The DOGEV Model
journal contributionposted on 05.06.2017, 02:53 by Fry, Tim R.L., Harris, Mark N.
At present, there appears to be no qualitative dependent model that can simultaneously account for data sets in which the variable of interest is potentially ordered but also has strong heterogeneity of the observed outcomes. This heterogeneity of particular outcomes, inherently attracts individuals to them, in addition to that determined by the individual's observed characteristics. An example of such unobserved heterogeneity would be brand-loyalty (or "captivity") in a model of consumer choice. Such heterogeneity of the outcomes, may well result in a pronounced multi-modal distribution of the variable of interest. This paper introduces the Dogit Ordered Generalized Extreme Value (DOGEV) model, which does account for both ordering and captivity (and/or multiple modes) in the data. In the spirit of Manski (1977), the DOGEV model combines a model for choice set generation with the Ordered Generalized Extreme Value model. We illustrate the model using three different empirical examples: a model of employment contract types; an inflationary expectations data set and; a survey of students' evaluations of teaching. These three examples are chosen as they represent different values that the additional ancillary parameters are likely to take in practice.