Electoral Vulnerability and Legislative Responsiveness

Electoral Vulnerability and Legislative Responsiveness Abstract This paper focuses on the relationship between legislators electoral vulnerability and their legislative behavior. I develop an alternative measure of electoral vulnerability: estimation based on several indicating variables including campaign expenditures, previous electoral margin, and district partisan support. Using this measure, I examine legislative responsiveness in the House during the 109 th and 110 th congresses, and find that electoral vulnerability is positively related to responsiveness. More vulnerable members better reflect constituency preferences in their legislative activities. Additionally, vulnerability plays an important role in individual roll call decisions. Vulnerable members are more likely to support bills when there is a bandwagon to jump on and shy away from bills that take place close to election time. This paper sheds light on our understanding of electoral vulnerability, finds evidence supporting the marginality thesis, and explores conditions under which vulnerability influences legislative votes. By Ruoxi Li Department of Political Science University of Wisconsin-Madison April 2013 Paper prepared for: American Politics Workshop University of Wisconsin-Madison

Introduction The relationship between electoral vulnerability and legislative behavior has important normative implications. Whether or not electoral vulnerability has direct influence on legislative outcomes is one of the fundamental questions of democratic representation. This topic first drew extensive scholarly attention in the 50's (e.g. Macrae 1952, Shannon 1968, Sullivan and Uslaner 1978, Cohen and Brunk 1983), and has experienced a revival during the past decade (e.g. Gulati 2004, Griffin 2006a, Gay 2007). No consensus in on this question has been reached. The best that can be said about the effect of electoral vulnerability on legislative outcome is that the results are mixed. This paper adds to this literature by first arguing for an alternative conceptualization and measurement of electoral vulnerability, then examining the effect of electoral vulnerability on legislative responsiveness, and also looking for conditions under which electoral vulnerability has a substantial effect on representatives' roll call voting decisions. First, I argue that electoral vulnerability can not be fully captured by the single variable of electoral margin. Electoral vulnerability is subjective and dynamic; I measure it as a latent variable indicated by several relevant factors including incumbent and challenger campaign spending, district partisan vote, seniority, and previous electoral margin. Second, I argue that electoral vulnerability does not have the same, universal impact on legislative decisions. Electoral vulnerability is likely to have a stronger effect on some bills and a weaker or no effect on others. The focus of previous scholarly debate is whether or not electoral vulnerability has an effect on legislative decisions; this is only a first step that needs to be supplemented by a more nuanced understanding of the 1

effect of electoral vulnerability. I hypothesize that the effect of electoral vulnerability is related to policy area, general level of bill support, and timing of the bill. The empirical analysis consists of three steps. The first step is to derive a new measure of electoral vulnerability. I use factor analysis to estimate an electoral vulnerability indicator for each House member of the 109 th and 110 th Congress. I find that most legislators are reasonably secure, with a few that are highly secure or highly vulnerable. The second step is to use the vulnerability indicator as an independent variable to explain legislative responsiveness. I find that vulnerable legislators follow district partisanship more closely with their own political ideology. Additionally, electoral vulnerability has a significant impact on about 12% of House roll call votes of the 110 th congress. The third step is to understand conditions under which electoral vulnerability has a substantive impact on legislative votes. In this step I let the coefficients of the vulnerability factor be the dependent variable, and find that level of bill support is positively related to the effect of vulnerability on legislative decisions. The findings suggest that it is too early to conclude that electoral vulnerably does not matter (e.g. Kuklinski 1977, Gay 2007) More effort is needed to better measure electoral vulnerability and to understand the conditions under which electoral vulnerability influences legislative voting decisions. Theoretical Considerations Literature review: an ongoing debate A repeating pattern in the marginality literature is the inconsistency between an intuitively straight-forward theory and the lack of supporting empirical evidence. The original marginality theory uses electoral margin to measure electoral vulnerability. Its proponents make the 2

argument that electoral margin is positively related to party loyalty and negatively related to constituency representation. Electoral margin, they hypothesize, [sensitizes the legislator] to the wishes of constituents in his quest for support at the next election (MacRae 1952: 1046). Early studies on the marginality theory find mixed results. On the one hand, Froman (1965) finds support for the marginality theory using congressional data from the 1960's. Bartlett (1979) finds that electoral margin is positively related to legislators' voting decisions on congressional pay raises. Kuklinski (1977) finds a limited, conditional relationship between marginality and policy responsiveness using state-level data from the late 1960's and the early 1970's. On the other hand, Cohen and Brunk (1983) find no relationship between marginality and party loyalty score using congressional data from the 1950's through the 1980's. Deckard (1976) similarly finds no relationship between electoral margin and party loyalty after controlling for district level demographics. In spite of the mixed empirical results, the concept of the marginality theory continues to play a role in explaining congressional voting behavior. Electoral margin is often included as a control variable explaining legislative voting decisions (e.g. Bartels 1994). The theory has traction because it makes intuitive sense and is supported by formal theoretical analysis. Anecdotal evidence too often points to the importance of electoral consideration in legislators' voting decisions. Controversial legislations are perceived to be situations where vulnerable legislators are in the unenviable position of having to choose between responsiveness and responsibility. News reports frequently quote legislators who cite electoral vulnerability as their main reason for refusing to support controversial bills. Campbell (1981) makes a formal theoretical argument showing that as long as voting decisions are perceived to be linked to 3

electoral results, 1 legislators will make voting decisions in accordance with their electoral margins. Thus, even though the marginality thesis has failed to gain sufficient empirical support, scholars are reluctant to conclude that electoral marginality does not matter in electoral voting decisions. A recent round of research brings the attention back to the marginality theory. The findings, again, are mixed. Griffin (2006) finds positive relationship between electoral competitiveness and incumbents' legislative responsiveness using congressional data from the 70's to the 2000's. Gay (2007) finds no relationship between district competitiveness and policy responsiveness in the context of majority-minority districts in the state legislature. Gulati (2004) even finds a negative relationship, that is, legislators from competitive districts are in fact less responsive to public opinion and more partisan compared to legislators from noncompetitive districts. The disagreement in regard to the marginality theory continues on with more advanced data and research methods. However, the problem with the literature may be less about data and methods but more about the conceptualization and measurement of electoral vulnerability. After all, electoral margin, albeit important, is only one way to conceptualize and measure electoral vulnerability. Electoral vulnerability: subjective and dynamic The core concept of the marginality theory is that electoral vulnerability matters in legislative voting decisions; those who are electorally vulnerable are more sensitive to voter preferences and more resistant to party pressure. A fundamental problem in the marginality literature is that electoral vulnerability is simply considered as legislators' vote share from the previous election, 1 Empirical research suggests that voting decisions are indeed directly related to election results, see (Bovitz and Carson 2006) 4

that electoral vulnerability equals electoral marginality. But it is not necessarily the case. Electoral statistics cannot capture the uncertainty members feel about their renomination and reelection. (Fenno (Fenno 1978):36) To conceptualize electoral vulnerability as marginality overlooks two important aspects of electoral vulnerability. The first is that electoral vulnerability is incumbents' subjective perception; the second is that there are possible discrepancies between vote margins from previous elections and electoral vulnerability in the current election cycle. To better measure electoral vulnerability, it is necessary to take into consideration its subjective and dynamic nature. As Fiorina (1974) points out, electoral vulnerability is a perception in the mindset of legislators. If electoral vulnerability affects legislative voting decisions, it is the subjective perception of vulnerability that matters. Few in the literature have attempted to measure the subjective perception of vulnerability. Some may argue that survey questions asking incumbent legislators about their perceived level of electoral vulnerability is a direct measure. The problem with this type of data is that they are difficult to obtain. Even if they are available for research, survey answers may not be a good indicator of electoral vulnerability because they can simply be cheap talk. Cohen (1984) shows that legislators across the board stated feeling electorally insecure regardless of their actual winning margin. These survey answers turned out to be uniform and lacked variance because they were costless and therefore unreliable signals. Legislators could easily overstate or understate their perceived level of electoral vulnerability without additional cost. Therefore, survey data, though seemingly straightforward, are not the best indicator of electoral vulnerability. I argue that an incumbent's campaign expenditure in the current election cycle is a useful indicator of the subjective perception of electoral vulnerability. The reasons are two-fold. First, 5

campaign expenditures are controlled by the incumbents and therefore subjective signals. When legislators perceive themselves as vulnerable, they are likely to raise and spend more money in election campaigns. Second, campaign expenditure is costly; it takes considerable effort to raise money and strategic consideration to spend money. Thus, large amount of incumbent campaign expenditure tends to reflect a sincere feeling of electoral vulnerability; it is too costly to be merely cheap talk. Another feature of electoral vulnerability is its dynamics. While many factors relating to electoral vulnerability, such as district partisanship, remain relatively constant from one election to the next, some important ones change across elections. For example, studies have shown that the emergence of strong challengers significantly affects election results and renders incumbents more vulnerable to losing (e.g. Jacobson and Kernell 1981, Jacobson 1990, Lublin 1994). Such dynamics can not be captured by electoral margin from the previous election cycle. While some scholarly work attempts to account for the dynamics of electoral vulnerability by including changes in electoral margin in addition to the absolute value of electoral margin (e.g. Cohen and Brunk 1983), these measures still fail to capture the dynamics of electoral vulnerability in the current election. Compared to electoral margin, which only indicates past electoral vulnerability, campaign expenditures towards an upcoming election indicate the current level of electoral vulnerability. In addition to incumbent spending, challenger spending is also an important indicator that captures the dynamics of incumbent's electoral vulnerability. It is both a reflection of incumbents' inherent electoral vulnerability and a contributing factor that increases incumbents' vulnerability. On the one hand, strong challengers emerge and spend significant amount of 6

