Identity as a Latent Variable

• In Event: Comparative Politics Applications of Latent Variable Models

Abstract

Many standard approaches to measuring different forms of identity are problematic from a theoretical perspective. For example, though scholars often measure ethnic identity as a dichotomous indicator of the group with which an individual most closely identifies, individuals can identify with multiple ethnic groups, as well as different aspects of these groups; the extent of their identification may also vary. As a result, ethnic identity may best be measured as multiple continuous variables that are not necessarily strongly correlated. Latent variable models provide a means for developing measures of identity that closely match this theoretical understanding, aggregating over multiple observed or self-reported indicators of ethnic identity. However, the degree to which the additional complexity of latent variable models improve upon standard methods for measuring identity is unclear. In this paper, I use original survey data from the Moldovan separatist region of Pridnestrovie to explore the advantages and disadvantages of measuring different forms of identity as latent variables, focusing on 1) ethnic, 2) regional and 3) linguistic identity. Specifically, I use item-response theory models to aggregate responses to multiple survey items, which I then compare conceptually and empirically to their standard dichotomous counterparts.

Author

• Kyle L. Marquardt, University of Bergen