6 minute read

Modelling an emotive topic – dealing with induced demand

No one should doubt the existence of generated traffic, and the need to reflect it in our models. But how?

It’s one of those topics that doesn’t want to go away – whether roads generate traffic and if transport modelling reflects the induced demand phenomenon well enough. Tom van Vuren considers the evidence

Advertisement

I was inspired to revisit the topic when I was pointed towards a paper that said “Ignoring induced demand is engineering malpractice”,* and there’s a lot in that document that I strongly disagree with. But it’s a good example of how emotive and, in my view, often ill-informed the discussion on the topic remains.

What’s in a name? Some talk about induced demand, others about induced traffic, and the original SACTRA Report “Trunk Roads and the Generation of Traffic” used the term generated traffic, and that’s the term I’m settling on. Does it matter? I would say yes. As so often is the case, terms that sound similar at first sight can mean something quite different to different people.

Generated traffic has been described slightly differently by many authors, but to me and others it reflects that investing in roads, increasing capacity and removing bottlenecks has been found to result in more car-based traffic on the road itself and/or in the overall study area, in the time period considered, negating some or most of the intended benefits. Where that demand comes from has been less concerning for many, particularly if the objective of the road scheme had been congestion reduction – the ‘what’ being more interesting than the ‘why’.

But for me, as a transport planner, it does matter where that new, generated traffic comes from. Did it previously use other modes, perhaps reducing the commercial base for public transport as an alternative? Have drivers changed destination, lengthening their distance travelled, but also potentially contributing to decongestion benefits elsewhere, outside of our immediate area of interest? Did they return to the peak? Or are these completely new trips, true induced travel demand?

Thinking along these lines, studies have shown that much generated traffic is actually redistributed from other alternatives (modes, destinations and time periods) that we already include as explicit choices in most of our (strategic) transport models. The Amsterdam Ring Road was an early case study in 1996, where an additional river crossing generated what was estimated to be 5% extra traffic after one year, and about 7% after five years.

A detailed analysis of the generated traffic suggested that some destination changes had taken place, a substantial amount of trip re-timing, hardly any mode shift and a surprising amount of changes in car occupancy. A later 2013 Department for Transport (DfT) study of completing the Manchester Motorway Box estimated that generated effects consisted mainly of destination changes (and that these were strongest for non-commute purposes), with some mode shift. In the Manchester case hardly any time period

shifting occurred. The actually induced demand – that is wholly new trips – was found to be small in both studies. And that’s why language is important.

An extensive recent review for DfT by WSP and RAND Europe finds that “state level road networks in the US and the national Dutch network indicate an elasticity of around 0.2 across the whole road network, i.e. a 10% increase in road capacity could lead to 2% induced demand on the network”.

Does that mean that approaching generated traffic through a more simplistic elasticity approach will suffice? I don’t think so – apart from the practical problem of reflecting road capacity changes in our models as discrete variables – I prefer to relate generated traffic to what the model calculates as generalised cost changes as experienced by travellers. Reflecting it through distributive processes enables us to do two things:

l It allows us to model and value the full impacts of the redistribution of travel demand

l It is much better at reflecting the local conditions that to a large extent determine the amount of traffic that is expected to be generated (as a function of the attractiveness of alternative modes, destinations and time periods).

A good example of this latter argument is the finding of the paper “Disappearing traffic? The story so far”, published by Cairns, Atkins and Goodwin in 2002, describing the opposite effect. The authors find for 70 case studies that, following a road-space reduction measure, the average drop in traffic flow (in the study area or crossing a cordon) is around 22%, but the figure ranges between studies from a more than 100% reduction to increases of 25%. To me that suggests that local conditions, and particularly the quality of alternatives, are of paramount importance for both generated and disappearing traffic, which can only be reflected in distributive model approaches that reflect all relevant choices explicitly.

Induced demand, generated traffic, does not have to be an emotive subject. The transport modelling and appraisal profession has responded to the recommendation of the 1994 SACTRA Report: “… that variable demand methods should now become the normal basis of trunk road traffic forecasts, and that these forecasts must be carried through into the operational, economic and environmental evaluation of schemes in a systematic way”. If you have designed and built your transport model aligned with TAG guidance (particularly Unit M2.1 – Variable Demand Modelling), you should already have considered and allowed for the responses that lead to the generation of traffic.

We have the modelling mechanisms, but we can do better. Reports that have looked at how well we have forecast generated traffic are “Beyond Transport Infrastructure” by Transport for Quality of Life and “The end of the road? Challenging the road-building consensus” by the Campaign to Protect Rural England. Some of the examples in these reports probably predate TAG. But they make it clear that one effect that is generally not well allowed for is location choice.

Generally only reflected in complex and quite data-hungry land use transport interaction models that are often poorly understood by end users, and normally applied as an optional add-on to transport models, it’s time to embrace location choice as an explicit choice into our 4, 5 or even 6 stage modelling framework.

Using recent advances in spatial computable generalised equilibrium framework, my VLC colleagues in Australia managed to represent location choice in a simpler way, faster and less data hungry than current practice. And it’s worthwhile. They calculated for a heavy rail scheme in Melbourne, using our in-house SPATIAL location choice model, a shift of population and employments closer to rail stations, and added scheme benefits, mainly due to decongestion, of 20%. I’d expect that the opposite would happen when testing a major road scheme.

No one should doubt the existence of generated traffic, and the need to reflect it in our models. And we generally do. How to improve? It all comes down to monitoring. The 2019 National Highways POPE report “Evaluation Insight Paper: Post Opening Project Evaluation of Major Schemes” states that less than 60% of 85 road schemes opened between 2002 and 2014 accurately forecast post-scheme traffic volumes (accurate meaning volumes within +/- 15%). There are many possible reasons for that discrepancy, but to increase the trust that planners have in our modelled forecasts we must point not just to the model mechanisms that are included but also critically assess their past effectiveness in forecasting generated traffic, and be willing to make changes.

If you want to read up further on the subject, the WSP/RAND Europe report has a good list of references. Names outside of the UK to look out for are Todd Litman, Robert Noland, Peter Naess and Greg Erhardt. *https://www.strongtowns.org/journal/20 22/4/4/ignoring-induced-demand-isengineering-malpractice n

The 2019 National Highways POPE report states that less than 60% of 85 road schemes opened between 2002 and 2014 accurately forecast postscheme traffic volumes (accurate meaning volumes within +/- 15%). To increase the trust that planners have in our modelled forecasts we must point not just to the model mechanisms that are included, but also critically assess their past effectiveness in forecasting generated traffic

Tom van Vuren is Chairman, Modelling World & Regional Director, UK & Europe, Veitch Lister Consulting www.veitchlister.com

This article is from: