Empirical Analyses of Politician Accountability in the United States: A Regression Discontinuity Analysis

几乎所有关于代议制民主中政治竞争的经济理论的一个重要含义是，当选政治领导人的行动在某种程度上受到选民愿望的约束。许多这样的理论表明，即使政治家和他们的政党想要实施截然相反的政策，他们仍然会被迫寻求更温和的立场，因为采取极端的政策可能会有被投票赶下台的风险。通过这种方式，政治竞争被认为使对立政党的立场更接近“中间选民”。但是这个概念在多大程度上与美国的政治经济有经验上的联系呢？选民是否暗中支持政党的立场？我们能否量化他们在多大程度上支持或不支持，这是否取决于选民的特征，比如教育水平？为这些问题提供可信的答案受到两个基本问题的阻碍：1)很难衡量两个对立候选人立场的“接近程度”，因为只观察到选举获胜者的政策决定；2)候选人赢得选举的可能性——这取决于选民——可能导致候选人采取某些立场；但这些立场会影响选民的投票方式，进而影响候选人赢得选举的可能性，因此很难从现实世界的数据中理清因果关系。该项目利用美国众议院议员的立法投票模式数据和选举返回数据，通过利用美国选举制度固有的准实验来解决这两个问题。这个项目的方法是研究以微弱优势（比如，不到1%）当选的美国众议院议员的唱名投票行为。在某些假设下，这实际上产生了哪一个政党赢得国会选区的随机分配，结果是选举结果很接近。勉强当选的民主党众议员的平均唱名投票行为应该反映出勉强当选的共和党人的对手如果没有输掉唱名投票会如何投票。这种所谓的“回归不连续设计”策略也被用来检查众议员是否对连任概率的急剧变化做出反应——换句话说，评估未当选的风险是否会迫使政治家缓和他们的立场。实证分析表明，几乎没有证据表明这种反应的形式有强有弱，这表明，党派分歧的力量似乎超过了选民的趋同力量。美国的政治和选举制度在多大程度上反映了“人民的意愿”，这是公众非常关注的问题，也激励了那些主张在从竞选财务法到任期限制到国会重新划分过程等领域进行政治改革的人。评估美国联邦制度运作好坏的一种方法是评估选民是否通过选举过程迫使党派政治家及其政党寻求“中间”立场，并缓和他们的政策选择。事实上，很多关于这个问题的理论文献都试图描述选民暗中“约束”政客决策的机制。然而，关于这一概念是否与经验相关的证据远没有定论，这主要是由于社会科学家面临的标准问题，即在非实验环境中解开因果关系。这个研究项目的发现可以在我们理解美国政治和选举制度的运作方面带来两个重要的进步：1)量化导致公共政策结果的趋同（选民意愿）和分歧（党派关系）力量的相对重要性；2)展示了研究人员如何利用选举制度中固有的“自然实验”来了解政治和选举背景下的因果关系。

An important implication of virtually every economic theory of political competition in a representative democracy is that the actions of elected political leaders are somewhat constrained by the desires of the electorate. Many of these theories suggest that even if politicians and their political parties would like to implement diametrically opposed policies, they would nonetheless be compelled to seek more moderate positions because to adopt extreme policies would risk being voted out of office. In this way, political competition is thought to bring opposing parties' positions closer to the "median voter". But to what extent is this notion empirically relevant for the political economy of the United States? Do voters implicitly moderate parties' positions? Can we quantify the extent to which they do or do not, and does it depend on characteristics of the voting populace, such as education levels? Providing credible answers to these questions is hindered by two fundamental problems: 1) it is difficult to measure the "closeness" of two opposing candidates' positions since only the policy decisions of the winner of the election is observed, and 2) a candidate's probability of winning an election - which depends on voters - may cause candidates to adopt certain positions; but those positions influence how voters vote, which in turn impacts the candidates' likelihood of winning the election, making it difficult to disentangle cause and effect from real-world data. Using data on the legislative voting patterns of members of the U.S. House of Representatives and election returns data, this project addresses these two problems by utilizing a quasi-experiment inherent in the U.S. electoral system. The approach of this project is to examine the roll call voting behavior of U.S. House members that were elected by a slim margin (say, by less than 1 percent). Under certain assumptions, this generates virtually random assignment of which party wins the Congressional District, among the elections that turned out to be close. The average roll call voting behavior of barely-elected Democratic Representatives should represent how the opponents of the barely-elected Republicans would have voted in roll call votes if they had not lost. This so-called "regression discontinuity design" strategy is also used to examine if Representatives respond to sharp changes in the probability of being re-elected - in other words, to assess whether or not the risk of not being elected compels politicians moderate their positions. The empirical analysis indicates very little evidence of strong or weak forms of this kind of responsiveness, suggesting that the diverging forces of partisanship seem to outweigh the converging forces of voters. The extent to which the political and electoral system in the U.S. operates to reflect the "will of the people" is of great concern to the public at large and motivates those who advocate political reform in areas ranging from campaign finance laws to term limits to the Congressional re-districting process. One way to evaluate how well the U.S. federal system functions is to assess whether or not voters, via the electoral process, compel partisan politicians and their parties to seek a "middle" ground, and moderate their policy choices. Indeed, much of the theoretical literature on this issue attempts to describe mechanisms by which voters can implicitly "discipline" politicians' decisions. However, the evidence on whether this notion is empirically relevant is far from conclusive, mostly due to the standard problem faced by social scientists, that of disentangling cause and effect in non-experimental settings. The findings of this research project can lead to two important advancements in our understanding of the operation of the U.S. political and electoral system: 1) a quantification of the relative importance of the convergent (the will of the electorate) and divergent (partisanship) forces that lead to public policy outcomes, and 2) a demonstration of how a "natural experiment" inherent in the electoral system can be used by researchers to learn about causal relationships in political and electoral contexts.