Dealing with Uncertainty in Travel Demand Forecasting

Sources of error

When forecasting for the medium to long term we need to be clear in acknowledging that we are dealing with an uncertain future. There are a number of factors which may contribute to errors and uncertainty in forecasts. Sources of uncertainty when forecasting travel demand include:

Demographic and socioeconomic input data: employment and population market segments; economic activity: impacts on drivers for freight and passenger demand.

Assumptions about physical networks (base year and future projects): observed counts may affect model calibration and validation; road and public transport network assumptions.

Values of travel time (VTT): strategic transport models use an imputed VTT for each main trip purpose. VTT are usually estimated from the coefficients of mode choice models.

Time of day effects: factors are usually used for different time periods, (peak, daily, weekly, monthly, and annually), based on historical data. Such factors will not deal with peak spreading or mixes in the traffic composition. Methods to model time-of-day choice explicitly are beginning be implemented in practice.

Optimism bias: Use should be made of strategic transport models to investigate the impact of proponent optimism bias on overall results. Patronage optimism bias should be examined using sensitivity testing.

Errors in strategic transport models

There are two types of errors in the outputs of such models, namely:

1. Measurement errors: these relate to inaccurate input data due to uncertainty in the forecasts of each of the inputs required to run the model for a given future scenario. For example there is significant uncertainty attached to 20 year forecasts for: social attitude changes; life style changes; employment patterns; funding levels; demographics; fuel price and levels of economic activity.

2. Specification errors: these relate to the fact that models are only an approximation of reality. Inaccurate assumptions made during the model building stage, as well as inaccurate model formulation may lead to specification errors.
The following figures shows the way in which model complexity and errors are related.

As model complexity increases (additional variables are included in the model), measurement error increases (there are more variables to forecast).

However, specification error is reduced as more variables are introduced in the formulation of the model.

The net result is that there is an optimum level of complexity beyond which the improvements due to better model formulation are more than offset by the increases in measurement errors.

Taking Advantage of Transport Models

Given that we are faced with an uncertain future, one way to take that into account is to undertake sensitivity analysis by varying the main inputs.

Another way is to devise a small number of future scenarios which cater for the likely range of uncertainties in each of the inputs.

We are usually interested in evaluating how a proposed strategy may perform under a number of alternative scenarios. The use of strategic demand models is well suited to the task of evaluating the robustness of a strategy under a range of likely futures.

The way in which modelling specification error and input data uncertainty are dealt with is one of the most important aspects of travel demand forecasting.

Risk assessment is very often dealt with by producing a range of outcomes to show the sensitivity of results to changes in each of the main inputs and modelling assumptions.

However, the risks associated with simultaneous changes in inputs/assumptions, are often not taken into account.

For example, changes in the central assumptions about population growth, level of economic activity, fuel prices, investment profiles for future road and public transport projects, are often varied one at a time.

Plausible combinations of a range of possible scenarios should also be presented.

Assessing how a strategy might perform under different scenarios can be seen as one example of sensitivity analysis.

In practice, the output from such sensitivity testing would be a better understanding of how vulnerable any particular strategy might be to changes in key input assumptions.

A strategy might be said to be robust if it is insensitive to changes in input assumptions.

References

Flyvbjerg, B. and COWI (2004). Procedure for dealing with optimism bias in transport planning. Guidance Document. Department for Transport, London.

Rand (2005). Uncertainty in traffic forecasts: Literature review and new results for the Netherlands. WR-268-AW (authors: Gerard De Jong, Marits Pieters, Stephen Miller, Andrew Daly, Ronald Plasmeijer, Irma Graafland, Abigail Lierens, Jaap Baak, Warren Walker, And Eric Kroes). Prepared for AVV Transport Research Centre, Rand Europe, The Netherlands.

guest post by Prof Luis Ferreira

Leave a Comment

This site uses Akismet to reduce spam. Learn how your comment data is processed.