money in election campaigns when the incumbent is weak; on the other hand, the emergence of and spending by strong challengers increases incumbents' vulnerability and decreases their chances of winning. Thus, incumbent campaign spending and challenger spending are important indicators that capture incumbents' electoral vulnerability. Some recent scholarly work has recognizes the limitation of electoral margin and uses alternative measures to capture district competitiveness. The alternative measures, however, are rather indirect; some are not even based on congressional elections. Gulati (2004) averages three indicators, presidential election returns, self-identified district partisanship, and congressional race returns, to measure district competitiveness. Griffin (2006) uses changes in previous presidential election results at the district level to measure competitiveness. These measures do not tap into the subjective and dynamic nature of electoral vulnerability; some attempt to capture competitiveness of congressional districts without using congressional election results. These measures are different, but not necessarily better indicators than electoral margin. Electoral vulnerability and responsiveness There is a long-standing debate in the literature about the effect of electoral vulnerability on legislative responsiveness. As discussed previously, recent evidence supports each of the three possible relationships between vulnerability and responsiveness possibilities. A positive relationship means support for the marginality thesis, that marginal members are more sensitive to constituency preferences (e.g. Griffin 2006). The lack of a relationship, in the context of a highly responsive legislative environment in recent decades, means that secure and marginal members represent their constituencies equally well (e.g Gay 2007). A negative relationship 7

means that intense competition may cause marginal legislators to be more responsive to their own partisans instead of to their district median (e.g. Gulati 2004). In this paper, I conceptualize and measure electoral vulnerability as a latent variable indicated by current campaign spending in addition to other relevant variables. With this measurement, the implications of the three possible relationships are somewhat different. A positive relationship will still imply that vulnerability contributes to responsiveness, as the reversed causal relation is unlikely. It is less plausible to argue that responsiveness produces vulnerability. The implication of a negative finding, however, can be different. If a negative relationship is found, it is possible that vulnerable members decide to appeal to partisans instead of district median, as discussed in the literature. It is also possible that legislators may try to buy their way out of out-of-step votes. In the latter explanation, legislators would first vote against constituency preferences, knowing that doing so may increase their electoral vulnerability, but then try to compensate by spending large amount of campaign for reelection in the hope that campaign spending may save them from electoral misfortune. This explanation argues for a reversed causal effect, that is, voting behavior leads to electoral vulnerability, instead of electoral vulnerability leads to voting behavior. When does vulnerability matter? The literature has largely focused on whether or not electoral vulnerability affects policy responsiveness. An important yet overlooked aspect of the effect of electoral vulnerability are the conditions under which electoral vulnerability has a significant effect. Although aggregate-level studies yielded mixed results, several individual case studies find significant effects of electoral vulnerability. Bartlett (1979) finds that electoral margin is positively related to legislators' voting 8

decisions on increase salaries for members of Congress. Jacobson (1993) finds electoral margin has a strong positive effect on support for the 1990 deficit cutting bill. Lanoue and Emmert (1999) find that multiple measures of electoral considerations had significant effect over the House members' decisions on holding hearings for the Clinton impeachment in 1998. The findings from the analysis of individual pieces of legislation suggest that it is highly likely that certain conditions highlight the importance of electoral vulnerability in the legislative decision-making process. In this paper I hypothesize that three variables, issue area, timing of the bill, and general level of support for the bill, are relevant to the effect of vulnerability. It is possible that bills in the category of government operation, such as the case of the congressional pay raise bill, are affected by electoral vulnerability. Governmental operation bills tend to only affect political institutions; voters are therefore able to take a clear position of either supporting or opposing the bills. Timing of the bills is also important since a hot topic near election time may attract more attention than bills that take place at other times. The deficit cutting bill, the Clinton impeachment bill, and the bailout legislation all occurred in a time close to a congressional election, when vulnerable members may be especially cautious against controversial legislations. Last but not least, if the general level of support to the bill is high, vulnerable members may be more likely to jump onto the bandwagon to support the legislation. Empirical Analysis Measuring electoral vulnerability I measure electoral vulnerability as a latent variable indicated by five variables: incumbent campaign spending, challenger campaign spending, previous electoral margin, district partisan 9

