Risk Aversion and the Incumbency Advantage

Transcription

1 Risk Aversion and the Incumbency Advantage David L. Eckles Assistant Professor Department of Risk Management & Insurance University of Georgia Athens, GA Brian F. Schaffner Associate Professor Department of Political Science University of Massachusetts, Amherst Amherst, MA January 29, 2011

2 Abstract Explanations for the incumbency advantage in American elections have typically pointed to the institutional advantages that incumbents enjoy over challengers. In this paper, we propose an individual-level explanation for this advantage. Specifically, we predict that voters who are less tolerant of risk will be more likely to support the incumbent, who will likely be seen as the safer choice. Using a module to the 2008 CCES survey which provides us with an individual-level measure of risk tolerance, we find support for these expectations. We show that risk averse voters are more likely to vote for incumbents and that the effect is at least as strong as having a favorable casework experience with the incumbent. However, we also find that the effect of risk tolerance is contingent on the choices the voter faces. Increasing risk aversion leads to increased support for the incumbent, but only when the challenger lacks experience in elected office.

3 Incumbents running for re-election in the United States consistently perform better than they would if they did not presently hold the position they were attempting to keep. A substantial body of research has developed to help explain why this incumbency advantage exists; in general, these explanations are elite-centered. For example, some scholars have argued that congressional incumbents are successful at winning re-election because they work hard to satisfy their constituents while in office (Fenno (1977); Fenno (1978); Cain, Ferejohn, and Fiorina (1987)). Others have noted that the calculations of strategic politicians determining when to run for a seat also contribute to this advantage (Jacobson and Kernell (1983)). Still others have pointed to the ability of incumbents to raise more money than would-be opponents (Jacobson and Kernell (1983); Abramowitz (1991)). While elite-centered explanations for the incumbency advantage are more common than voter-based explanations, there is reason to think that the latter may play an important role as well. Ansolabehere, Snyder, and Stewart (2000) use the redistricting process as a quasiexperiment to find that approximately half of the incumbency advantage can be explained by a personal vote cultivated by the incumbent member. While the personal vote does not account for all elite-based theories, this finding does suggest that at least part of the incumbency advantage may come from characteristics of voters rather than the behavior of elites. This notion is further buttressed by the fact that an incumbency advantage is observed for many different types of offices (e.g. Ansolabehere and Snyder (2002)). In this paper, we analyze one potential voter-based explanation for the incumbency advantage. Specifically, we examine whether widespread risk aversion among the American public allows incumbents to receive more support from voters than they might otherwise (Quattrone and Tversky (1988)). Using survey data from the 2008 Cooperative Congressional Election Study (CCES), we show that voters who are more apt to act risk averse will over-value their incumbent relative to the challenger. In fact, we find that the most risk averse voters are over 30 points more likely to vote for their incumbent relative to voters 1

4 who are risk neutral; this effect is stronger than the increase in incumbent support generated by a favorable casework experience. However, we also show that the effect of risk aversion is contingent on the nature of the challenger. Risk aversion has a strong effect on incumbent support when the challenger lacks experience, but that effect is nullified when a more experienced candidate is running against the incumbent. Risk Aversion and Support for Incumbents Since Robert Erikson s seminal article in 1971, political scientists have amassed a substantial body of research studying the incumbency advantage (see Carson and Roberts (Forthcoming) for a review). 1 Early explanations of the incumbency advantage focused on the incumbent s use of office to maximize the likelihood of re-election through assisting constituents (i.e. pork barrel politics), generating increased general visibility, and better fundraising opportunities (Erikson (1971)). Further, Erikson (1971) points out that the incumbent, by definition, is likely to be a high quality candidate, and therefore more likely to be re-elected. Since Erikson (1971), numerous scholars have considered these rationales, and have put forth alternative explanations for the incumbency advantage. Hood and McKee (2010) provide a useful summary of these explanations: There are numerous (often complementary) explanations for the incumbency advantage: credit claiming, position taking, and advertising (Mayhew (1974)), constituency service (Fiorina (1977)), declining party attachments (Ferejohn (1977)), strategic retirements (Cox and Katz (2002)) and strategic challenger entry (Cox and Katz (1996); Jacobson and Kernell (1983)), and the declining ability of challengers to raise enough money for competitive campaigns (Abramowitz (1991)) 1 Erikson (1971) was not the first study to show an incumbency advantage. Erikson, himself, points to Cummings (1966), Stokes and Miller (1962), and Jones (1966) as early research, either directly or indirectly, examining the incumbency advantage issue. 2

5 (p. 346). In addition to candidate-specific explanations, several systematic explanations have been offered for the incumbency advantage. McKelvey and Riezman (1992) suggest that the incumbency advantage is an artifact of the seniority systems employed in legislatures. Recently, general changes in long-term (Ansolabehere, Snyder, and Stewart (2000)) and shortterm (Ansolabehere and Snyder (2002); Desposato and Petrocik (2003)) conditions (e.g. economic events) and redistricting (Cox and Katz (2002); Hood and McKee (2010)) have been examined as components of the incumbency advantage. Largely missing from the study of the incumbency advantage are individual-level (or voter-level) explanations. In particular, there has been relatively little work seeking to understand the psychology of the decision to vote for an incumbent rather than a challenger (though see Quattrone and Tversky (1988)). One possible mass-based explanation for the incumbency advantage is risk aversion. Economists have long studied how individuals deal with risk when making decisions. This research has demonstrated that individuals faced with risky propositions often act in irrational ways depending on considerations such as their wealth, the size of the potential gambles, or even the framing of the gamble (Kahneman and Tversky (1979)). Such fallibilities explain why we see an individual buy an insurance policy to protect his home in the event of a fire and also engage in casino gambling. The former behavior is one of a risk averter attempting to remove the risk (at an unfair price) associated with his home burning down while the latter behavior is that of a risk seeker purposely engaging in an unfair gamble (again, at an unfair price). While the influence of risk on decision making has been extensively studied in the economic and public health fields, individuals must deal with risk in nearly every aspect of their lives, including the political realm. For example, citizens must consider the risk that a particular policy may not work as intended once enacted (or the risks involved in failing to enact a particular policy). Likewise, voters must consider the risk that a particular can- 3

6 didate will behave differently than expected once in office. The act of voting in an election where there is an incumbent places the voter in a situation where one choice may appear to be more risky than the other. In this situation the voter has experience with how the incumbent operates when holding that office, but the voter generally must make an uncertain prospective judgement about how the challenger will behave if elected. Notably, risk aversion has been promoted as an explanation for an incumbency advantage in non-political decisions. For example, Muthukrishnan (1995) conducted a series of experiments to determine the factors that led people to chose a new consumer product over the one they were currently using. The study found that nearly 40% of subjects who stayed with their incumbent brand did so despite reporting that they thought the challenging brand was superior. The author cites risk aversion as the likely reason that subjects would maintain loyalty to their brand despite recognizing that there appeared to be a better alternative. Of course, not all individuals are equally averse to risks. Indeed, a growing body of research recognizes that some people are more tolerant of risks than others and attempts to measure these differences. As Ehrlich and Maestas (Forthcoming) point out, orientation towards risk is, at least in part, an underlying trait rather than wholly dependent upon context or framing (see also Filbeck, Hatfield, and Horvath (2005); Kowert and Hermann (1997); Cesarini, Dawes, Johannesson, Lichtenstein, and Wallace (2009); Nicholson, Soane, Fenton-O Creevy, and Willman (2005)). These differences in risk orientation appear to have consequences for whether citizens participate in politics by voting (Berinsky (2000)) and the positions that they take on political issues (Ehrlich and Maestas (Forthcoming); Kam and Simas (2010)). Generally speaking, risk averse individuals tend to favor policy options that are framed as less risky. The fact that not all individuals are equally susceptible to risk is significant for two reasons. First, if risk aversion is an important determinant of incumbent support, then most existing individual-level vote choice models are under-specified. Second, the fact that orientation toward risk varies across individuals provides us with the ability to 4

7 examine the extent to which risk aversion is a factor in the incumbency advantage through the comparison of risk averse voters with those who are risk neutral. Expectations We expect that risk aversion will influence an individual s propensity to vote for an incumbent. There are two reasons that this effect might occur. First, voters generally know more about the incumbent s positions and capabilities than they do about the challenger. 2 This added uncertainty about the challenger might lead a risk averse voter to favor the option he or she knows more about. However, existing research is mixed on this point. Some scholars have found that voters are less willing to cast their ballots for a candidate whose policy positions they are uncertain about (Bartels (1986); Brady and Ansolabehere (1989); Alvarez (1998)). However, more recent research has challenged that finding (Tomz and Van Houweling (2008); Berinsky and Lewis (2007)). Indeed, Tomz and Van Houweling (2008) find that subjects in an experimental setting actually favored the candidate who issue positions were more ambiguous. Even if uncertainty about the challenger does not generate more support for the incumbent, we still expect risk averse citizens to be more supportive of the incumbent. After all, a voter may be well-informed about which policies a challenger will support once in office, but there is still some risk that he will not carry through on those promises (or not be effective in pursuing those policies) once elected. Since a voter has already had experience with the incumbent, they will likely feel more confident about projecting what that candidate will actually do if she is re-elected. Thus, even after we account for the fact that voters will be less certain about the challenger s positions and traits than the incumbent s, we still expect to find that risk averse citizens will prefer incumbents more than those who are risk neutral. 2 In the dataset we analyze for this project, respondents were almost twice as likely to place the incumbent on the ideological scale as they were to place the challenger, providing support for the notion that they were generally less certain about where the challenger stood. 5

8 Data and Methodology The data we analyze come from a module to the 2008 CCES. The CCES is a cooperative survey project that allows teams to purchase individual module surveys of 1,000 respondents. The survey was conducted via the Internet by YouGov/Polimetrix using a matched random sample design. A subset of respondents recruited for online surveys were selected by matching them on a set of demographic characteristics to a randomly selected set of individuals from the population of American adults. Propensity score weights for the samples were developed so as to ensure that the sample represents the demographic characteristics of the adult population as reflected in the 2004 and 2008 Current Population Survey. The 2008 CCES was conducted from October 8th to November 3rd for the pre-election survey and November 5th to December 1st for the post-election battery. Additional information about the sampling methodology and the total survey error for vote and other objective indicators is presented in the guides to each of the surveys, posted at Our analysis is limited to a subset of 649 respondents living in districts where their incumbent House member was facing a major-party challenger. The dependent variable in our analysis is simply whether the respondent voted for the incumbent (1 if they voted for the incumbent, 0 if they voted for a challenger). We control for a number of factors that are likely to influence an individual s vote for their incumbent. We include in the model two dummy variables one indicating whether the respondent shares the incumbent s party affiliation and the other indicating whether the respondent belongs to the opposite party from the incumbent. We expect that partisans are much more likely to vote for an incumbent from their own party and substantially less likely to vote for an incumbent from the other party. In addition to this variable, we also include a dummy variable indicating whether the respondent s incumbent House member is a Republican, to account for the fact that 2008 was a particularly difficult year for Republican candidates. Also included in the model is a 6

9 control for the respondent s evaluation of the national economy. The economic evaluation variable is a retrospective measure capturing whether the respondent thinks the economy has gotten much worse (coded 1) to gotten much better (coded 5). Voters may be less supportive of incumbents when they view the economy as performing relatively poorly. The final control variable in the model accounts for how closely the respondent follows politics. This variable takes on three values: very much interested, somewhat interested, and not much interested. We expect that respondents who are less interested in politics may be more reliant on the incumbency heuristic when casting their ballots for Congress. 3 Fortunately, the CCES also includes two measures that allow us to account for institutionalbased explanations for the incumbency advantage. First, respondents were asked whether they had ever contacted their House member or anyone in the House member s office. The 30% of respondents who indicated that they had contacted their Representative were then asked whether they were satisfied with that contact. About 70% of those who contacted their member reported that they were satisfied with that contact. We use these questions to control for casework effects with two dummy variables. The first dummy variable indicates whether a respondent had satisfactory contact with their member while the second variable indicates whether they had contact that they were dissatisfied with. We expect those who had favorable contact will be more likely to vote for their incumbent House member, while those who did not have contact or who had unsatisfactory contact will be less supportive of the incumbent. The second measure that allows us to account for institutional-based explanations is based on a question asking respondents whether they could recall any specific projects that their members of Congress had brought back to their area. Only 14% of respondents could 3 As a more objective indicator of political attentiveness, we also estimated a model that included a measure of how many factual questions about politics the respondent answered correctly. This variable failed to approach conventional levels of statistical significance and the other estimates in the model did not change substantially with its inclusion. Thus, for the sake of parsimony, we excluded this measure from the final model. 7

10 recall a project, but we expect this group to be more supportive of their incumbent than those who could not think of any projects the incumbent had brought back to the area. The variable we are primarily interested in is our measure of risk aversion. A number of different survey instruments have been created by political and social scientists to measure an individual s tolerance for risk (see Kam and Simas (2010) for a useful review of these approaches). In general, measures of risk tolerance take one of two approaches. The first approach is to rely on respondents to rate themselves with regard to how comfortable they are taking risks. For example, Ehrlich and Maestas (Forthcoming) ask respondents to place themselves on a scale that ranges from extremely comfortable taking risks to extremely uncomfortable taking risks. Kam and Simas (2010) use an index of questions asking whether they agree with statements such as I would like to explore strange places or I like new and exciting experiences, even if I have to break the rules. Since respondents are sometimes unwilling or unable to provide accurate responses to such self-evaluative queries, we rely on the second approach, which captures a respondent s orientation to risk by asking him/her to participate in a hypothetical situation where he/she chooses between a probabilistic outcome and one that is certain. The two-question battery we use was pioneered by Barsky, Juster, Kimball, and Shapiro (1997), and a similar question was asked on the 1996 Panel Study of Income Dynamics. Figure 1 presents the questions used to determine each respondent s level of risk tolerance. As the figure indicates, the two question survey instrument generates a four point scale of risk aversion ranging from risk neutral to high risk aversion. While the distribution of responses to our question is very similar to that reported by Barsky, Juster, Kimball, and Shapiro (1997), our respondents appear to be slightly more tolerant of risk. Barsky et al. report that 65% of their respondents fell in the most risk averse category, while we find that 53% of our respondents fall into that group. Barsky et al. find 13% of their respondents falling into the risk neutral grouping, whereas 17% of our sample falls into that category. 8

11 Figure 1: Risk Tolerance Battery On one hand, the question we use in this paper avoids potential complications related to asking respondents for a self-assessment of their orientations toward risk. On the other hand, there may be a concern that we are only measuring one s orientation to taking financial risks while people may be more or less tolerant when it comes to other types of risks. While we cannot fully address that possibility in this paper, we can address whether our measure has properties that are generally consistent with other measures that have been used in previous research. Kam and Simas (2010) find that measures of risk tolerance are usually negatively correlated with age and positively correlated with education and income. They also find that women are less tolerant of risk than men and respondents who are married are less tolerant than those who are not married. With the exception of income, our measure of risk tolerance is consistent with these patterns. Notably, income was uncorrelated with our risk tolerance scale, indicating that responses to the question about taking a financial risk were not influenced by the respondent s current financial situation. Finally, as we noted above, risk aversion may lead voters to be more supportive of the incumbent because they are more certain about where the incumbent stands relative to the 9

12 challenger. To account for this possibility, we create a measure of the amount of uncertainty a respondent has about the challenger relative to the incumbent. The CCES asked respondents to rate both candidates on an ideological scale and asked respondents to indicate whether they thought each candidate was honest, knowledgeable, and experienced. Don t know responses were frequent for these questions, particularly when it came to evaluating challengers. In fact, only 14% of respondents answered each of the four questions about the challenger while 46% did the same for the incumbent. To generate our measure of relative uncertainty, we subtract the number of don t know responses for the incumbent from the number of don t know responses for the challenger. Thus, a value of 0 on this measure would mean that a respondent was no more uncertain about the challenger than the incumbent, while a value of 4 would mean that the respondent gave a response to each of the four questions when evaluating the incumbent, but was not able to give a response for any of the four questions when evaluating the challenger. Over half of respondents received a positive value on this measure, indicating that they were more uncertain about the challenger than the incumbent, while 40% were equally uncertain (or certain) about the candidates. Fewer than 6% received a negative value on this measure, indicating that they were actually more uncertain about the incumbent than the challenger. Results Table 1 presents the results from two incumbent vote choice models. The first model excludes the measure of relative uncertainty while the second model includes this variable along with an interaction term to determine whether the effect of risk aversion is conditioned on uncertainty. Both sets of estimates are generated using sampling weights and the standard errors are adjusted to account for clustering by congressional district. The fist model correctly predicts 86% of the cases and the adjusted count R 2 of.65 indicates that it also offers a 10

13 substantial improvement over a naive prediction of incumbent support. The second model improves little over the first, with 87% of the cases correctly predicted and an adjusted count R 2 of.66. Table 1: Logit Estimates of Vote for Incumbent Model 1 Model 2 Variable Coef. SE P-value Coef. SE P-value Republican Incumbent Eval. Of Economy Attention to Politics Opposite Party than Inc < <.000 Same Party as Inc Satisfied with Contact Dissatisifed with Contact Recall Project by Inc Risk Aversion < Relative Uncertainty Risk X Rel. Uncertainty Constant Note: N = 649. Robust standard errors are reported. The Adjusted Count R 2 are and for Models 1 and 2, respectively. We begin by discussing the results from the first model in Table 1. As expected, a respondent s party affiliation has a strong role in determining his probability of voting for the incumbent. Holding all the other variables in the model at their means, when the voter affiliated with the same party as the incumbent, he or she had a predicted probability of.91 of voting for that incumbent. However, when a respondent affiliated with the opposing party, the probability of voting for the incumbent dropped to.19. The coefficients for the other control variables in the model (the respondent s economic perceptions, attention to politics, and the party of the incumbent) failed to attain conventional levels statistical significance. The two variables capturing the effect of an incumbent s institutional advantages had mixed results. The coefficient for the respondent s recollection of any projects his or her incumbent brought back to the district was negative and failed to approach traditional 11

14 Figure 2: Risk Tolerance Battery levels of statistical significance. This does not necessarily mean that bringing such projects back to the district fails to generate greater support for the incumbent, but actively recalling such projects does not generate more support for the incumbent. Casework, on the other hand, had a statistically significant and strong effect on support for the incumbent House member. Figure 2 presents the influence of these variables on the predicted probability of voting for the incumbent. 4 When an individual reported no contact with their Representative, the probability that she would vote for the incumbent was.66. A favorable experience with the incumbent increased the probability of support to.84, an increase of nearly 20 points. However, the figure also demonstrates that incumbents do not always benefit from casework. For the small percentage of respondents who contacted their Representative but were not satisfied with the response, the probability of supporting that House member s reelection campaign dropped to just.42. Most pertinent to our analysis is the effect of the risk aversion scale. The coefficient for this variable is statistically significant (p<.001) and in the expected direction, indicating 4 These estimates were generated while holding all other variables in the model at their mean values. 12

15 Figure 3: Risk Tolerance Battery that respondents registering higher levels of risk aversion are more likely to vote for the incumbent, even after controlling for other factors. We should note that this finding persists even after controlling for the demographic factors that are correlated with risk aversion (e.g. gender, income, gender, marital status, and age). 5 Figure 3 presents the predicted probability of voting for the incumbent depending on an individual s level of risk aversion. According to the predicted probabilities, an individual s risk aversion had a strong impact on their support for the incumbent. Holding all other variables in the model at their means, a risk neutral voter had a predicted probability of.44 of voting for the incumbent House member while the predicted probability for the most risk averse respondent was.77. Thus, risk neutral voters were more than 30 points less supportive of their House incumbent relative to the most risk averse voters. Not only does this finding provide initial support for the notion that risk aversion among American voters can help to explain at least part of the incumbency advantage, but the 5 We exclude these variables from the final model because they are not significant in the model and their inclusion does not alter our results in any meaningful way. When each of these demographic controls is included, the coefficient for the risk aversion measure is.309 and the p-value is

