Logit models: smallest versus largest extreme value error distributions

Abstract

The general term ‘type I extreme value distribution’ underlying the logit model is not fully precise. Through a case study, this study compares two models based on the distribution – one with the smallest specification and one with the largest. Results show these two models are different.

Acknowledgement

Comments from Drs. Wiktor Adamowicz and Michele Veeman are greatly appreciated.

Notes

¹Function f and F are standardized versions of the distribution; i.e., the location parameter is set to zero and the scale parameter is set to one.

² Other variables, such as respondents’ attitudes, perceptions and demographic characteristics were also collected in the data. Given that the purpose of this study is not to address consumer preferences and willingness to pay, these variables were not included in the analysis.