Last year Matt wrote a post about NZTA’s Post-implementation reviews. These reviews are undertaken a few years after a project is completed and perform three functions:

  • To see how the outcomes compare what was expected before the project was built
  • To explain any variation in those measures
  • To identify lessons that can be learned to improve other projects.

We have also written before about them here.

One example of an interesting finding, is from the post implementation review of the Victoria Park – Newmarket project, which is no longer generating any travel time savings. In fact, now travel times are worse.

This is a major problem because the biggest benefit in business cases for general traffic capacity projects is time travel savings. For some projects travel time savings make up nearly all of the benefit. If we are finding that projects are not resulting in these forecast travel time savings, then we need to be updating our models and our process for analysing projects.

These questions about travel time savings benefits have been around for a while. In a fairly ground-breaking 2008 paper, UK academic David Metz pointed out that while travel time savings benefits make up the vast majority of the expected benefits from road investment, actually average travel times are staying the same.

It should be possible to measure time saving if this is a significant part of the benefits to travellers of new investment in transport infrastructure. Travel time is measured in surveys of personal travel behaviour, typically using 7-day travel diaries. In Britain, for instance, average travel time (per person per year) has been reported since 1972/73 as one output of the National Travel Survey (NTS) (see most recently Department for Transport, 2006a). This household survey covers personal travel by residents of Britain along the public highway, by rail and by air within Britain, including walks of more than 50 yards. The most recent value of average travel time is 385 hours per person per year, or just over 1 hour per day. As indicated in Figure 1, this has changed rather little over 30 years, during which period car ownership has more than doubled and the average distance travelled has increased by 60%.

It seems as though this trend is also true in New Zealand, with changes in travel time per person over time mainly being due to modal shift (we walk less than we used to) rather than any actual decrease in overall time:

Metz points out that we should be surprised by these results, given the great promises of travel time savings in the business cases for all the transport investment we have been making:

These data on average travel time offer no obvious support to the idea that travel time savings comprise the dominant element of the benefits from investment in the transport system. Indeed, Figure 1 prompts the following question. What has happened to all the travel time savings claimed to justify public expenditure on British roads of around £100 billion over the past 20 years at current prices? One possible answer would be that had it not been for the time savings associated with this investment, average travel time would have been higher than it has been. The pattern of investment in road infrastructure in Britain over the past 20 years has shown marked swings in expenditure, between £3.5 and £6.4 billion per year (at constant 2004/05 prices) (Department for Transport, 2007) and, hence, in new capacity becoming available. The steady trend of travel time seen in Figure 1 shows no suggestion of a reflection of such variation in new capacity, and hence offers no support for the idea that average travel time would have been higher in the absence of new road construction.

An alternative interpretation of Figure 1 is that people take the benefit of investment in the transport system—private investment in vehicles as well as public investment in infrastructure—in the form of additional access to desirable destinations, made possible by higher speeds in the time available for travel. From this viewpoint travel time savings would be at best transient phenomena. Light might be shed on this possibility by empirical studies of travel time savings putatively associated with infrastructure investment, such as a new or widened road that has been built with the intention of generating such savings.

It is this important distinction between “saved time” and “better access” that sits at the heart of our need to update business case processes to actually match with the real impact of investment. While the two seem similar, there are many ways of improving access that don’t require building new roads – things like better land use policies that allow housing to be built in areas that already have a lot of access. Or investment in public transport improvements that improve the access of an area in a way that can be maintained over time and doesn’t get eaten away by induced demand.

Our strategy documents have caught up with this new paradigm. ATAP, the GPS and the Auckland Plan all emphasise the importance of accessibility. But the cost-benefit analysis process still lags behind with its outdated focus on travel time savings. This means we are continuing to have a situation where roading projects appear to score decent BCRs even though they will almost certainly not deliver long-lasting gains. Getting past this will require an acceptance of two key points:

  1. Increasing general traffic capacity does not result in beyond short-term travel time savings (due to induced demand);
  2. Many people will bank the time savings and simply travel further (known as the Marchetti Constant);

If we are to build the right projects we need to ask the right questions and update the way we measure benefits of transport investment. At the moment we are still asking the wrong questions when doing this analysis, which is going to undermine the efforts of our strategic documents to focus more on projects that generate long-lasting accessibility gains.

Share this


  1. Lie one: “It is industry standard that newly created trips are the least likely response to new infrastructure and it is not normal practice to include them.” [OIA response from NZTA] leads to:
    Lie two: “Travel times will decrease.”

    So Betteridge’s Law does not always apply! Thanks, Harriet, this is at the heart of bad transport planning.

    Is GA putting together a submission on the Draft Investment
    Assessment Framework? I’m imagining my submission will talk quite a bit about this lie.

  2. Even NZTA communications implicitly acknowledge this. You see new projects being described as delivering “better travel time reliability” rather than promising travel time savings. But obviously travel time savings are still being promised in the business case.

    1. Most of the other benefits in the business case for roads stem from the supposed travel time benefits, too. Here’s Mill Rd’s benefits table:

      When it is understood that induced demand will fill the roading capacity eventually, and increase traffic elsewhere as a result as well, the whole thing becomes a joke. There will be no long-term benefit due to any of those listed items. Agglomeration benefits? That one was really attempting to pull the wool over someone’s eyes.

  3. So, the Law that does apply then, is “Traffic always rises to fill what is available” and that seems to be happening in Auckland. Clearly then, there is absolutely no point in building new roads, as they will only fill up in short time, to be as bad as what is there now.

    Seems 100% logical to me then, that the only path to take is alternative means of transport, and especially Public Transport that does not travel on the same roads that those congested cars use. That equals Trains, Light Rail, or segregated high speed busways.

    End of story.

    1. Yep, that is indeed the logical conclusion. I’ve wondered for sometime whether it is now time to create new public transport infrastructure only, and to require car drivers to go cold turkey?

      1. What if AT took the $217 million that they spent on new roads last year and reduced all fares above two stages by up to $2 per fare, with no fare falling below the two stage fare? Would that clean the motorways and arterials out so that we don’t need more roads in the near future? Obviously the balance of the $217 million would be used for more buses, bus lanes etc.

  4. At last some real sense Mill Rd is a lie start to finish
    1 it wont stop or alleviate congestion it will create more as the northern outlet is already completely jammed with existing traffic at Redoubt everglade intersection.
    2 It will be another slow moving carpark at both peaks of the day
    3 The better would be to replace existing road with separate dedicated public transport and let single occupant cars get log jammed to get a behavioural shift to bums on seats of PT
    4 $500 will build a lot of kiwi build home which is a better use of the money.
    Good on you average human.

    1. 5 It will induce traffic throughout the city, for which we will have to pay more money to try to make our streets safer. But as they know in Denmark and the Netherlands, to make a city safe, you have to actually reduce the traffic volume.

    2. Yes the Mill Road ‘upgrade’ should consist of some road safety improvements.
      Then bus lanes each way.
      that’s all that’s required – then monitor to see how many people the corridor begins to move.

  5. Some might argue there is still the “benefit” of more traffic overall being accommodated on an expanded road system, even if it travels no faster than before.

    But my question is, what are the values of all of these journeys? How many are high-value “life-or-death missions”, and how many are completely trivial and not deserving of anything like the road investment that has been lavished upon them?

    If each journey required the payment of a toll commensurate with the cost of providing and maintaining the infrastructure, how many would be considered not worth making?

    Our road systems have all been built without consideration of these questions. Every car or truck trip is assumed to be of sufficient value not-only to justify the billions spent, but also to justify not pursuing alternative policies which might provide a better alternative to all these car and truck trips.

    The phenomenon whereby traffic can simply ‘disappear’ when roads become unavailable for any reason is a strong indication of the ‘optional’ nature of many journeys and of the folly of assuming they all need to be catered for by a continual programme of multi-$billion road-expansion schemes.

    1. I think your first sentence is getting closer to the crux of the issue. Certainly on NZTA projects I’ve worked on there’s been no denial that building roads induces traffic. And that has been reflected in the modelling. But it is very difficult for a transport organisation to accurately predict landuse and behavioural changes due to a transport project. And it’s been easy to just assume that, even if the travel time benefits eventually disappear in exchange for more people driving, the value of the travel time savings is a reasonable proxy for the other economic benefits gained by more people being able to travel. I think it’s that assumption that probably needs challenging. I thought this guy said it well…

      1. Chris, can you expand on this?: “Certainly on NZTA projects I’ve worked on there’s been no denial that building roads induces traffic. And that has been reflected in the modelling.”

        I’ve pulled apart the modelling for Waterview Connection. They do not include new trips resulting from the increased road capacity. They use the words “induced traffic” to mean trips that have been shifted from other modes or times, etc. When I questioned this, they confirmed it, and when I questioned if this was the case for all projects, said:

        “The Transport Agency is not aware of any models or projects that include newly generated trips in addition to other induced traffic.”