16 strength of the effect rivals, if not surpasses, that of casework. Consider that a risk neutral voter who did not have contact with her house member had a probability of.41 of voting for that incumbent; but this probability increases to.75 for a voter with the highest level of risk aversion. Given that more than half of voters were highly risk averse, incumbents gain a major benefit from risk aversion. However, if a voter is already risk averse, having a favorable casework experience with the incumbent increases the probability of supporting that incumbent by just 14 points. Thus, incumbents appear to benefit more from the public s risk aversion than they do from casework. While the results from the first model in Table 1 demonstrate that individual risk aversion has a strong effect on an individual s propensity to vote for the incumbent, it is not clear whether this effect stems from voters being less certain about the challenger s issue positions and qualifications relative to the incumbent or if risk averse voters prefer the incumbent even in the absence of such uncertainty. The second model attempts to make this determination by adding our measure of relative uncertainty along with a term interacting relative uncertainty with risk aversion. Notably, the coefficient for this interaction term is close to zero and falls well short of attaining statistical significance, indicating that the effect of risk aversion is not dependent on how uncertain a voter is about the challenger relative to the incumbent. Furthermore, the size and statistical significance of the coefficient for the risk aversion measure is essentially unchanged from model 1 while the coefficient for the relative uncertainty variable is also positive and statistically significant. Altogether, these coefficients indicate that voters are more likely to vote for the incumbent when they are less certain about the challenger and when they are more averse to risk, but that these effects are not interrelated. In other words, more risk averse voters are more likely to prefer the incumbent, even if they are no more uncertain about the challenger s ideology and qualifications than they are about the incumbent. 14

17 Accounting for the Choice Context The findings in the previous section indicate that risk averse voters (who comprise a majority of the electorate) are more likely to vote for incumbents than those who are risk neutral. This pattern appears not to be attributable to heightened uncertainty about the challenger s positions or qualifications; rather, voters may feel that the incumbent represents a less risky choice because they have experience with her in office and they do not know how the challenger would behave if elected. But if this explanation is correct, then it suggests that some challengers may fare better among risk averse voters than others. A challenger s background may be a particularly important factor in this regard. When a challenger has held (or currently holds) another elected office, voters may be more confident in making prospective judgments about what that candidate will do once elected to this new position. But when a challenger has never served as an elected official, there is more uncertainty about how he or she will behave once in office and voters may perceive the challenger as more risky. Jacobson and Kernell (1983) show that challengers with experience in elected office perform significantly better than those without such experience; part of this advantage may arise from the electorate s aversion to risk. To test this possibility, we re-estimate the model in Table 1 but add a dummy variable that equals 0 if the challenger has not held previous elected office and 1 if she has. In our dataset, just 16.4% of respondents lived in a district where an experienced challenger was running against the incumbent. The model also includes an interaction term between the challenger experience variable and the individual-level measure of risk aversion. As noted above, we expect that an individual s aversion to risk will be a stronger determinant of their support for the incumbent when the challenger lacks experience (and is therefore riskier). The estimates from this new model are presented in Table 2, but Figure 4 plots predicted probabilities to show the relationship between risk aversion, challenger experience, and support for the incumbent. The figure shows that the relationship between risk aversion 15

18 Table 2: Logit Estimate of Model Including Challenger Experience Variable Coef. SE P-value Republican Incumbent Eval. Of Economy Attention to Politics Opposite Party than Inc <.001 Same Party as Inc Satisfied with Inc. Contact Dissatisifed with Inc. Contact Recall Project by Inc Risk Aversion <.001 Experienced Challenger Risk X Experienced Challenger Constant Note: N = 649. Robust standard errors are reported. The Adjusted Count R 2 is and support for the incumbent House member is contingent on the challenger s experience. When an individual is voting in a race where the challenger has not held elected office, risk aversion has a strong influence on vote choice. Specifically, an individual registering the highest level of risk aversion is about twice as likely to vote for the incumbent as a risk neutral voter. However, when an experienced candidate is challenging in the incumbent, the effect of risk aversion on incumbent support is nearly flat and the confidence intervals around the predictions overlap. In sum, our analysis shows a strong relationship between a voter s aversion to risk and the probability that he or she will support the incumbent. In fact, the increased support that an incumbent receives from a highly risk averse voter is larger than the payoff from a favorable casework experience. This relationship between risk aversion and incumbent support persists regardless of how much the voter knows about the challenger (relative to the incumbent), but it depends on the choice context. Risk averse voters prefer an incumbent to a challenger that has no experience in elected office, but the incumbent gains no advantage among risk averse voters when the challenger does have this experience. 16

