POLICY VOTING IN SENATE ELECTIONS: The Case of Abortion

Political Behavior, Vol. 26, No. 2, June 2004 (Ó 2004) POLICY VOTING IN SENATE ELECTIONS: The Case of Abortion Benjamin Highton Questions about whether voters rely on their policy preferences when casting ballots have been present since scholars first began examining the determinants of voting behavior. This paper seeks to contribute to research in this area by analyzing abortion policy voting in Senate elections. Specifically, I investigate how the effects of national party position divergence, candidate position divergence, and voter information and salience moderate the relationship between abortion policy preferences and vote choice. The results suggest that the national parties divergence on abortion does not directly strengthen the connection between abortion policy preferences and ballot decisions. Instead, candidate contrast appears to be the key. And, well informed and motivated voters are especially responsive. Taken together, the findings illuminate the nature of abortion policy voting and also inform the burgeoning scholarship on campaign effects, the role of information, and issue publics in American politics. Key words: senate; elections; abortion; policy voting; issues. Whether or not citizens engage in issue voting is a question of longstanding importance for students of voting behavior. Early investigations of voting in presidential elections cast doubt on the proposition that policy preferences were a significant factor affecting vote choice. For example, Berelson, Lazarsfeld, and McPhee (1954) showed that few members of the electorate knew Truman s and Dewey s positions on two major issues in 1948, the Taft Hartley Act and price controls. And, Campbell, et al. (1960) found only a modest articulation between party policy and voter response (p. 183). Despite the pessimistic implications of these early and influential studies, subsequent research suggests Department of Political Science, One Shields Avenue, University of California, Davis, CA 95616-8682 (bhighton@ucdavis.edu). I appreciate comments and advice from Henry E. Brady, J. Merrill Shanks, Walter J. Stone, and Raymond E. Wolfinger. 181 0190-9320/04/0600-0181/0 Ó 2004 Plenum Publishing Corporation

182 HIGHTON that the question is far from settled. Perhaps the most important development has been the expansion of scholars attention to contextual factors and voter heterogeneity when considering the nature of policy voting. 1 This paper builds on previous research and analyzes the factors that moderate the relationship between voters abortion policy preferences and their ballot choices in Senate elections. Since the Supreme Court s Roe v. Wade decision in 1973 and the subsequent evolution of party polarization on the abortion issue, the electoral relevance of abortion has steadily grown (Abramowitz, 1995; Adams, 1997). The first question I consider is how Senate candidates and their national parties positions relate to the observed levels of policy voting in Senate elections. Then, at the level of individual voters, I investigate how abortion policy voting depends on information and salience. Finally, I consider whether the results reported in this paper generalize to other issues and different types of elections. OPPORTUNITY FOR POLICY VOTING Policy, or issue, voting is commonly defined as the extent to which ballot choices are caused by the voter acting in accord with his [or her] policy preferences (Brody and Page, 1972, p. 450). Although it is impossible to determine whether any particular individual engages in issue voting, one can determine whether there are meaningful relationships between policy preferences and ballot decisions among groups of voters or in the electorate overall. When considering the factors that might influence the magnitude of the relationship, aspects of the electoral context and voters gain prominence because they are related to the opportunities, information, and motivation voters have to engage in policy voting. The opportunity to cast a ballot based on policy preferences depends on the characteristics of the choice with which voters are presented. Suppose two competing candidates hold identical positions regarding issue X and that the candidates positions perfectly reflect their parties positions. In this instance, the election outcome has no discernable relationship to policy regarding X. The choice faced by voters presents no opportunity instrumental, expressive, or otherwise for their preferences about X to help them pick one candidate over the other. Divergence between candidates and/or parties policy positions thus appears to be a necessary condition for voters policy preferences to be related to their ballot choices. As V.O. Key (1966) famously put it in his posthumous defense of the American electorate: The voice of the people is but an echo. The output of an echo chamber bears an inevitable and invariable relation to the input. As candidates and parties clamor for attention and vie for popular support, the people s verdict can be no more than a

POLICY VOTING IN SENATE ELECTIONS 183 selective reflection from among the alternatives and outlooks presented to them (Key, 1966, p. 2, italics added). Consequently, it is important to examine the electoral context in order to understand the nature of the opportunity for voters policy preferences to help them choose among competing electoral options. What kind of alternatives do voters have regarding abortion policy? One relates to the choice presented by the national parties. Since the Supreme Court s Roe v. Wade decision in 1973 the national parties have been polarizing around the abortion issue (Adams, 1997). In 1980 they formalized their positions in their party platforms; the Democratic party took on the pro-choice position and the Republicans adopted the pro-life one. For example, in the presidential election year covered in this study (1992) the Republican platform said, in part, that We therefore reaffirm our support for a human life amendment to the Constitution, and we endorse legislation to make clear that the 14th Amendment s protections apply to unborn children. Democrats, on the other hand, stood behind the right of every woman to choose, consistent with Roe v. Wade, regardless of ability to pay, and support a national law to protect that right. Because the national parties are so clearly split on abortion, the opportunity for voters abortion policy preferences to influence their senatorial ballot decisions presents itself in every election where candidates from the two parties compete. Whether there is an observable relationship depends on if the national parties positions contribute to the structure of voters decisions. In Senate elections, one alternative presented to voters is from the national parties. Another relates to the positions of the candidates, which are not always consistent with those of their national parties. As I will show, it is not uncommon to observe pro-life Democratic candidates and pro-choice Republican ones. As a result, voters sometimes, but not always, are presented with a choice based on the positions of the candidates. In some elections the candidates present alternative abortion positions as one is pro-choice and the other is pro-life. But, in other elections both candidates have similar abortion positions; either both are pro-choice or both are pro-life. If voters respond to the positions of the candidates, then contrast in candidates positions could strengthen the observed relationship between abortion policy preferences and vote choice. However, it is important to keep in mind the strategic incentives of competing candidates. When candidates hold opposing positions and one position is favored by the electorate, then the candidate holding the other position has a strong incentive to pursue a strategy of ambiguity (Page, 1976; Shepsle, 1972) or obfuscation (Franklin, 1991) so voters do not perceive the candidate s losing position. 2 If the strategy is successful, then the apparent level of policy voting will be lower than it otherwise would have been. If completely successful, then no policy voting will be observed. The point is that the opportunity for policy voting may not translate into its occurrence, even if

184 HIGHTON voters are predisposed to bring their ballot choices in line with their policy preferences. Previous research has investigated the role of candidate divergence, or contrast, on vote choice in Senate elections (Abramowitz, 1981; Wright and Berkman, 1986). This study complements that work and distinguishes itself by its focus on a particular issue whereas Abramowitz (1981) and Wright and Berkman (1986) emphasize the general level of ideological similarity and difference among candidates. In addition, because I analyze elections in 2 years, there is leverage to address the question of whether voters actually respond to the candidates positions or whether candidates positions and observed patterns of voting behavior are both caused by fixed characteristics of state electorates, a point I elaborate further, below. INFORMATION, SALIENCE, AND POLICY VOTING Although opportunity is a necessary condition for policy voting, it is not the only factor that might influence the relationship between policy preferences and voting behavior. If candidates and/or parties positions diverge, voters can only be expected to be responsive to the extent that this information reaches them in some form. Absent reception and incorporation, a voter is no different from one in an electoral context with no issue position contrast. The importance of information has been studied extensively in a variety of electoral contexts (e.g., Alvarez, 1998; Delli Carpini and Keeter, 1996; Kahn and Kenney, 1999; Zaller, 1992). In congressional elections, both campaign intensity (Gronke, 2000; Krasno, 1994; Kahn and Kenney, 1999; Westlye, 1991) and voters political awareness (Kahn and Kenney, 1999) contribute to and facilitate the acquisition of information about candidates. Here, I investigate how information influences the relationship between abortion policy preferences and vote choice. Consistent with studies of information effects in presidential voting (e.g., Abramowitz, 1995; Alvarez, 1998; Delli Carpini and Keeter, 1996; Jacoby, 1986; Knight, 1985; Miller and Shanks, 1996), I hypothesize that greater voter information levels strengthen the relationship between abortion policy preferences and vote choice. In addition to analyzing how the determinants of vote choice depend on information, I also estimate how the choices voters make would change if they were more fully informed about the candidates, which provides insight into the electoral consequences of information effects (Althaus, 2001; Bartels, 1996). Clearly, heterogeneity among voters is an important factor to consider. 3 Information is one dimension of heterogeneity. Another relates to the motivation people have for voting on the basis of policy considerations. Converse (1964) introduced the term issue public to describe people who had particular interest in narrower belief areas... people who had shown a self starting concern about particular controversies (p. 244). Members of an issue

