Campaign Contributions and Congressional Voting: A Simultaneous Probit-Tobit Model
构建了一个同时概率比-托比特模型，联合估计国会投票与利益集团捐款的关系，以解决内生性问题，发现捐款显著影响投票。
M /[ ANY citizens and scholars have been troubled by the influence that special interest groups appear to exert over policymaking as a result of their roles in the financing of political campaigns. These concerns inspired the enactment of a number of election law reforms in the 1970s, and have led to increasing popular support for public financing of electoral campaigns. Despite the public attention these issues have attracted, few social scientists have attempted to analyze the financial relationships between interest groups and policymakers empirically.1 One such study was reported by Durden and Silberman (1976), who included contributions from the AFL-CIO political action committee as an independent variable in an equation explaining congressional voting on minimum wage legislation. Their results indicated that campaign contributions significantly affected voting on that issue. The work I have done differs substantially from that of Durden and Silberman in two ways. First, I have chosen to examine issues which were of considerably narrower concern than the minimum wage issue. In the interest of simplicity, I examine issues of concern to just one or a few groups. I have also used a different econometric technique. If campaign contributions are actually endogenous (as seems plausible), the single equation estimation technique employed by Durden and Silberman is subject to a possible simultaneous equations bias. For each issue studied, I have therefore jointly estimated a two equation system explaining both votes on the bill and contributions from an associated interest group. The 'simultaneous probit-Tobit model which I use takes into account the dichotomous nature of the variable indicating a congressman's vote, the non-negativity constraint imposed on the contribution variable, and the possibility of correlation between the error terms for the equations explaining these two variables.2 Full-information maximum likelihood (FIML) estimates for the model are consistent and asymptotically efficient.