support, and seniority. District-level partisan support and incumbent seniority are also variables that indicate incumbents' electoral vulnerability. District-level partisan support is the baseline strength of incumbents' party among constituencies, and arguably the long-term force at work in the absence of personal factors related to congressional candidates (see, Converse 1966, Goldenberg and Traugott 1981). When district-level partisan support is strong, incumbents are more secure from opposite party challengers. Seniority directly taps into the long-observed phenomenon of incumbency safety, where incumbents win elections consecutively and arguably gain electoral advantage over time (e.g. Mayhew 1974). 2 Seniority may also contributes to incumbents' subjective sense of electoral vulnerability. Holding electoral margin and campaign spending constant, a junior member may feel more vulnerable compared to a senior, more seasoned member of Congress. Campaign spending is measured as all spending reported in the current electoral cycle; for the 109 th Congress it covers January 2005 to December 2006, for the 110 th Congress January 2007 to December 2008. Electoral margin is measured as the percentage of vote share obtained by the incumbents in the previous congressional election. If the election was a two-round system, vote percentage from the first round is used. The first round vote provides a better indicator of the incumbents' level of vulnerability, because it is an environment in which all candidates are eligible to compete for votes. District partisan support is measured as the incumbent party's presidential candidate vote share in the 2004 presidential election. Seniority is measured as terms served in Congress. 2 There is a long-standing debate about whether or not incumbents have been gaining electoral advantage over time. For counter argument, see (Jacobson 1987). 10

Statistically, incumbent spending and challenger spending are highly correlated. The reasons are two-fold. First, both are indicators of incumbents' electoral vulnerability; second, incumbents anticipate and react to challenger spending in election campaigns (e.g. Goldenberg, Traugott, and Baumgartner 1986, Jacobson 1978). The correlation requires data reduction methods, in the case of this paper factor analysis, that take into consideration variable correlation. [TABLE 1 HERE] Results of the exploratory factor analysis, presented in table 1, provide strong evidence of one common factor underlying the five indicating variables; the factor has the expected relationships with all of the indicating variables, although seniority seems to be a less relevant one. 3 To determine the number of latent factors, the rule of thumb Kaiser criterion is to retain factors with eigenvalues greater than 1, since eigenvalues indicate the amount of variance accounted for by each factor. The eigenvalue for the first factor is 1.49 and for the second one 0.47, which suggests that there is most likely one common factor, and one factor only, that explains most of the variance. Based on theories discussed previously, the common factor is electoral vulnerability, which should have negative relationships with the campaigns spending variables and positive relationships with previous electoral margin, district partisan support, and seniority. The factor loadings confirm the directionality of the relationships, which suggests that the common factor is indeed electoral vulnerability. 4 Moreover, the absolute values of the factor loadings suggest that the campaign spending variables are the most relevant indicators and seniority the least relevant one. This finding has 3 For the sake of simplicity, only results of the 110 th Congress are presented in table 1; results of the 109 th Congress are very similar. 4 It is also possible that the signs of the factor loadings are reversed, where the estimated factor has negative relationships with the spending variables and positive relationships with the other variables. A case like this is also consistent with the theory; the estimated factor would be electoral security, instead of electoral vulnerability. 11

two implications. First, it suggests that campaign spending variables need to be taken into consideration in conceptualizing and measuring electoral vulnerability; previous electoral margin alone could not fully account for variance of electoral vulnerability, which is subjective and dynamic. Second, given the low factor loading (an absolute value of 0.23), high uniqueness (0.9), as well as less theoretical importance of the seniority variable, it is not as relevant to the latent factor of electoral vulnerability. Consequently, seniority is not included in the estimation of electoral vulnerability in the next step. [TABLE 2 HERE] [FIGURE 1 HERE] Table 2 presents the scoring coefficients of the variables in the estimation of the electoral vulnerability score. The coefficients are rather similar between the 109 th and the 110 th Congress. Ranking from the most important to the least important is challenger campaign spending, incumbent campaign spending, previous electoral margin, and district partisan support. It is not surprising that challenger spending is more relevant than incumbent spending in terms of measuring vulnerability, because challenger spending is the less noisy signal. Incumbents engage in preemptive spending even when they are safe; challengers do not have similar behavior. When we observe the top five most secure and most vulnerable representatives from each Congress, which are listed in Table A1 in the Appendix, the story of electoral vulnerability is the story of Happy families are all alike; every unhappy family is unhappy in its own way (Tolstoy). In the 109 th Congress, the most secure legislators all won previous election with large electoral margin and ran unopposed in the 2006 election. Among the most vulnerable legislators, two faced strong challengers and were defeated in the 2006 election. One managed to survive the 12

