dollars prior to undergoing log transformations.

80


VOLUME V

The data from BLS is the county-level annual unemployment rate (Urit) for each institutionâ&#x20AC;&#x2122;s county in each year. This serves to provide an estimate of the economic situation in each institutionâ&#x20AC;&#x2122;s local area and is the independent variable of primary interest. To account for lag effects, the value of every control variable and of unemployment rates for each observation is that of four years prior, since it is likely that the conditions of four years ago will have a stronger effect on degrees conferred compared to the university conditions of today. Additionally, an interaction variable of the unemployment rate times the natural logarithm of enrollment is used to account for the interrelated nature of these variables. The reasoning is that if increases in unemployment are linked to increases in enrollment as the literature indicates, then this will allow for the increase in enrollment that results from unemployment. Conversely, this also models the effect of unemployed individuals dropping out of the labor force to attend university fulltime. To examine the question of each fieldâ&#x20AC;&#x2122;s direction of cyclicality, I employ the fixed effects regression model accounting for both entity and time fixed effects. It is important to account for entity fixed effects because there are many qualities about each institution of higher education unique to that institution, which do not change over time. Time fixed effects are important in accounting for the many unmeasurable changes all institutions face as time changes. Given the large number of entities (907), however, entity fixed effects are accounted for by using the entitydemeaned ordinary least squares (OLS) method of fixed effects regression. This is done by subtracting the entity-specific average of each variable over time from the equation, and using these entity-demeaned variables to estimate the regression of interest as shown below:

81


THE BERKELEY ECONOMIC REVIEW

In the above equations, Yit represents any of the five dependent variables, Xit represents the independent and control variables, uit represents the error term, αit represents the entity fixed effect plus the intercept β0, and λt represents time fixed effects. This results in equation (5) where the control and independent variables which my analysis covers are substituted back in, resulting in the population regression I attempt toestimate. To verify the robustness of the unemployment variable, a regression of just this variable with fixed effects is estimated, and then four more regressions are estimated adding one of the excluded controls at a time. On the fourth and fifth estimation, year fixed effects are added. These five regressions are conducted for each of the five dependent variables separately with special attention paid to how the unemployment coefficient changes for each estimation. While the data possesses many strengths, there are also weaknesses. By using institution-level data, I can only speak to general trends in degrees conferred by institution. Additionally, access to student-level data might have enabled an examination of the fields individuals apply to in response to the economic conditions in each individual’s local area at the time of application. Further, individual student data may have enabled observation of students switching majors in response to changing economic conditions.

82


VOLUME V

IV. Results and Discussion I test the hypothesis that the proportion of STEM degrees awarded is countercyclical and the hypothesis that the proportion of degrees awarded in all other fields is pro-cyclical. The method employed to test these hypotheses is the entity-demeaned fixed effects regression with year fixed effects as presented in Table 3 through Table 7. Entitydemeaned fixed effects are used due to the large number of entities (907). For each field, the initial estimation includes only unemployment and adds one control variable for each subsequent estimation. Regression (4) and (5) include year fixed effects. Table 1 presents descriptive statistics of all variables used. Several tests are employed on each of the five regressions to determine the validity of the model for each dependent variable as shown in Table 2. To determine whether entity random effects or fixed effects were preferable, the Hausman test is used. The results for all five regressions provide sufficient evidence to reject the null hypothesis that random effects were better suited at the 1%, 5%, and 10% levels of significance. Therefore, entity fixed effects are used. To determine the validity of the model specification, the Ramsey RESET test is used. The result for all regressions, except for social sciences, indicate there is insufficient evidence to reject the null hypothesis that the model is correctly specified at the 1% and 5% level of significance. The third and final test is a test of joint significance of year fixed effects to determine the appropriateness of including them. The results of this joint significance test provide sufficient evidence to reject the null hypothesis that the year fixed effects are jointly equal to zero at the 1%, 5%, and 10% levels of significance for all five regressions. For all regressions, university-level clustered standard errors are used to address autocorrelation likely to be present in the data. The coefficients of interest for the STEM regressions in Table 3 were significant. The coefficient on the unemployment rate is positive and statistically significant at the 5% and 10% level of significance for every estimation performed. The interaction of unemployment with the natural logarithm of enrollment in columns (4) and (5) is negative and statistically significant at the 5% and 10% level of significance. The results of columns (4) and (5) imply that for every 1 percentage point increase in unemployment there is a 0.697 percentage point increase in the proportion of STEM degrees awarded, less 0.084 percentage points for every 1% increase in enrollment, holding all else constant. The negative coefficient on the interaction term is expected, since an increase in unemployed workers enrolling full-time in university would