19 Figure 4: Predicted Probability of Voting for Incumbent Based on Risk Aversion and Presence of Quality Challenger Conclusion and Discussion Explanations for the incumbency advantage have traditionally focused on the institutional advantages that incumbents hold over challengers. However, our empirical results demonstrate the important role that risk aversion plays in bolstering incumbent office holders. Using our individual-level estimates we can hazard a rough estimate of how much of the incumbency advantage can be attributed to risk aversion among American voters. As a baseline, we used the vote choice model in Table 1 to generate predictions for each respondent based on their actual values on the independent variables. Assuming anyone with a predicted probability above.5 would have cast a vote for the incumbent, we estimate that baseline incumbent support among voters in our sample was 64.6%. We then adjusted the data so that every respondent was assumed to be risk neutral and we generated a second set of predicted probabilities. The second set of probabilities indicate that if everyone was risk 17

20 neutral, incumbents would have won 56.8% of the vote among respondents in our sample. Thus, incumbents performed about 7%-8% better than they would have if the electorate was risk neutral. 6 This difference is significant, particularly given that 18 House incumbents won re-election by a margin of less than 8% over their challengers in In a recent review of the literature on congressional elections, Carson and Roberts (Forthcoming) note that the idea that incumbent office-holders have an advantage over their would-be challengers runs afoul of many tenants of democratic theory. [...] If incumbent legislators are gaining an electoral advantage due to the rules of the game being skewed in their favor, then this would be especially problematic and require some form of change. Indeed, by focusing mostly on elite-centered explanations for the incumbency advantage, the institutional advantages enjoyed by incumbents become a frequent target for reformers seeking to make elections more competitive. Specifically, the incumbency advantage has been cited as an impetus for reforms like term limits, public funding of campaigns, and campaign finance regulations. Yet, our findings suggest that even in the absence of institutional advantages, incumbents would still hold an edge in their reelection campaigns because of widespread aversion to risk in the electorate. Importantly, however, our analysis also demonstrates that the extent to which risk aversion affects the public s support for incumbents is contingent on the choices they face in the voting booth. Risk averse voters favor incumbents more than risk neutral voters, but only when the alternative is an inexperienced challenger. This provides an additional explanation for the consistent finding that experienced challengers tend to fare better than those lacking experience, but it also suggests that the most fruitful approaches to undercutting the incumbency advantage in American elections will be those that encourage experienced politicians to challenger incumbents more frequently. 6 It is worth noting that a similar calculation for the casework variables indicates that incumbents would lose less than one percentage point from their vote totals if none of their constituents could recall any casework experiences. 18

21 Bibliography Abramowitz, Alan I., 1991, Incumbency, campaign spending, and the decline of competition in U.S. House elections, The Journal of Politics 53, Alvarez, R.M., 1998, Information and elections (Univ of Michigan Pr). Ansolabehere, Stephen, and James M. Snyder, 2002, The incumbency advantage in U.S. elections: An analysis of State and Federal offices, , Election Law Journal: Rules, Politics, and Policy 1, , and Charles Stewart, 2000, Old voters, new voters, and the personal vote: Using redistricting to measure the incumbency advantage, American Journal of Political Science 44, Barsky, R.B., F.T. Juster, M.S. Kimball, and M.D. Shapiro, 1997, Preference Parameters and Behavioral Heterogeneity: An Experimental Approach in the Health and Retirement Study*, Quarterly Journal of Economics 112, Bartels, L.M., 1986, Issue voting under uncertainty: An empirical test, American Journal of Political Science 30, Berinsky, A., 2000, Rolling the Dice on Election Day: Risk Seeking and Political Participation, Unpublished Manuscript. 19