2006 election with a very close margin, and decided to forfeit the 2008 election to run for other office. Two (Jim Gerlach R-FL and Melissa Bean D-IL) were incumbents who continued to serve in the House after 2006 but had particularly tough races in the 2006 election. Their electoral margins do not vary much over the years; for Gerlach it varies from 51% to 57%, and for Bean 48% to 52%. Their campaign spendings, however, are much higher in the 05-06 cycle than any of their previous or following elections. 5 This is an example of how the dynamic of incumbent's vulnerability can be much better captured by the spending variables. Figure 1 shows the distribution of the estimated vulnerability score for each Congress. Most legislators clustered around 0. Intuitively, one might suspect that there is a relationship between electoral vulnerability and ideological moderateness, since anecdotal stories highlight the vulnerability of moderate legislators such as bluedog Democrats. Using the absolute value of the DW-NOMIATE score as the measure of ideological moderateness, the scatterplots presented in figure 2, however, seem to suggest that there does not exist a relationship between electoral vulnerability and ideological moderateness. [FIGURE 2 HERE] Electoral vulnerability and responsiveness In this step of the analysis I examine the effect of electoral vulnerability on legislative responsiveness. The dependent variable of legislative responsiveness is measured as each representative's DW-NOMINATE score (Poole and Rosenthal 2000), which is derived based on legislative roll call votes. There are four explanatory variables: legislators' partisan affiliation, coded 0 for Democrats and 1 for Republicans; district ideological preferences, measured as the Republican presidential candidate's vote share from the 2004 election; electoral vulnerability 5 See figure A1 in the appendix. 13

score, and the interaction term between electoral vulnerability and district preferences. I include the interaction term so as to examine whether electoral vulnerability increases responsiveness. The theoretical discussion of electoral vulnerability and responsiveness poses three possible relationships. Each of the three possibilities has at least some empirical support. If the interaction term turns out to be positive and significant in this analysis, it is evidence supporting the marginality thesis, that is, more vulnerable members better represent constituency preferences. [TABLE 3 HERE] [FIGURE 3 HERE] The analysis uses OLS regression with robust standard errors to account for heteroskedasticity. Results presented in table 3 show that the interaction term is positive and significant at 0.01 level, meaning that vulnerability has a positive effect on responsiveness. More vulnerable members better reflect district partisan preferences in their own ideology. Figure 3 shows the increase of the marginal effect of district partisanship as vulnerability increases. It is worth noting that, in spite of the change of majority party in the House from the 109 th to the 110 th Congress, the effect of vulnerability on responsiveness persists to be positive and significant. This is evidence suggesting that the effect of vulnerability does not depend on majority party status. Note that the substantive effect of vulnerability is fairly small; it is not surprising since legislators' partisan affiliation and district ideological preference already account for most of the variance in the dependent variable. A nested model using only these two explanatory variables has R^2 values of 0.925 and 0.938 for the 110 th and 109 th congresses respectively. Legislators are already highly responsive to district preferences; therefore, the additional effect of vulnerability 14

on responsiveness is limited. However, the statistical significance and directionality of the effect of vulnerability on responsiveness is clear, which is evidence supporting the marginality thesis. Effect of vulnerability on roll call votes The next step in understanding the effect of vulnerability on legislative behavior is to obtain a more nuanced understanding of the conditions under which electoral vulnerability influences voting decisions. The analysis consists of two steps. In the first step I regress each legislators' vulnerability score on non-unanimous final passage bills voted in the 110 th congress, and obtain logistic coefficients of vulnerability on each bill. In the second step I let the regression coefficient be the dependent variable, and use issue area, levels of bill support, and timing of the bill to explain the size of the coefficients. 6 The first step analysis uses logistic regression because of the binary nature of the dependent variable. After controlling for partisanship and ideology, the results show that electoral vulnerability has a statistically significant effect on 15 out of the 119, or 12.6% final passage bills. 7 This finding is consistent with previous results that electoral vulnerability affects legislative behavior. In the second step, issue area is measured using the subject coding of the Congressional Bill Project (Adler and Wilkerson 2008); levels of bill support is measured as the total number of yea votes; and the timing of the bill is measured as the number of month into the congressional session, which ranges from 1 to 24. [TABLE 4 HERE] Results in table 4 show that neither government operation issues nor civil rights and civil liberty issues are related to the effect of electoral vulnerability. When there is a high level of 6 This research design is similar to that of Bovitz and Carson (2006) in their analysis of the effect of individual roll call votes on electoral outcomes. Instead of using statistical significance as the dependent variable, I use regression coefficients, because the question of interest is what affects the size of the vulnerability effect. 7 In the appendix I list the bills and the vulnerability coefficient for each bill. 15