        Litman’s research shows quite clearly that the newly generated trips are the ones of significance, that the shift from other modes or times, etc, are only of slight significance.

        When you say that there’s no denial that building roads induces traffic, has this been as a result of fudged nomenclature? Have even the people working on the projects had the wool pulled over their eyes and been convinced that the models include the effects of induced demand?

  6. some of these new roads would be acceptable if the old ones went on a road diet, less lanes more cycle/pt. Hutt Rd in Wellington is finally going on such a diet, forty odd years after the motorway passed thru next door.

  7. I agree and disagree with David Metz.

    1) I agree that people’s trip lengths will lengthen in response to additional capacity. I.e. I might go to a shopping mall that is further away when a congestion bottleneck is removed. This is both a travel time budget issue (I have time to travel that far) and calibration of a trip length distribution to generalized cost issue (models assume a constant function – it would be an interesting research topic to determine based on real costs whether this is true over a long period of time).

    2) I disagree in that, in the case of no population growth – no capacity projects would be justified at all unless they are residual network deficiencies from previous population growth. Once they are resolved there should be no further induced traffic growth beyond what the transport model shows assuming its well calibrated as provides a reasonable approximation of reality.

    3) In the face of massive population growth e.g. Bangkok the transport networks do grind to a halt – people were travelling for four hours in the morning peak to get to work. Discretionary trips don’t happen & trip lengths will be short as the generalized cost of the delays is so high.

    4) In the case of average growth rates networks will continue to fill up again as we can never afford to build everything we need to, to remove all residual congestion & very few places have congestion tolling.

    5) There are two other key issues:

    5a) The real cost of travel continues to fall. GDP per capita keeps rising (proxy for income) and the real cost of cars keeps falling. The real cost of petrol is variable but broadly constant within a very wide range. This means that given travel is (assumed) to be a constant function of generalized cost trip lengths will continue to increase over time as we get continue to get wealthier. This is part of the longer term induced traffic effect.

    5b) Transport models in general still don’t use mesoscopic or microsimulation at the strategic level (there are some places that do & the trend is in that direction). The aggregate static transport assignment is very poor at showing the network effects of relieving congestion bottleneck.

    1. 2/ I’m not sure from your tense exactly what you mean but induced traffic is a separate phenomenon from population growth. Traffic volumes rise and fall according to the availability of road capacity.

      5a/ The real cost to society and the environment of travel continues to rise. The real cost to a pedestrian has been rising steadily with Auckland’s road building programme. The real cost to the planet has been rising due to carbon emissions and road runoff. The real cost to a teenager has been rising with reduced access to the city. So which real cost are you talking about? A car driver’s cost per km? That’s a pretty narrow view. It also ignores a whole heap of externalities.

      1. Heidi,

        I agree with you. We should be pricing externalities into the costs that people face when using the transport system.

        However most business cases do try to take them into account to some degree.

        One major issue I do have with the approach to transport business cases in NZ, is that as far as I know the still use real costs from the base year.

        This means that the real future increases in carbon are not captured. The expected future costs (circa > NZD200 – NZD250 nominal (still > NZD100 real at 25 years & 2% inflation)) are a magnitude greater than the current values (circa NZD21)

        Also we do not properly price in the health dis-benefits of road improvements (or low density land development). By decreasing congestion we increase trip length distributions. This means more and more trips are truly car captive (i.e. greater than 1-2km walk and > typical 5km cycle) & except for a few O-Ds eg CBD focused not viable as PT mode.

        In this respect one can consider the base case as a very high density city where everyone is able to walk and cycle efficiently between O-Ds and there are no cars (but some freight vehicles) vs the health dis-benefits of any car dependent modern city.

        This is no different than applying a social cost to crashes where the base case is assumed to be 0 crashes in terms of absolute social costs.

    1. But Grant, that is a disbenefit. It is that extra general traffic that is reducing safety for vulnerable road users and increasing carbon emissions and car dependency.

      1. Yes I agree but in itself an economic etc benefit of doing business from point A to B. This extra general traffic then should have all the relevant costs weighed against it. I would think this means if the business cases are done properly & account for everything properly that only some rural or quite outer area roads would pass. Most other urban areas would only pass active & public transport mode items or upgrades.