83


THE BERKELEY ECONOMIC REVIEW

decrease the number of individuals looking for work and would therefore reduce unemployment through a reduction in labor force participation. Additionally, this coefficient illustrates that as more students enroll, there may be institutional constraints or incentive structures in place that induce diminishing marginal returns to STEM degrees as unemployment driven enrollment grows. The remaining variable coefficients are not statistically different from zero. The coefficients of interest for the business regressions are also significant. The coefficient on unemployment in columns (4) and (5) is negative and statistically significant at the 5% and 10% percent level of significance. The coefficient on the interaction term in columns (4) and (5) is positive and statistically significant at the 1%, 5%, and 10% levels of significance. The coefficient on the interaction term is expected, for the same reasons as the STEM interaction term in the opposite direction. For every 1 percentage point increase in unemployment, there is an associated 0.714 percentage point decrease in the proportion of business degrees awarded, less 0.096 percentage points for every 1 percentage point increase in enrollment, holding all else constant. Similar to the STEM case, all other variable coefficients are statistically insignificant. For the remainder of the degree fields, the results are inconclusive. The humanities regressions of Table 4 and the arts and design regressions of Table 7 both have no statistically significant variable coefficients in columns (4) and (5). For the social sciences, the results are significant at the 10% level of significance, but given the regression lacked sufficient evidence to conclude proper specification, interpretations drawn from the regression are dubious. One possible explanation for the lack of significance in the humanities is countervailing responses to unemployment for the specific majors within the general field that negate much of the variability when the majors are aggregated. For the arts and design field, it may be the case that the range of proportions of degrees awarded for the field is too small to allow the variability necessary to achieve significant results. For social sciences, it seems likely this field has a different set of variables that are more relevant to changes in proportion of degrees awarded, which may not be measured at the institution-level. My hypothesis stating STEM degrees are counter-cyclical is supported by the results. However, it is not clear if these results support the hypothesis that all other fields are pro-cyclical. While business degrees are shown to be pro-cyclical, there is insufficient evidence to conclude that the remaining fields are pro-cyclical at any level of significance as the model is currently specified. The challenges with the

84


VOLUME V

humanities, social sciences, and arts and design regressions illustrate the limitations of my analysis. While some significant results are found, more significant results might have been possible with data on individual students. If individual student data had been used, it would be possible to link the economic conditions in the location where each student resided at the time of college application to the field the student applied to or switched to after acceptance into a university. An additional layer of complexity is the substitution of STEM for business degrees during economic downturns. This is somewhat puzzling, since business graduates fall just behind STEM graduates in lifetime earnings (Webber, 2014). In examining higher education as an investment decision, it would make more sense for the relatively lower earning fields, such as humanities or social sciences, to be the fields predominantly switched away from. More research and detailed data is needed to fully examine this relationship.

