Transport models and the Representation Void: How models represent the world

On most infrastructure projects, there’s not a lot of time to ponder the nuances and? meanings of modelling and forecasting. I think that it is important to understand not only our models and what they tell us, but also the effect of our own role in modelling. I’ve spent the last 42 years or so either directly or indirectly involved with transport demand models. Now as I ride into the sunset of my career, I’m trying to work out what the hell I was really doing and how I could have done it better. It is all taking a bit of a philosophical turn, from which I don’t resile.

I’ve become interested in what transport demand models represent in the world and how they do it.? Words represent things or actions. We understand words even when what they represent doesn’t exist. Everybody knows (spoiler alert) that Santa Claus doesn’t exist. Also, everyone also knows that he’s a portly gentleman, with a white beard, has a whim for wearing a red suit with a black belt, has a sleigh that is pulled by flying reindeer and has the mother of all logistical systems. That’s quite a detailed portrait of a man who doesn't exist. This knowledge tells us nothing about the real world though. If we say “Santa will leave from the North Pole on Christmas Eve” we’re saying nothing about the real world. In the end, our familiarity with Santa and his travels tells us more about ourselves than the world.

What our models represent and how they represent it

In our models, we create spatially accurate transport networks, residential areas and employment areas.? We build true-to-life household characteristics and employment types as far as we can.? We do all this so that the models represent the modelled city or region. There is a direct representational link between each road, bus service, train, intersection and their real-life counterparts. The relationship may not be perfect.? We may leave out some entities, like local roads and taxis.? Omissions and? errors that arise from imperfect data and reduced detail may spoil the quality of the representation.

Some of the model’s components don’t represent anything real. In agent based models, the agents are created randomly and? don’t represent real persons. There is no direct representational link between the model’s agents and people. Agents in a model are more like sprites or avatars whose movements and impacts can be aggregated to help the? model’s transport networks represent THEIR real life counterparts.?

In traditional four-step models, trips are the counterparts of agents and don't represent real people. The concept of a trip is deceptively simple. When we deal with, say 60 trips, it could denote, 20 people completing a journey 3 times each, or 15 people completing the journey 4 times each or any of many alternatives. Trips are gangs of movements that help the representation of real roads and public transport services.

Travel zones don’t exist outside of models. They’re just a convenient way of aggregating travel behaviour in a spatial way. We’re a bit hazy about the sort of day that? we model. We usually define it as a weekday in a school term, but we don't specify whether it’s in winter or summer, or whether it's an average or a typical day (which are not always the same).?

In the end, the representation may be nebulous but a city would usually be recognizable from its model, even if it’s in the same way that we recognise the subject of a caricature.

The representation void

The transport models’ phantom cities of the future are always conceived in the image of the current city. When we use our models to forecast, we strike a representation void. The transport networks of the future models contain new roads, stations, rail and bus services and so on. They don’t have the direct representative relationship that existing elements do. The new parts represent only concepts in human minds oron drawings. Forecast population and employment are like the current population and jobs but with more of them. Calibration parameters ensure that the future population makes transport choices in the same way as the original population. Jobs are in much the same industries as there are today. The average/typical day that we calibrated may turn out in the future not not to be an average/typical day.

In short, the forecast model represents, a “straw” city or region that is a construct of our speculation.? The Sydney Area Traffic Study of 1980 provides a clear illustration of this. The aim of the study was to? plan the provision of transport infrastructure for the Sydney Metropolitan area in 2000 with transport modelling underpinning the work. We know now that the model included some transport infrastructure that was not provided by 2000 and omitted some that was. Demographers who provided the model’s population and employment numbers did not foresee the demise of the manufacturing industry in Sydney and the rise of the technical and communications industries. As a result, forecasts of workers and employment bore little resemblance to those in 2000. Planners also missed the Sydney Olympics in 2000 and construction boom in the preparation for the games.? The future model for 2000 represented a phantom, an assumed city that was no more real than Santa's workshop. None of the forecasts were credible; the planners just didn’t know it. That was not their fault, though.