  8. When faced with 2 alternatives, only one of which will be implemented, there is no way of knowing what would have happened if the alternative had been followed. Thus, for this project to have been successful, it did not need to “shorten travel times”. What it needed to do was to “shorten travel times at the time the project is completed, compared with the times which would otherwise have applied had the change not occurred” . An entirely different proposition, and to my mind proves the value of this project without a doubt.

    1. Are you assuming that travel times would have just got worse, so that a project that keeps them the same or only a little poorer is a success?

      I’m not sure I agree with that, there’s little evidence to suggest that actually happens in practice. Most of the time travel times stabilize around a limit that reflects what people are happy to put up with, which generally maxes out at peak journeys of about an hour regardless.

    2. Are you referring to the Victoria Park to Newmarket project? That’s a pretty wild conclusion you’ve come to. You’ve fallen for the same fallacy – that the increase will happen regardless of whether the project is undertaken or not. That’s not true. We will see some residual increases due to the road building of the last few decades. But building new road capacity will have a huge effect. NZTA models that the number of trips will be the same regardless of whether the project is undertaken or not, and that is at the core of what’s wrong here.

      If we want to understand why pedestrian deaths and significant injuries have increased by 44% in 8 years, we need to understand induced traffic. The travel times might have ended up being the same on the Victoria Park to Newmarket stretch of motorway. But car and truck traffic volumes increased. And it didn’t evaporate at the motorway offramps! It spilled onto all of our roading network.

      And it killed and injured people.

      Mill Rd will do the same. SH1 and SH20 ‘improvements’ will do the same. New roads to greenfields sites will do the same. Penlink would do the same if it was ever built.

      How much money are we willing to waste to try to improve safety while we’re still creating the traffic that is the problem?

  9. Harriet another fine piece of writing, worthy of wider publication…

    Looking back over the evolution of the practice of transport modelling, I think it was not so much dishonesty as ignorance at work. Or perhaps an extrapolation too far. Transport models started out as ways to assist traffic engineers work out how traffic flowed around a road network. Their use for economic justifications to expand those road networks was a later application. There is always a danger in such exercises that the original limitations of an analysis technique get forgotten by those who apply it to something else. The “value of time” in transport models is a device to distribute traffic, no more. Using it as a measure of reality, rather than a neat mathematical trick, is dangerous.

    The other huge danger in traditional four step transport models is that they do not allow for changes in travel behavioural choice either with the introduction of new modes, or over time. So policies that achieve that cannot be easily measured.

    The original strategic transport models were developed (by an English analyst named Wardrop) to analyse route choice. They were never designed to analyse mode choice, or travel demand itself. Those are later additions to what became the classical four step model. Hence, an extrapolation too far.

    That being said, I think by now the problems with travel time savings and four step models are well known to both transport economists and traffic modellers. Their continued use implies a reluctance to “look under the hood” for fear their current work might be useless.

  10. Penlink is a similar work of fiction to Mill Road – over 90% of the benefits claimed for Penlink are mere travel time savings.

    The Productivity Commission report on Low emissions economy called out the many road projects that use mostly travel time savings as benefits as one that are suspect and that such projects will have to stop being pursued sooner than later.

    1. Incidentally there are some jurisdictions now where for project benefits they ask the proponents to do two calculations – one with all travel time benefits counted, and the other only including commercial vehicle travel time. That is, commuter travel time “saving” is excluded as a benefit. So if you can’t fix the models easily, fix the way you count benefits.

      There are other things this approach solves too. For example, is walking to the bus a dis- benefit, or badly needed exercise in a society with a growing rate of obesity?

      1. Thanks, SP. What do you know about the models themselves in other jurisdictions? Criticism of the four step model has been strong for a long time now.

        1. In France and a growing number of European countries they still use them, but only as a comparator for evaluating options, not to develop strategy. Their process is more top down and less model driven.

          In Canada and several US states they are having some success with more complex activity or tour based models (these can include a travel time budget) and land use (luti) models in an iterative process. I.e. feed in base land use and year one infrastructure and growth, get year two land use, feed that into model to get year three and so on. Vancouver has been doing that since the 90s. It requires a clear picture of what your long term infrastructure program will be and a predictable land use approval system. Even then the luti models tend to be the weak link.

          Not sure what UK is doing now but I could inquire. They take WEBs and Luti models seriously, but not sure about tour based models.

  11. Could be that the focus on travel time is the wrong measure to focus on. If any of you knew your economics (which I very much doubt) you would recognise the issue as being the application of Say’s Law, formulated by Say (a French economist) in 1803 (yes, 215 years ago).

