When we talk about transport, it’s natural for our minds to focus on moving people from A to B. That is, we tend to think about how to meet the demand for travel between locations, assuming that the demand for being in those locations is unaffected by the transport policies that we implement. In this post I present some evidence that questions this assumption, and argue there is a need to re-frame how we think about the economic effects of transport policy.
In some situations the effects of transport infrastructure on location choice is readily apparent. The Auckland Harbour Bridge, for example, opened up vast swathes of extremely amenable land close to Auckland’s City Centre. Surprise, surprise: After the Bridge opened many people and firms chose to locate on the North Shore, increasing the demand for travel between there and other locations. If we had better understood the effects of transport on location choice at that time (no doubt some people did), then perhaps we would have retained the ability to add rapid transit to the AHB?
A growing number of economic studies analyse the locational effects of transport investment in a systematic way, and find that they are indeed significant. Let’s consider a couple of recent studies.
Duranton and Turner (2011) analyse the relationship between the supply of road capacity (measured in lane-kilometres) and the demand for vehicle travel (measured in vehicle kilometres travelled) in the U.S. The authors find a nearly perfectly elastic relationship between supply and demand, which they refer to as the “fundamental law of traffic congestion”. That is, a 10% increase in supply is met by a ~10% increase in demand. Duranton and Turner present evidence that a large chunk of this demand response is caused by changes in location, with the balance associated with changes in travel behaviour (such as mode and route choice).
Teulings et al (2014) analyse the effects of rail services on location choice in the Netherlands. After motivating and estimating their economic framework, the authors then simulate how people relocate in response to the withdrawal of rail services to a region north of Amsterdam. They find relatively large changes in the location of people and firms when rail services are withdrawn. The economic value of the land use effects of rail services are approximately 25 percent of the transport (modal split) benefits. And in this context, the main effects are felt by the highly-educated, who are more likely to live and work in the locations where rail is available.
While not published in an academic journal, Auckland Council’s recent (excellent) research into the effects of walking times on effective density and productivity sit on the same tentacle of the economic octopus.
In a similar vein, my recently completed masters thesis uses commuting data to analyse the effects of walking and cycling travel-time on where people choose to live and work in three Australian cities. The intuition underlying my analysis is that commuting data reveals information on people’s preferences for home and work locations, and the perceived costs of travelling between the two. For those of you who are short on time, here’s my results in a nut-shell:
I find that walking and cycling travel-time has an economically and statistically significant effect on where people choose to live and/or work. A one minute saving on a 15 minute journey between two locations is predicted to cause a 1-8 percent increase in commuting flows between those locations. The magnitude of the effect decreases with commute length, and varies by city and model specification. I find some evidence that low-income commuting flows are more sensitive to walking and cycling travel-time.
In my thesis I focus on Australia in general and Brisbane in particular. Australia provides a useful setting for such analysis due to its comprehensive census data and high levels of residential mobility. People change where they live and work relatively frequently, which in turn enables them to take advantage of differences between locations, e.g. due to changes in walking and cycling travel-time. Brisbane is particularly interesting because of the Brisbane River, which carves a torturous path through inner-city suburbs and introduces considerable variation in travel-times between transport modes and locations. I exploit the quasi-experimental setting created by the Brisbane River to identify the effects of walking and cycling separately from other transport modes.
The effect of the River on my data can be seen in the following two figures. The left-hand figure plots car travel-times (horizontal axis) versus cycle travel-times (vertical axis) between origins and destinations in my data set (NB: Locations are defined as the centroids of census areas). The 45-degree line demarcates journeys for which cycle and car travel-times are equal; points that lie below the line relate to journeys for which cycling is faster and vice versa for points that lie above the line. The right-hand figure shows the ratio of cycle travel-times to car travel-times. While car is, on average, approximately twice as fast as cycling, I find 155 origin-destination pairs for which cycling is faster than car. Many of these locations relate to suburbs that are close to the Brisbane River.
As noted above I find that reducing the walking and cycling travel-time between two locations causes an increase in commuting flows between those locations. The magnitude of the effect is relatively large for short journeys: A one minute saving on a 15 minute journey between two locations increases commuting flows by 1-8 percent. Somewhat intuitively, the size of this effect decreases (linearly) with distance. I find some evidence that low-income (possibly part-time?) commuters are more sensitive to walking and cycling travel-time, although the difference is not always significant. The effect in Brisbane > Perth > Adelaide, which I hypothesise may reflect differences in congestion levels in each city, with larger locational responses in larger cities.
