A question often asked of modellers is “does it really make a difference to the final result whether you use technique A or B or C?” A common response is “I think so but I have not made the comparisons; however it can be argued on theoretical grounds that technique B is likely to be superior.” This paper documents the theoretical merits of seven econometric model forms and explores the empirical differences among them using a well‐specified set of explanatory variables associated with the mode‐choice decision for a cross‐section of Sydney commuters.
The empirical results suggest that the standard goodness‐of‐fit criteria (t‐values, prediction success and elasticities) provide an inadequate basis for selecting a modelling procedure from a set of non‐nested models which are estimated using different statistical methods (i.e. OLS and MLE). Simplified models are also a dangerous mechanism for screening explanatory variables for inclusion/exclusion in more appropriate models. There is a need for research to establish other criteria and tests for selection of a model form.