Interpretation of Decision Maker Attributes in Discrete Choice Models Simplified

• In Event: Measurement of Segregation and Partisanship

Abstract

In political science, we have many discrete choice situations that involve both the attributes of choice objects and decision makers. Regarding the latter, discrete choice models require the introduction of a normalization to ensure model identification, which is typically the definition of a reference category. Even though the selection of normalization is completely arbitrary and all types of normalization contain the same information, it has direct consequences upon parameter interpretation. Depending on what pairwise-comparison is considered, the estimates change in size, direction, and statistical significance. This makes interpretation challenging. This paper provides a simple solution that can be easily implemented: the use of the symmetric side constraint. Under the symmetric normalization, the model does not increase in complexity, and parameter interpretation refers to the geometric mean category instead of a pre-selected arbitrary reference category. For illustration purposes, I use a neo-Downsian probabilistic spatial voting model where voter choice is a function of spatial proximity and voter-specific nonpolicy attributes.

Author

• Ingrid Mauerer, University of Munich