So what does all this mean for other places? While one must always be careful when extrapolating analytical results beyond the context in which they are derived, there are three fairly general implications that seem likely to transfer to other settings, such as Auckland.
First, my results align with the growing body of economic evidence which finds that transport investment has significant effects on location choice. When thinking about transport policies, we should consider not just how they relate to the demand for travel, but also the demand for locations. So my pitch to Mayor Goff and other councillors is that — when looking to prioritise the transport budget — it would seem wise to cull transport investment in areas that will not accommodate much growth. I realise this is a bitter pill for some, but the reality is that in the context of limited transport budgets, we should seek to reduce the costs of travel between locations that are able to accomodate development.
I note this finding is general insofar as it extends beyond walking and cycling. Public transport fare policies that reduce the costs of long distance journeys, for example, will increase the demand for those journeys by virtue of their effects on location choice. And if we are to extend rail services to regions south of Auckland, then this investment should be accommmodated by supportive land use policies. I’d go as far to say that such policies should be a pre-requisite for rail service. And if towns aren’t able or willing to support land use development, then perhaps they should be dropped from the network?
Second, my results suggest that improvements in walking and cycling infrastructure have significant effects on location choice, especially where journeys are short. The implication is that I am more excited about the effects of pedestrian access to tunnels under Albert Park than, for example, the SkyPath walking and cycling connection over the Harbour Bridge. Now don’t get me wrong, I think the latter is desperately needed, and is likely to be well-used. However, I don’t expect many people will re-locate to the North Shore because of SkyPath, mainly because the journey is, for many people, a relatively long one. On the other hand, the Albert Park tunnels will shorten walking and cycling journeys between the city centre and Parnell, which are already well within walking and cycling distance.
Third, in congested urban areas even small-scale changes in walking and cycling travel-time, can – at the margin – have relatively large effects on location choice. A one minute saving could arise from changes in the phasing of traffic signals, for example. This effect works in both directions: By reducing walking and cycling connectivity, the Central Motorway Junction may have increased walking and cycling travel-times and reduced the relative attractiveness of Newton. If we now want these areas to intensify, perhaps we should look to mitigate the effect of the CMJ as best we can?
What do you think? Comments, criticisms, and witty lyrical waxing are — as always — welcome.
Shameless, self-promoting postscript: Some of you may be wondering what comes next with my research. The short story is that I’m currently trying to secure funding for a PhD, in which I hope to extend and build on this research — among other things. If you know of any universities in New Zealand and/or Australia who might be keen to support my research, then please put them in touch with me. I’m relatively flexible, and open to propositions. Ka kite.
Processing...
Great to see lots of this kind of research going on, and the land use change implications are for me a hugely important missing element of transport modelling.
I’m intrigued by your second observed implication around short route impacts, and wonder if that’s an area for further research?
Clearly shorter commutes are the more likely territory of walkers and cyclists and offer big % journey time wins, but I suspect time savings from upgrades to “backbone” bike routes might also give rise to increased cycle commuting and location choices over much longer distances. This could be quite significant as an economic issue, as it _could_ affect much larger quantum within a city’s land use markets.
Auckland seems ripe for testing that hypothesis. Carrington-Avondale-New Lynn-onwards, Tamaki Drive-Glen Innes, and Waterview-Te Atatu-Lincoln-Westgate, coupled with the dynamics of house prices on the Isthmus will probably extend the spread of what have traditionally been considered “bike suburbs”.
Yes I’m also enthused by the degree to which economics research is gravitating towards sustainability issues.
In terms of the second finding, I think it’s important to carefully distinguish between volume effects and distance effects. Your comment on “backbone” routes seems to me as if you are talking about routes that have high volumes, such as SH16. Naturally, routes with high volumes will have more potential to attract demand.
But my findings are standardised for volume, which is why I talk about effects in percentage terms. I find that a marginal (one-minute saving) on a short route will lead to a larger percentage increase than the same saving on a longer route ceteris paribus.