One of my early mentors warned me that if a model's output ever surprises me, it’s not giving me some deep insight, it’s telling me there's a mistake. In 2006 public transport patronage in Melbourne and Sydney unexpectedly started increasing. Studies found some reasons for it: the coming GFC, higher car and petrol prices and so on. Yet the size and nature of the change was never fully explained. Whatever the reason, no transport model forecast the change in mode choice. I suspect that had the models shown the change , we transport planners would have dug into the models and corrected “mistakes”. Or perhaps we’d have taken our findings to decision makers who, I have no doubt, would have laughed us out of their offices.

This is the credibility pincer we work within:

• Models forecast the present in the future, but everybody's comfortable with that
• In the unlikely event that models hint at an unexpected change, we fix them.

The representation void does not help with this because it doesn’t give us a solid foundation to back the veracity of our forecasts.

Modelling and forecasting are different activities. The problem of representation of future infrastructure is one for forecasting, not modelling . Most modelling studies compare the forecast transport conditions for two or more scenarios.? The problem is that these comparisons are merely comparisons of? fantasies.

?What can we do about it?

Almost any improvement in forecasting transport demand involves doing a lot more modelling, and this is no exception. We really need to bring our forecasting into the high-stakes, uncertainty-riddled 21st century. Making decisions using a single future is not good enough, even if? we test a few scenarios. We need to model several different futures to add credibility to forecasts. A future is not a scenario. Scenarios are about variations in whatever we can influence directly in our planning. For transport, scenarios only test changes in? transport networks. I think that we need to test several futures defined by:?

• Alternative technological futures (demand management, work from home strategies and infrastructure, path-finding):
• Economic alternative futures (fares, fuel prices, wages, value of time/willingness to pay)?
• Alternative societal futures (attitudes and priorities in travel choices)?
• Alternative demographic futures (distribution of population, employment, age, gender, household and family makeup)?

The clearest example I can find of a defined future is from a study in Johannesburg in the 1970s. The study looked? at the transport needs of the metropolitan area in 2005 at a time when the city faced many uncertainties: apartheid was still an unshakeable policy of the government, there was political unrest, economic difficulties, sanctions disrupted oil supplies and limited international trade. The planning team, with expert advice, set out the conditions for which they were forecasting.? The political and economic aspects of the future foresaw the end of apartheid and part of it looked like this:

The red-ringed assumption indicates the end of apartheid.

There are several advantages to this approach:?

• It facilitates Triple Access Planning, because it specifically lincludes spatial, movement and digital aspects of transport within a city
• It conforms to the principles of Design Thinking as it works from a single point (base year) to many futures and solutions that are whittled to an optimum solution for many futures
• It? acknowledges the possibility of alternative futures and? allows us to make probabilistic forecasts rather than single numbers or ranges. We could (and should) move to probabilistic forecasts like weather forecasters use and that “superforecasters” embrace. It will need more work, but will yield more meaningful forecasts.?

Most importantly, it provides us the opportunity to test futures that we normally wouldn’t, like the potential migration that may occur as people move to avoid the impacts of climate change. Like it or not, we also need to test catastrophic futures to provide resilience to the transport networks.?

I am confident that this approach would work. I had the opportunity on one project to model alternative futures, even if in a brief way. The approach lowered the forecast patronage to a third of earlier estimates. Our forecasts for the second year of operation were in the right ballpark but still 50% higher than? actual patronage. Previous forecasts had been nearly 350% higher.?

To finish

I think we sometimes overlook the truism that there are no facts about the future, only opinions. While we look to our models for evidence for our plans, the models only give us an outline of the eventualities of a story they “imagine” based on our inputs. We need to interpret these outlines to build up a picture of the best way to provide infrastructure for the future.?

It is risky and, in these days of uncertainty, unacceptable to plan future transport networks based on a few network and land use scenarios for a single set of future conditions.? The process of many futures will need more modelling and more analysis of the results than what we usually do. It’s also true that modelling many futures doesn’t make individual forecasts less fantasy than any of the others. In fact, we could say that we’re filling the representation void with rubbish, but at least we can cross the void on the rubbish that fills it. Somewhere in the multitude of futures are some that will resemble the future. We will have had the opportunity to choose and design the best set of infrastructure for that future.

More questions?

Here are some other questions I’m working to answer:

• Are there better ways of representing the world than the models we have?
• Can models actually teach us about the world and its future?
• Should we be using less detailed models that provide answers more quickly than the ones we use at the moment?
• Can we use our models to represent a future world that is very different to the one we live in now, where people live and work differently because of climate change, different technology and changed attitudes?