    In short; supply creates its own demand. Build more roads and what do you know but people will use them! Bastards!!

    Travel times are used as a key metric because they are easy to measure and can be readily seen from users of a route. However, they are a poor proxy (I would say very poor) for the value that is created from a road or a roading network. If you are going to have a bunch of people in a place then you need to connect them to other people and roads are a part of that connection.

    Despite the overall view of roads as evil on this blog they are still fundamental communication devices. We use them to get places.

    The more difficult, but probably economically useful strategy, is to assess the impact of the costs of not building a road (or any asset really) and then working out whether the cost of the build is greater or smaller than the cost incurred by the road/asset not being there…

    Rather than what benefits we can dream up, however, dubious, focus on the costs avoided by not doing it.

    Just a thought

    1. This is a pretty good idea actually. Instead of planners constantly dreaming up “do minimum” counterfactual scenarios that are impossible to do accurately, you compare what happens if you do *everything except this project* and see whether it’s still OK.

    2. Fat Bastard says: “The more difficult, but probably economically useful strategy, is to assess the impact of the costs of not building a road (or any asset really) and then working out whether the cost of the build is greater or smaller than the cost incurred by the road/asset not being there…” I’d suggest that’s exactly what is done, where “costs incurred” with/without the project are measured in units of travel-time.

      When models (of any type) appear to generate poor predictions, I like to start by considering their underlying assumptions. Understanding when these assumptions may be invalidated can help you identify what may be giving rise to the observed errors. With regards to transport models, I find several behavioural (microeconomic) channels that are usually assumed to be constant that are in fact not. Specifically:
      — land use: is assumed to be independent from transport infra/services. LUTI models try and model this interaction, although their ability to do so is a moot point.
      — vehicle ownership: is assumed to be independent of land use and transport outcomes. So people still own the same number of cars even when PT/active modes are improved.
      — trip generation rates: is assumed to be constant when evidence suggests they are a function of income, technologies, accessibility, generalised costs of transport etc.
      — trip lengths (or more specifically, distribution) are typically assumed to be constant, although induced demand models do try to take this account in some way.
      — trip-chaining: not considered in most models, which means that they do not capture interdependencies between trip-legs that may respond to investment.
      — timing of trips: are assumed to be constant. So people cannot respond to better transport infra/services by changing their time of departure.

      There’s also the gnarly question of population preferences, which are assumed to be constant, when they may in fact be endogeneously determined within the system being modelled. This could occur in (at least) two ways. First, there’s the possibility for a learning effect, whereby investment encourages people to try PT and active modes and this subsequently affects their preferences. Second, there’s the possibility that investment to affect the composition of the population. For example, if Auckland improves its PT and active models then, at the margin, the city may attract people and businesses that prefer these modes to cars.

      So I think the main problem is that the behavioural (microeconomic) responses to transport infrastructure/services are more complex/nuanced than the models assume, especially in the medium to long run. If we were using the models for 5-10 year forecasts then I think we’d be OK. Problem is we’re asking them to generate predictions 30-40 years into the future, at which point the aforementioned assumptions seem likely to become more problematic.

      P.s. Take this comment with a grain of salt. As you note in your comment, it’s unlikely that any of us “know anything about economics.”

    3. I know a little economics and that you don’t understand Say’s Law. Supply creates its own demand for inputs to that supply and creates incomes for the suppliers who then spend in the economy. Yes, since using roads is free there will be a demand for that product, but it is not the case that Say’s Law is predicting that it creates its own demand for itself. Try building a road to nowhere and see how well your interpretation of Say’s Law explains its use.

  12. Newmarket viaduct was replaced because it was literally falling to pieces and its piers were bordering on lethal. Problem with the materials used that were porous and such like.

    However. Is it 800 extra cars a week, in Auckland, or NZ? Bad in any case, worse if its just Auckland. And this explains the issues and possibly why those doing the projections were so wrong.

    Rapid car uptake seems to match Aucklands rapid, almost unconstrained population growth, a brilliant sugar high for electable economic statistics, terrible for the rest of us.

    Someone tell the economic morons, MP’s Simeon Brown and Judith Collins and hopeful Northcote candidate, Dan Bidois, who might I add is putting the boot into Auckland Council about traffic issues.

    It appears road planners are trying to solve historic problems, not future proof!

Leave a Reply

Your email address will not be published. Required fields are marked *