support, electoral vulnerability is positively related to support for the bill. The finding is not surprising in that vulnerable members are likely to be more sensitive to political conflict and shy away from legislations when there is a close call. The sign of the bill timing coefficient is negative, 8 which suggests that vulnerable legislators may be less likely to support legislations when the votes take place close to the election time. In future versions of the paper, I plan to expand this section on the effect of vulnerability on roll call votes by using data from more congresses and including additional explanatory variables. Conclusion This paper contributes to the literature in two accounts. First it sheds light on our understanding of electoral vulnerability. When it comes to House elections, political scientists have observed a decline of electoral competition and an increase of incumbents' winning margins in recent decades. Some argue that the vanishing marginals increase incumbency safety and decrease electoral responsiveness. However, representatives themselves are unlikely to agree to the statement that elections are becoming easier and incumbents safer. One of the signs of competitive elections is the increasing amount of money that goes into election campaigns from both incumbents and challengers. To better understand and measure electoral vulnerability, we should not focus on previous electoral margin as the sole indicator, but expand the measurement to include important indicating variables of campaign expenditures. Second, it is too early to conclude that electoral vulnerability does not affect legislative behavior. Previous literature testing the marginality theory yielded mixed results; but it is likely to be a result of inadequate measurement and not because of erroneous theory. With the more robust measurement developed in this paper, I find strong evidence suggesting that electoral 8 But it falls just short of achieving statistical significance at 0.05 level. 16

vulnerability indeed increases legislative responsiveness. Vulnerability is not irrelevant, as some suggest, and certainly not detrimental, as others argue, to legislative responsiveness. This finding has important normative implications. Classic democratic theories rely on competitive elections for accountability and responsiveness, which seems to be the mechanism at work in the U.S. Congress. To conclude, more research is necessary to further examine the relationship between electoral vulnerability and legislative decisions, which is essential to representative democracy. Given the normative importance of the topic, there has been surprisingly little research on this subject in recent years. The next step of this paper is to further develop theories and empirical analysis to examine the conditions under which electoral vulnerability affects roll call votes. 17

Reference Bartels, Larry M. 1994. The American Public s Defense Spending Preferences in the Post-Cold War Era. The Public Opinion Quarterly 58 (4) (December 1): 479 508. Bartlett, Robert V. 1979. The Marginality Hypothesis Electoral Insecurity, Self-Interest, and Voting Behavior. American Politics Research 7 (4) (October 1): 498 508. doi:10.1177/1532673x7900700407. Bovitz, Gregory L., and Jamie L. Carson. 2006. Position-Taking and Electoral Accountability in the U.S. House of Representatives. Political Research Quarterly 59 (2) (June 1): 297 312. Campbell, James. 1981. Electoral Competition & the Congressional Connection: The Marginality Hypothesis Reconsidered. Political Methodology 7 (2): 55 70. doi:10.2307/25791109. Cohen, Jeffrey E. 1984. Perceptions of Electoral Insecurity Among Members Holding Safe Seats in a U. S. State Legislature. Legislative Studies Quarterly 9 (2) (May 1): 365 369. doi:10.2307/439397. Cohen, Jeffrey E., and Gregory G. Brunk. 1983. A Dynamic Test of the Marginality Hypothesis. Political Behavior 5 (3) (January 1): 293 307. Converse, Phillip. 1966. The Concept of Normal Vote. In Elections and the Political Order, edited by Phillip Converse, Angus Campbell, Warren Miller, and Warren Stokes. New York: John Wiley & Sons. Deckard, Barbara Sinclair. 1976. Electoral Marginality and Party Loyalty in House Roll Call Voting. American Journal of Political Science 20 (3) (August 1): 469 481. doi:10.2307/2110683. Fenno, Richard F. 1978. Home Style: House Members in Their Districts. Little, Brown. Fiorina, Morris P. 1974. Representatives, Roll Calls, and Constituencies. Lexington Books. Froman, Lewis A. 1965. Congressmen and Their Constituencies. Gay, Claudine. 2007. Legislating Without Constraints: The Effect of Minority Districting on Legislators Responsiveness to Constituency Preferences. The Journal of Politics 69 (02): 442 456. doi:10.1111/j.1468-2508.2007.00542.x. Goldenberg, Edie N., and Michael W. Traugott. 1981. Normal Vote Analysis of U. S. Congressional Elections. Legislative Studies Quarterly 6 (2) (May 1): 247 257. doi:10.2307/439648. Goldenberg, Edie N., Michael W. Traugott, and Frank R. Baumgartner. 1986. Preemptive and Reactive Spending in U. S. House Races. Political Behavior 8 (1) (January 1): 3 20. Griffin, John D. 2006a. Electoral Competition and Democratic Responsiveness: A Defense of the Marginality Hypothesis. The Journal of Politics 68 (04): 911 921. doi:10.1111/j.1468-2508.2006.00479.x.. 2006b. Senate Apportionment as a Source of Political Inequality. Legislative Studies Quarterly 31 (3) (August 1): 405 432. doi:10.3162/036298006x201869. Gulati, Girish J. 2004. Revisiting the Link Between Electoral Competition and Policy Extremism in the U.S. Congress. American Politics Research 32 (5) (September 1): 495 520. doi:10.1177/1532673x04263820. 18

