Constituency influence: A generalized model and its implications for statistical studies of roll-call behavior

Abstract

This paper uses a simple analytical model to illustrate the consequences of statistical model misspecification. The first part of the article presents and extends an earlier model of constituency influence on legislative roll-call voting. Unfortunately, the implications of this extended model have no clear empirical interpretations. This fact is turned to advantage, however, by using the model to generate an artificial data set which is then analyzed within a design common to the area of roll-call analysis. Although the data theoretically represent nothing but constituency influence, the usual analysis does not uncover that fact. Moreover, the usual analysis may produce serious errors in conclusions about the relative influence of party and constituency. The importance of the paper lies less in the general example of model misspecification than in the application to a substantive area where misspecification probably is both pervasive and productive of mistaken inferences.

Additional details