Big data is not a replacement for behavioural models - Tom van Vuren

Thirty years ago, data on travel behaviour was difficult and expensive to come by, and computers were big, expensive and slow. Mathematical models were a necessary way of generalising the few observations we had into operational traffic analyses and patterns – such as congestion and delays – across the whole population and network.

Transport modelling pioneers developed these models in the 1960s, but because traffic and travel are dynamic phenomena covering large spatial areas, the models were complex and cumbersome. For example, they often contained feedback loops that oversimplified complex interrelations between choices.

Although models have advanced over time – including improvements to the representation of true passenger behaviour, speed of calculations and the user-friendliness of models – tools used to analyse and predict travel behaviour today are still very similar to their 1960s predecessors.
Some big data evangelists would like us to believe that modelling is on the way to becoming obsolete, due to the sheer volume of data now available. Why model when you can measure everything and create beautiful data visualisations showing large quantities of observations, spatially distributed and moving in time?

But simply observing is not particularly helpful for planning purposes. For example, it does not allow you to understand what-if scenarios and develop response strategies. Understanding is essential before a problem can be solved – and that is exactly the point of models: using what we observe to help us understand how the system under scrutiny works so that we can, if necessary, respond. Displaying how many people use the London Underground may help identify capacity problems, but we cannot begin investigating strategies to overcome them without a behavioural model.

Rather than replacing behavioural models in transport planning, big data will enhance them. We can now access unprecedented amounts of data at much faster speeds than ever before. We can now observe the travel choices of a much larger proportion of the population and see in real time how they actively respond to changing travel conditions. This empowers us to develop and estimate new, different and hopefully better models of travel behaviour.

This is our chance to start considering whether and how to replace the data-poor models of the past with the data-rich models of the future. Big data will not make modelling obsolete – it will make it even more exciting!