Jacobson, Gary C. 1978. The Effects of Campaign Spending in Congressional Elections. The American Political Science Review 72 (2) (June 1): 469 491. doi:10.2307/1954105.. 1987. The Marginals Never Vanished: Incumbency and Competition in Elections to the U.S. House of Representatives, 1952-82. American Journal of Political Science 31 (1) (February 1): 126 141. doi:10.2307/2111327.. 1990. The Effects of Campaign Spending in House Elections: New Evidence for Old Arguments. American Journal of Political Science 34 (2) (May 1): 334 362. doi:10.2307/2111450.. 1993. Deficit-Cutting Politics and Congressional Elections. Political Science Quarterly 108 (3) (October 1): 375 402. doi:10.2307/2151696. Jacobson, Gary C., and Samuel Kernell. 1981. Strategy and Choice in Congressional Elections. Yale University Press. Kuklinski, James H. 1977. District Competitiveness and Legislative Roll-Call Behavior: A Reassessment of the Marginality Hypothesis. American Journal of Political Science 21 (3): 627 638. doi:10.2307/2110584. Lanoue, David J., and Craig F. Emmert. 1999. Voting in the Glare of the Spotlight: Representatives Votes on the Impeachment of President Clinton. Polity 32 (2) (December 1): 253 269. doi:10.2307/3235285. Lublin, David Ian. 1994. Quality, Not Quantity: Strategic Politicians in U.S. Senate Elections, 1952-1990. The Journal of Politics 56 (1) (February 1): 228 241. doi:10.2307/2132355. Macrae, Duncan. 1952. The Relation Between Roll Call Votes and Constituencies in the Massachusetts House of Representatives. The American Political Science Review 46 (4) (December 1): 1046 1055. doi:10.2307/1952111. Mayhew, David R. 1974. Congressional Elections: The Case of the Vanishing Marginals. Polity 6 (3) (April 1): 295 317. doi:10.2307/3233931. Poole, Keith T., and Howard Rosenthal. 2000. Congress: a Political-economic History of Roll Call Voting. Oxford University Press US. Shannon, Wayne. 1968. Electoral Margins and Voting Behavior in the House of Representatives: The Case of the Eighty-Sixth and Eighty-Seventh Congresses. The Journal of Politics 30 (04): 1028 1045. doi:10.2307/2128687. Sullivan, John L., and Eric M. Uslaner. 1978. Congressional Behavior and Electoral Marginality. American Journal of Political Science 22 (3): 536 553. doi:10.2307/2110460. 19

Appendix Table A1: Top 5 List State District Name 110 th Congress Electoral margin 06 Incumbent spending 07-08 (thousands of dollars) Challenger spending 07-08 (thousands of dollars) Most secure representatives California 33rd Watson, Diane E. 100% 229.692 5.773 New Jersey 10th Payne, Donald M. 100% 502.611 0 Pennsylvani a 1st Brady, Robert A. 100% 1013.84 0 Illinois 4th Gutierrez, Luis V. 86% 188.438 0 Illinois 7th Davis, Danny K. 87% 413.001 0 Most vulnerable representatives Washington 8th Reichert, David G. 51% 2852.51 4322.88 Connecticut 4th Shays, Christopher 51% 3771.64 3828.3 Illinois 10th Kirk, Mark Steven 53% 5444.41 3566.12 Illinois 14th Foster, Bill 55% 4904.27 5064.49 New York 20th Gillibrand, Kirsten E. 50% 4489.39 7038.5 State District Name 109 th Congress Electoral margin 04 Incumbent spending 05-06 (thousands of dollars) Challenger spending 05-06 (thousands of dollars) Most secure representatives New York 16th Serrano, José E. 91% 314.38 0 New Jersey 10th Payne, Donald M. 97% 495.562 0 California 33rd Watson, Diane E. 89% 181.051 0 Florida 23rd Hastings, Alcee L. 100% 427.924 0 Massachuset Capuano, Michael 8th ts E. 99% 626.795 0 Most vulnerable representatives Pennsylvani a 6th Gerlach, Jim 51% 3492.4 4097.66 New Mexico 1st Wilson, Heather A. 50% 4906.6 3386.54 Florida 22nd Shaw, Clay 63% 4185.92 5226.16 Illinois 8th Bean, Melissa 52% 4294.59 5058.73 Florida 13th Harris, Katherine 55% 8112.75 3002.8 20