However, the same percentage change can lead to vastly different absolute numbers depending on the volume. So you’re correct to note that volume is important, it’s just not what I was particularly interested in here. Does that make sense?
A current example is the success of the Northern Busway; a relative with a child at Rosmini School Takapuna is looking for somewhere to live so she will be looking for places walking distance to the busway stops (there are pupils commuting from Warkworth).
If PT is overloaded then farther out locations can be more attractive. My experience was the London Tube and living in Walthamstow at the end of the Victoria line meant I would always get a seat into town. Similarly living in Birkdale my student kids used to walk away from the city to catch a bus to the city because they would get a seat whereas in Highbury or Onewa Rd there were more buses but they could be full.
Great research Stu.
This type of spatial general equilibrium model seems well suited to predicting land use responses from transport improvements. And if I’m reading the underlying work correctly, you’ve found evidence that this is a causal effect – ie improvements to walking/cycling conditions will *cause* more people to live or work in the affected locations.
I had a couple practical questions. Question 1: Have you incorporated any productivity responses (eg increased wages) from improvements to accessibility? Or are these controlled out / held fixed in the econometric model?
Question 2: You’ve modelled a land use response from walking/cycling improvements – both modes that are unlikely to get congested with increased demand, at least in the current context of Australian cities. Would this type of model be as applicable to car travel times (where peak congestion matters) or PT travel times (where vehicle crowding is an issue)?
Thanks Peter.
Yes *in general* I think this kind of SGE modelling framework is suited to predicting the causal effect of transport improvements on land use.
The devil is in detail, as your follow-up questions note.
To answer them:
1) Productivity related effects, such as increased wages from agglomeration economies, are implicitly captured in the location-specific fixed effects (ASCs), which I include for home and work locations. Here I don’t model wages endogenously but — to the degree that walking and cycling travel-times affect wages and in turn location choice — they are controlled for. You could take my results one step further and look to explain the employment ASCs as functions of wages and other productivity policy related parameters that were of interest. I hope to do this in the PhD.
2) Car travel-times: The short answer is yes. Car travel-times are endgeneous, but I present a method for controlling for endogeneity (using an instrument and a control function), so there’s no prima facie reason why you couldn’t use the same methods for car. The main limitation is data quality: Rather than assuming symmetric travel-times between locations (as I do here) you’d need to extract travel-time data for each direction. But that’s not a big problem …
Interesting stuff, Stu. What sorts of improvements in walking infrastructure were you able to record? Were you able to distinguish between the effects of different types of improvement (eg a new pedestrian bridge, a widened footpath, a pedestrianised area, new street trees, etc.)
One typo to fix: cycling is… twice as fast as cycling.
Thanks Heidi; typo fixed.
Here, I don’t explicitly measure infrastructure quality or improvements. I use a cross-sectional spatial model, which estimates the effects of walking/cycling travel-time on commuting flows bu comparing across locations, rather than across time. I hope to extend the models to include temporal analysis in subsequent research.
Infrastructure quality would also ne nice to have, but data on this is hard to get in a systematic way.
Thanks Stu, yes I should have realised this. I’d imagine research looking at differences in infra quality and temporal changes would have to be much smaller in geographic scope.
Would there be any point in someone producing some guidelines for what would be the best data to record on these more local studies, so that it could more easily be incorporated into a larger study later?
You were looking at commuting – so were the car and cycling travel time graphs for peak hour? Which demographic for the cycling fitness? On the isthmus, I’d imagine Auckland’s figures would be different at peak hour for an able-bodied confident cyclist – with much lower cycling to driving time ratios. Do you know if anyone is studying it here?
The example of landuse changes due to the Auckland Harbour Bridge is a good one, Stu, to show the deficiencies in NZTA’s modelling assumptions.
If, for the Harbour Bridge, the current NZTA modelling process had been used (which assumes one land-use scenario and applies it to both the with and without project scenarios) the planners would have looked like fools. Do NZTA get away with doing so for Mill Rd, EWL, WC because each road only causes incremental land-use changes, not whole scale changes like opening up the North Shore?
And what’s the situation for a new harbour crossing? Will they eschew the current process as deficient, or will they plug ahead with showing no landuse differences between with and without project scenarios? (And does Jonathon Coleman know?)