22 Berinsky, Adam J., and Jeffrey B. Lewis, 2007, An estimate of risk aversion in the U.S. electorate, Quarterly Journal of Political Science pp Brady, HE, and S. Ansolabehere, 1989, The nature of utility functions in mass politics, Am. Polit. Sci. Rev 83, Cain, B., J. Ferejohn, and M. Fiorina, 1987, The personal vote (Harvard University Press). Carson, James M., and Jason Roberts, Forthcoming, Handbook of Congress. chap. House and Senate Elections (Oxford University Press). Cesarini, D., C.T. Dawes, M. Johannesson, P. Lichtenstein, and B. Wallace, 2009, Genetic variation in preferences for giving and risk taking*, Quarterly Journal of Economics 124, Cox, Gary W., and Jonathan N. Katz, 1996, Why did the incumbency advantage in U.S. House elections grow?, American Journal of Political Science 40, , 2002, Elbridge Gerry s Salamander: The Electoral Consequences of the Reapportionment Revolution (Cambridge University Press: Cambridge). Cummings, Milton C., 1966, Congressmen and the electorate; elections for the U.S. House and the President, [by] Milton C. Cummings, Jr (Free Press: New York). Desposato, Scott W., and John R. Petrocik, 2003, The variable incumbency advantage: New voters, redistricting, and the personal vote, American Journal of Political Science 47, Ehrlich, S., and C. Maestas, Forthcoming, Risk exposure, risk orientation, and policy opinions: The case of free trade., Political Psychology. 20

23 Erikson, Robert S., 1971, The advantage of incumbency in Congressional elections, Polity 3, Fenno, R.F., 1977, US House members in their constituencies: An exploration, The American Political Science Review 71, , 1978, Home style: House members in their districts (Scott Foresman & Co). Ferejohn, John A., 1977, On the decline of competition in Congressional elections, The American Political Science Review 71, Filbeck, G., P. Hatfield, and P. Horvath, 2005, Risk aversion and personality type, Journal of Behavioral Finance 6, Fiorina, Morris P., 1977, Congress, keystone of the Washington establishment / Morris P. Fiorina. Yale fastback ; 18 (Yale University Press: New Haven :). Hood, M. V., and Seth C. McKee, 2010, Stranger danger: Redistricting, incumbent recognition, and vote choice, Social Science Quarterly 91, Jacobson, Gary C., and Samuel Kernell, 1983, Strategy and Choice in Congressional Elections (Yale University Press: New Haven, CT). Jones, Charles, 1966, The Role of the Campaign in Congressional Politics. pp (Prentice-Hall: Englewood Cliffs, N.J.). Kahneman, Daniel, and Amos Tversky, 1979, Prospect theory: An analysis of decision under risk, Econometrica 47, Kam, C.D., and E.N. Simas, 2010, Risk orientations and policy frames, The Journal of Politics 72,

24 Kowert, P.A., and M.G. Hermann, 1997, Who takes risks? Daring and caution in foreign policy making, Journal of conflict Resolution 41, Mayhew, David R., 1974, Congress : the electoral connection / David R. Mayhew. Yale studies in political science ; 26 (Yale University Press: New Haven :). McKelvey, Richard D., and Raymond Riezman, 1992, Seniority in legislatures, The American Political Science Review 86, Muthukrishnan, A.V., 1995, Decision ambiguity and incumbent brand advantage, Journal of Consumer Research 22, Nicholson, N., E. Soane, M. Fenton-O Creevy, and P. Willman, 2005, Personality and domain-specific risk taking, Journal of Risk Research 8, Quattrone, G.A., and A. Tversky, 1988, Contrasting rational and psychological analyses of political choice, The American political science review 82, Stokes, Donald E., and Warren E. Miller, 1962, Party government and the saliency of congress, The Public Opinion Quarterly 26, Tomz, M., and R.P. Van Houweling, 2008, Candidate positioning and voter choice, American Political Science Review 102,