Figure A2: Dynamics of Electoral Vulnerability Congressional Election Spending: PA 6 th Winner: Jim Gerlach Congressional Election Spending: IL 8 th Winner: Melissa Bean 21

Coefficient of vulnerability Table A2: Bill List Bill title 0.60 Implementing the 9/11Commission Recommendations Act 0.72 Rail and Public Transportation Security Act 0.93 To provide Federal assistance to States, local jurisdictions, and Indian tribes to prosecute hate crimes. 0.82 Department of Homeland Security Authorization Act 0.35 Department of Homeland Security Appropriations for FY 2008 0.53 0.37 Making appropriations for the Department of Labor, Health and Human Services, and Education, and related agencies for fiscal year ending September 30, 2008, and for other purposes Departments of Transportation, and Housing and Urban Development and Related Agencies Appropriations for FY 2008 0.85 Flood Insurance Reform and Modernization Act of 2007 0.68 Free Flow of Information Act of 2007 1.07 Saving Energy Through Public Transportation Act 0.50 National Highway Bridge Reconstruction and Inspection Act -1.32 Paycheck Fairness Act 1.86 Commodity Markets Transparency and Accountability Act -0.51 Employment Non-Discrimination Act (ENDA) (Listed are bills in which vulnerability significantly affects voting decisions.) 22

Table 1: Exploratory Factor Analysis Method: Principle Axis Factor # of observations: 436 Factor Eigenvalue Difference Proportion Cumulative Factor 1 1.495 1.017 0.970 0.970 Factor 2 0.477 0.480 0.310 1.280 Factor 3-0.003 0.195-0.002 1.278 Factor 4-0.198 0.032-0.129 1.149 Factor Loading Variable Factor 1 Factor 2 Uniqueness Opponent spending 07-08 0.791 0.165 0.345 Incumbent spending 07-08 0.685 0.341 0.413 Electoral margin 2006-0.451 0.402 0.634 District partisan support 2004-0.373 0.357 0.732 Seniority -0.233 0.210 0.901 (with 110 th Congress data) Table 2: Factor Analysis Estimation 110 th Congress 109 th Congress Variable Factor Loading Uniqueness Scoring Coefficient Variable Factor Loading Uniqueness Scoring Coefficient Opponent spending 07-08 0.794 0.369 0.481 Opponent spending 05-06 0.743 0.447 0.395 Incumbent spending 07-08 0.706 0.501 0.327 Incumbent spending 05-06 0.681 0.535 0.307 Electoral margin 2006-0.419 0.823-0.160 Electoral margin 2004-0.530 0.718-0.208 District partisan support 2004-0.360 0.869-0.138 District partisan support 2004-0.527 0.721-0.208 23

Figure 1: Estimated Electoral Vulnerability Figure 2: Electoral Vulnerability and Ideological Moderateness 24

Table 3: Electoral vulnerability and Legislative Responsiveness 110 th Congress 109 th Congress Variable Party ID Bush vote share 2004 Electoral vulnerability Vote share * Vulnerability Constant OLS Coefficient (Robust Standard Error) 0.799* (0.017) 0.010* (0.001) -0.123* (0.044) 0.003* (0.001) -0.839* (0.041) 0.844* (0.017) 0.008* (0.001) -0.097* (0.035) 0.002* (0.001) -0.713* (0.036) R^2 = 0.927 R^2 = 0.939 * Statistically significant at 0.05 level Figure 3: Electoral vulnerability and Legislative Responsiveness 25

Table 4: Explaining Effect of Vulnerability Variable Timing Bill support Government operation Civil rights and civil liberties Constant # of observations: 118 OLS Coefficient -0.0009ᶵ (0.0005) 0.001* (0.000) 0.007 (0.023) -0.004 (0.010) -0.010 (0.015) R^2 = 0.093 ᶵ t-value =0.06 * Statistically significant at 0.05 level 26