“And if towns aren’t able or willing to support land use development, then perhaps they should be dropped from the network”
Unfortunately that’s exactly the opposite. Central suburbs which are anti growth gets better and better public transport that they doesn’t deserve.
Therefore there is benefit to have a coordinated/merged management of AT, Panaku, and unitary plan’s department together so that we can have a transit oriented development.
Excellent work Stu. I am night shift jet lagged so cannot think clearly enough to add anything to the debate. But clearly this study is opening up eyes into different ways to see transport, housing and urbanisation in NZ and Australia. Good luck with finding support to continue this important study : )
This is an excellent piece of research and should be an integral part of deciding routes for cycleways/paths.
“A one minute saving on a 15 minute journey between two locations is predicted to cause a 1-8 percent increase in commuting flows”
The 700m longer Glen Innes to Tamaki Dr path is an increase of about 2 minutes by bicycle and 8 minutes on foot.
So that’s a 2-16% decrease in cyclists and 8-64% decrease in pedestrians.
https://www.greaterauckland.org.nz/2017/09/13/big-changes-tamaki-dr/
People don’t commute to the beach, main flow is to/from town.
Good point about the potential application for cycleway planning but I’m not sure your calculations in relation to Tamaki Drive are quite right.
Stu also concludes that the magnitude of the effect decreases with commute length. Most trips between the residential catchment for GI-Tamaki route and CBD will be over 15 mins. So I think it could potentially have an effect on those trips but not as pronounced as you’ve suggested.
“People don’t commute to the beach, main flow *is* to/from town.”
Key word; is.
Eastern Bays to Newmarket, Greenlane, Sylvia Park, Manukau, or airport are all shorter via this route and everyone riding from Orakei to town rides past a train station that will almost certainly offer a faster trip.
A fascinating read Stu and I would love to talk with you about this some time. Coincidentally I did some work years ago on walking trips to public transport stops in Brisbane and we found a marked effect near the river then too. People near the river were willing to walk much longer to PT stops, especially the ferry terminals. The riverside pedestrian connections were more direct with no road interruptions and we hypothesised the pleasant walking environment made people more willing to walk as well.
I have been reading on benefits from light rail projects and it is relevant. Some of the best performing French cities found that up to half the benefit from LRT schemes resulted from creating a more walkable environment along the LRT corridor, as opposed to the LRT services themselves. Mode choice shift to walking and cycling may have resulted in almost as much reduction in traffic as the LRT. I think this is very relevant to the prospect of running such a system through Auckland CBD.
Finally I was thinking that this is also important for centres outside the CBD. Places like Albany, Takapuna, Henderson and Manukau City all suffer in terms of walkability, having all been built when the car was dominant. They can tend to be a bit fractured, which limits their pedestrian activity levels and potential to really grow into full regional business centres. The obvious question is: can we tweak their design or traffic management to get better pedestrian and cycling connection times, to encourage more interaction between major campuses? That interaction might make these centres economically stronger.
I think any transport authority working on calibrating mode choice models including walking and cycling trips might be interested in your research. Calibrating a logit curve for choice between walking and other modes in short trips is often a weakness for such models. I don’t know about universities but Luis Ferreira at QUT/UQ used to do a lot of this sort of work.
Thanks, Scott. “we hypothesised the pleasant walking environment made people more willing to walk as well.” I’d be interested in reading your research, too, or would love to see links to the most recent research. Certainly in the cities I’ve lived, I’ve chosen to walk when I could do so along a canal, through a park, where the trees are inspiring, or along a street of interesting shops or activities.
Certainly – to come back to one of my bug-bears – removing mature, beautiful trees will decrease rates of walking, and shouldn’t be considered in the provision of cycleways. Instead, they should be supplemented by more trees to create a special boulevard which really invites walking. It’s the traffic lanes that need to be converted to cyclelanes, not the tree space.
Hi Scott! Good to hear from you and thanks for the hat-tip on QUT/UQ. And yes our findings so sound well-aligned.
The general message, I think, is that locations are — to some degree — substitutes, so if we invest in the transport infrastructure/services in an area then we can expect more people to live/work in the affected locations. And of course if that investment makes one mode relatively more attractive, then we can expect more people to use it.
The complementarity you note between PT and walking/cycling is something that I have also noticed when analysing Australian data: The two appear to be net (aggregate) complements rather than substitutes. More work on this is needed!