85


THE BERKELEY ECONOMIC REVIEW

V. Conclusion Given the evidence from the literature that university enrollment increases during downturns in the business cycle, and that STEM degrees have higher returns relative to other fields, my analysis tests the conception of higher education as an investment. To do this, the entity-demeaned fixed effects regression model is used with year fixed effects. The results of the estimation suggest the proportion of STEM degrees conferred to be counter-cyclical, while the proportion of business degrees conferred is pro-cyclical. The counter-cyclicality seems to support the conception of higher education as an investment, due to the relatively high earnings of graduates in this field, but the result of business being the only significantly pro-cyclical field is somewhat puzzling within the investment framework. One possible reason for this puzzle could be the use of universitylevel data rather than individual student-level data. These results likely apply to universities within the United States. It seems unlikely the results would be valid outside of the United States, given the myriad differences in how education is structured in other countries, from early childhood education all the way to post-secondary education. Additional items preventing external validity would likely include differences in culture and attitudes towards education and different industries. Future research could expand upon and improve these results by using studentlevel data to more closely match the economic conditions of each individual student to the field each student selects at the time of application to a university. This would likely allow for better specification of the models, especially for social sciences. Additionally, future research could delve deeper into the relationship between STEM and business. There may be characteristics of the students typically applying to these majors making them appear to be substitutes for one another. Student-level data would also assist in this analysis.

86


VOLUME V

VI. Bibliography Abel, Jaison, & Richard Deitz. 2013. “Do the Benefits of College Still Outweigh the Costs?” Current Issues in Economics and Finance, 20 (3). https://papers.ssrn.com/sol3/papers.cfm?abstract_ id=2477864. Altringer, Levi, & Jeffrey Summers. 2015. “Is College Pricing Power Pro cyclical?”  Research in Higher Education, 56 (8): 777-792. doi:10.1007/s11162-015-9373-z. American Institutes for Research. 2017. Delta Cost Project Database, 1987-2015. Washington, DC. http://www.deltacostproject.org/delta cost-project-database. Arcidiacono, Peter, Joseph Hotz, & Songman Kang. 2012. “Modeling College Major Choices Using Elicited Measures of Expectations and Counterfactuals.” Journal of Econometrics, 166 (1): 3-16. doi: 10.1016/j.jeconom.2011.06.002. Barrow, Lisa, & Jonathan Davis. 2012. “The upside of down: Postsecondary enrollment in the Great Recession.” Economic Perspectives, 36 (4): 117-129. https://www.chicagofed.org/~/media/ publications/economic-perspectives/2012/4q2012-part1-barrow davis-pdf.pdf. Bleemer, Zachary, Brown, M., Lee, D., Strair, K., Klaauw, W. 2017. “Echoes of Rising Tuition in Students’ Borrowing, Educational Attainment, and Homeownership in Post-Recession America.” Federal Reserve Bank of New York Staff Reports, no. 820. Federal Reserve Bank of New York. New York, NY. https://www.newyorkfed.org medialibrary/media/research/staff_reports/sr820.pdf. Ehrenberg, Ronald, & Robert Smith. 2012. Modern Labor Economics: Theory and Public Policy. Upper Saddle River: Prentice Hall, 2012. 87


THE BERKELEY ECONOMIC REVIEW

Farber, Henry. 2011. “Job Loss in the Great Recession: Historical Perspective from the Displaced Workers Survey, 1984-2010.” NBER Working Paper 17040. National Bureau of Economic Research, Cambridge, MA. https://www.nber.org/papers/w17040. Hamermesh, Daniel, & Stephen Donald. 2008. “The Effect of Curriculum on Earnings: An Affinity Identifier for Non-Ignorable Non-Response Bias.” Journal of Econometrics, 144 (2): 479-491. doi:10.1016/j.jeconom.2008.04.007. Hillman, Nicholas, & Erica Orians. 2013. “Community Colleges and Labor Market Conditions: How Does Enrollment Demand Change Relative to Local Unemployment Rates?” Research in Higher Education, 54(7): 765-780. doi:10.1007/s11162-013-9294-7. Kinsler, Josh, & Ronni Pavan. 2015. “The Specificity of General Human Capital: Evidence from College Major Choice.” Journal of Labor Economics, 33 (4): 933-972. https://www.journals.uchicago.edu/ doi/10.1086/681206. Long, Bridget. 2014. “The Financial Crisis and College Enrollment: How Have Students and Their Families Responded?” NBER Chapters, 209. National Bureau of Economic Research, Cambridge, MA. https://www. nber.org/chapters/c12862. Lucca, David, Taylor Nadauld, & Karen Shen. 2015. “Credit Supply and the Rise in College Tuition: Evidence from the Expansion in Federal Student Aid Programs.” Federal Reserve Bank of New York Staff Reports, no. 733. Federal Reserve Bank of New York. New York, NY. https://www.newyorkfed.org/research/staff_reports/sr733.html. Main, Joyce, & Ben Ost. 2014. “The Impact of Letter Grades on Student Effort, Course Selection, and Major Choice: A Regression Discontinuity Analysis.” Journal of Economic Education, 45 (1): 1-10. doi:10.1080/00220485.2014.859953. 88


VOLUME V

Martin, Robert. 2003. “Pricing and Enrollment Planning.” Planning for Higher Education, 31 (4): 29-37. https://eric.ed.gov/?id=EJ676397. Messer, Dolores, & Stefan Wolter. 2010. “Time-to-Degree and the Business Cycle.” Education Economics, 18 (1): 111-123. doi:10.1080/09645290903102860. Montmarquette, Claude, Kathy Cannings, & Sophie Mahseredjian. 2002. “How do Young People Choose College Majors?” Economics of Education Review, 21 (6): 543-556. doi:10.1016/S0272-7757(01)00054-1. Niu, Lian. 2017. “Family Socioeconomic Status and Choice of STEM Major in College: An Analysis of a National Sample.” College Student Journal, 51 (2): 298-312. https:// eric.ed.gov/?id=EJ1144312. Nordin, Martin, Inga Persson, & Dan-Olof Rooth. 2010. “Education–Occupation Mismatch: Is there an Income Penalty?” Economics of Education Review, 29 (6): 1047-1059. doi:10.1016/j.econedurev.2010.05.005. Reiling, Rune, & Bjarne Strøm. 2015. “Upper Secondary School Completion and the Business Cycle.” Scandinavian Journal of Economics, 117 (1): 195-219. doi:10.1111/ sjoe.12088. Robst, J. 2007. “Education and job match: The Relatedness of College Major and Work.” Economics of Education Review, 26 (4): 397-407. doi:10.1016/j.econedurev.2006.08.003. Rumberger, Russell, & Scott Thomas. 1993. “The Economic Returns to College Major, Quality and Performance: A Multilevel Analysis of Recent Graduates.” Economics of Education Review, 12 (1): 1-19. https://ideas.repec.org/a/eee/ecoedu/ v12y1993i1p1-19.html. U.S. Department of Education. 2017. College Scorecard Data 1996-2016. Washington, DC. https://collegescorecard.ed.gov/data/.

89


THE BERKELEY ECONOMIC REVIEW

U.S. Department of Labor. 2017. Local Area Unemployment Statistics 1990-2015. Washington, DC. https://data.bls.gov/cgi-bin/dsrv?la. Webber, Douglas. 2014. “The Lifetime Earnings Premia of Different Majors: Correcting for Selection Based on Cognitive, Noncognitive, and Unobserved Factors.” Labour Economics, 28: 14-23. doi:10.1016/j.labeco.2014.03.009. Yakusheva, Olga. 2010. “Return to College Education Revisited: Is Relevance Relevant?” Economics of Education Review, 29 (6): 1125-1142. doi:10.1016/ j.econedurev.2010.06.006.

90


VOLUME V

[THIS PAGE INTENTIONALLY LEFT BLANK]

91


THE BERKELEY ECONOMIC REVIEW

Profile for Berkeley Economic Review

Berkeley Economic Review Volume V (Spring 2018)  

The Berkeley Economic Review is the University of California at Berkeley’s premier undergraduate, peer-reviewed, academic economics journal....

Berkeley Economic Review Volume V (Spring 2018)  

The Berkeley Economic Review is the University of California at Berkeley’s premier undergraduate, peer-reviewed, academic economics journal....

Advertisement

Recommendations could not be loaded

Recommendations could not be loaded

Recommendations could not be loaded

Recommendations could not be loaded