Seth J. Hill

Sources of Bias in Retrospective Decision Making: Experimental Evidence on Voters' Limitations in Controlling Incumbents

Are citizens competent to assess the performance of incumbent politicians? Observational studies cast doubt on voter competence by documenting several biases in retrospective assessments of performance. However, these studies are open to alternative interpretations because of the complexity of the real world. In this article, we show that these biases in retrospective evaluations occur even in the simplified setting of experimental games. In three experiments, our participants (1) overweighted recent relative to overall incumbent performance when made aware of an election closer rather than more distant from that event, (2) allowed an unrelated lottery that affected their welfare to influence their choices, and (3) were influenced by rhetoric to give more weight to recent rather than overall incumbent performance. These biases were apparent even though we informed and incentivized respondents to weight all performance equally. Our findings point to key limitations in voters’ ability to use a retrospective decision rule.

A Disconnect in Representation? Comparison of Trends in Congressional and Public Polarization

While it is widely agreed that Congress has polarized over the past 40 years, there is considerable disagreement about the extent of public polarization and its connection to congressional polarization. We present the first estimation of time series of polarization using the same method on the most comprehensive data for both the public and the Senate. With statistics of various definitions of polarization, we find little increase in the dispersion of views in the public from 1956 to 2012 but do find an increase in ideological sorting starting around 1980. The two time series bear little resemblance to one another with respect to divergence. Further, while congressional sorting exceeds that in the public today, we find that Congress has always been unrepresentative of the public. These results suggest that it is unlikely that changes in public preferences alone explain the widening gulf between the two parties in Congress.

Institution of Nomination and the Policy Ideology of Primary Electorates

Many hypothesize that the divergence between Democratic and Republican members of Congress is partly attributable to partisan primary elections. Yet most empirical evidence on the influence of primary elections finds small to no effect on member behavior. I argue that existing designs that compare members elected out of nomination systems with more open rules of access to members elected out of more closed systems rest on the crucial and untested assumption that more closed institutions lead to more polarized primary electorates. With survey opinions, turnout validated to voter files, and an IRT model of ideology, I characterize the preferences of Democratic and Republican primary electorates and general electorates in each House district in 2010 and 2012. To the extent that there is a relationship between primary ideology and closed primary institution, it is in the direction opposite that hypothesized. I then show that the primary electorate diverges from the general electorate in every House district and even from supporters of the party in the general election in almost every district, which is consistent with a centrifugal influence of primary voters. These results suggest that institution of nomination may not have a large influence on the type of voters who turn out, and that some other feature of nominating contests must be implicated in polarized primary voters.

Learning Together Slowly: Bayesian Learning about Political Facts

Although many studies suggest that voters learn about political facts with prejudice toward their preexisting beliefs, none have fully characterized all inputs to Bayes’ Rule, leaving uncertainty about the magnitude of bias. This paper evaluates political learning by first highlighting the importance of careful measures of each input and then presenting a statistical model and experiment that measure the magnitude of departure from Bayesian learning. Subjects learn as cautious Bayesians, updating their beliefs at about 73% of perfect application of Bayes’ Rule. They are also modestly biased. For information consistent with prior beliefs, subject learning is not statistically distinguishable from perfect Bayesian. Inconsistent information, however, corresponds to learning less than perfect. Despite bias, beliefs do not polarize. With small monetary incentives for accuracy, aggregate beliefs converge toward common truth. Cautious Bayesian learning appears to be a reasonable model of how citizens process political information.