Heterogeneity and Representation Reconsidered: A Replication and Extension Benjamin G. Bishin University of Miami Christopher D. Dennis California State University at Long Beach This paper replicates and extends the article “Heterogeneity and Representation: The Senate and Free Trade” which appeared in the American Journal of Political Science (42: 524-544) in which Michael Bailey and David W. Brady argue that legislative representation is dyadic. Constituency matters in homogeneous states but is less important in heterogeneous ones. The implication is that scholars who fail to disaggregate states by heterogeneity conflate the two distinct types of representation that occur and reach conflicting results. We replicate the authors’ statistical analysis and confirm their results. However, disaggregating the votes used as the dependent variable shows important differences. While the results for homogeneous states remain the same, the results for heterogeneous states change. More specifically, constituency influence varies across votes. We suggest that the limited finding for dyadic representation only in homogenous states is premature. The results seem as likely to stem from measurement problems as from differences in the representation process. The authors’ conclusions are likely understated. The real impact of accounting for heterogeneity is probably even larger than the authors suggest. 1 Introduction An understanding of the manner in which elected representatives reflect constituents’ preferences is central in evaluating to democracy efficiency. In an important article, Bailey and Brady (1998) suggest that constituent influence on trade issues varies depending on the heterogeneity of a legislator’s constituency.1 Specifically, the authors find that in homogeneous states constituency self-interest variables drive senatorial behavior. Homogeneous states produce less diverse constituency cues and, hence, clearer signals to their representatives. In these states, senators vote in agreement with the dominant constituency cue. However, the authors find a less clear role for constituency in heterogeneous states. In heterogeneous constituencies, senators are more likely to be given conflicting cues. Since a senator cannot simultaneously vote in accordance with conflicting cues, they are more likely to vote on the basis of their own ideology. Bailey and Brady conclude that dyadic representation occurs-- constituents in homogeneous districts receive representation, constituents in heterogeneous districts who do not share the legislator’s ideology, are less well represented.2 1 We thank Michael Bailey for both his data and his exceedingly helpful comments. We would also like to thank David Karol, and Vince Hutchings for their comments on this paper. 2 The term dyadic representation refers to the idea that legislators directly represent he preferences of their states and districts (e.g. Weissberg 1978, Hurley 1982). However, one implication of Bailey and Bradys’ results is dyadic in that it occurs directly in homogenous states and indirectly in heterogeneous ones. In this paper, we rely exclusively on the former description.2 In this paper we replicate the authors statistical analysis and confirm their results. However, the authors’ findings depend on their aggregation of individual trade votes into indices. Disaggregating the votes used as the dependent variables show important differences. We argue that the finding that constituency influence is less important in heterogeneous constituencies results partly from vote aggregation. When the votes are disaggregated, constituency variables are more important in heterogeneous constituencies than the authors analysis suggests. Consequently, a finding for dyadic representation only in homogeneous states, a direct implication of Bailey and Bradys’ statistical results is premature. Indeed, the authors’ conclusions are likely understated. This replication proceeds as follows: first, we summarize Bailey and Bradys’ methods and conclusions. Then we replicate their statistical results and identify a potential problem with their dependent variable. Extending these results, we show that their findings may depend on the construction of their dependent variables. We conclude that since there is no theoretical justification for the operationalization of their dependent variable, their findings are premature. 2 Heterogeneity and Representation Reviewed Examining why constituency variables are typically insignificant in roll call voting studies, Bailey and Brady (1998) fill an important void in the congressional voting literature. Bailey and Brady suggest the reason is that constituency effects are easier to determine in some situations than others. Specifically, they suggest that constituency effects are easier to discern as opinion within the legislator’s re-election constituency becomes unified. Thus, as the number of conflicting cues within the legislator’s re-election constituency increases, it becomes difficult to discern the impact of constituency. To test this expectation, Bailey and Brady use Senate votes on international trade: the North American Free Trade Agreement (NAFTA) and the General Agreement on Tariffs and Trade (GATT). They form an index of support for both NAFTA and GATT (i.e., the two votes on NAFTA form one index while the three votes on GATT form a second index). Since these issues received great publicity and there are state-wide variables that reflect the probable impact of trade on constituencies that are likely part of a senator’s re-election constituency, they offer a good opportunity to examine the impact of constituency on senatorial voting. Bailey and Brady (1998) test the hypothesis that the impact of constituency on legislator behavior is easier to discern the more unified the legislator’s re-election constituency, in two ways. First, since some constituencies are traditionally associated with certain political parties, one should expect a model that interacts the party specific constituency with party will find that the interactive effect matters. Bailey and Brady’s results offer strong support for this hypothesis. While some constituency variables, as well as senator ideology and party are important in explaining both NAFTA and GATT votes, the addition of an interaction term between party and constituency (union strength