Effect of analytical units and aggregation rules on mode choice models

Abstract

The quality of mode choice prediction is essential on the way to a resource efficient transport system. Commonly, mode choice is modelled on the trip or tour level, but a systematic evaluation of the definition of appropriate analytical units and aggregation rules is missing so far. This is in analogy to the ’modifiable areal unit problem’ (MAUP) in statistical geography, in the sense that the definitions applied can effect the result. In order to find evidence for this issue with respect to mode choice, a choice set is generated on the basis of the Swiss Microcensus 2010 data, and multinomial logit-models are estimated for different aggregation levels. Results show that marginal substitution rates derived from coefficient estimates do not always show plausible values, which could be due to the known correlation between attributes in RP data. It can be concluded that regarding the trade-off between computation cost and prediction precision, the tour level might be the appropriate level for mode choice. It has to be noted however, that meaningful definitions will be subject to change with a more multi-modal behaviour.