POLICY VOTING IN SENATE ELECTIONS 185 public, because they are more concerned about a particular policy area, have greater motivation to vote on the basis of their policy preferences. Several studies (Abramowitz, 1995; Krosnick, 1990; Rabinowitz, Prothro, and Jacoby, 1982; Repas, 1971), all of presidential voting, provide empirical evidence consistent with this claim that issue salience influences the level of policy voting. Whether or not issue publics, in general, and the abortion issue public, in particular, are important for congressional voting remains an open question. Also important, previous investigations of information and salience effects tend to consider them separately rather than simultaneously, which poses a problem of interpretation. The difficulty with analyzing them separately is that the two effects may be conflated if salience and information are interrelated. For example, Krosnick (1990) finds that attitudes toward Russia were more strongly related to presidential voting in 1980 among those who indicated that it was an important issue to them. However, if the issue was more important to voters with greater information, then it unclear if the effect should be understood as a salience or information effect. As Althaus (2001) puts it: There are any number of relevant variables associated with political knowledge that might be contributing to the appearance of information effects. For example, highly knowledgeable people perceive a greater number of policy issues as important or salient than do less-knowledgeable people. Perhaps the perceived salience of an issue or campaign, rather than knowledge itself, encourages well-informed people to engage in a more thorough and thoughtful analysis of a candidate s policy positions (p. 51). Because information and salience are distinct but correlated concepts, it is preferable to construct and estimate a voting model that permits the effects of policy preferences to vary with both information and salience. In the analysis reported here, I do this by using separate measures of each concept and estimating the effects of both on the relationship between abortion policy preferences and vote choice. RESEARCH DESIGN AND DATA In their wide-ranging study of Senate elections, Kahn and Kenney (1999) closely examine Senate elections between 1988 and 1992. They collected a wide variety of contextual data on these elections and merged it with the individual-level data collected by the National Election Studies. Although they make many important contributions, their investigation of policy voting was limited by the fact that it is not easy to isolate the important issues in all Senate contests in a given year, since some of the salient issues in Senate campaigns vary from state to state (Kahn and Kenney 1999, p. 211). By focusing on a single issue, abortion, the problem of differences across elections encountered by Kahn and Kenney becomes an asset in this study.

186 HIGHTON As I have explained, because of the national party divergence on abortion, the opportunity for abortion policy preferences to influence vote choice is present in all senatorial elections. Moreover, in elections where the candidates hold similar positions, it is the only source. Thus elections where the candidates abortion positions are similar provide a useful context to test the hypothesis that voters decisionmaking is influenced by national party positions. At the same time, by comparing the magnitude of the relationship between abortion policy preferences and vote choice in these elections to its size in elections where the candidates positions are different, it is possible to assess how the relationship depends on the candidates positions. 4 To classify the elections based on whether or not candidates abortion positions were similar or different, I conducted an extensive search of numerous national campaign periodicals like the Cook Political Report, Hotline, and the Abortion Report in addition to a large number of local newspapers. 5 For instance, consider the 1990 South Dakota election where Larry Pressler and Ted Meunster held contrasting abortion positions. On October 15, the Abortion Report (1990) provided an update on the election, with quotes from a debate between the candidates. Pressler declared that he was a pro-life, pro-adoption senator while Meunster said that I believe that decision is best left to the woman. The 1992 Illinois election is an example of one where the candidates positions were similar. The Democrat, Carol Moseley Braun, ran a television ad in which she emphasized a woman s right to choose (Abortion Report 1992). Her opponent, Richard Williamson held a news conference announcing his endorsement from Republicans for Choice and was careful to distinguish his pro-choice position from the pro-life position of the president: He is comfortable with this [pro-life] platform and that s fine with me...my obligation is to let the people of Illinois know where I stand (Abortion Report 1992). Overall, 35 of the 66 contested elections involved candidates who held different abortion positions. 6 I rely on the Senate Election Studies (SES) for individual-level data (Miller et al., 1999). The SES includes 2,567 voters of whom 1,386 (54%) cast ballots in elections where the candidates positions were different. To measure voters policy preferences regarding abortion, their answers to three questions were combined into a single scale of abortion policy preferences. Details about the questions and coding are in the appendix. In short, I coded each question on a )1 (conservative/pro-life) to +1 (liberal/prochoice) scale and then added respondents answers to produce a measure ranging from )3 to +3. Isolating the relationship between abortion policy preferences and vote choice requires a model that takes into account other political attitudes. Otherwise, a serious problem of interpretation results: the connection between policy preferences and vote choice might indicate the role of policy

POLICY VOTING IN SENATE ELECTIONS 187 preferences or the relationship might be the result of the effect of more general policy liberalism/conservatism, for example. 7 In this study I employ three variables tapping voters general political orientations in order to distinguish the independent effects of voters abortion policy preferences. Party identification is based on the traditional NES measure of party identification and ranges from )2 (strong Republican) to +2 (strong Democrat). 8 Ideological identification is based on respondents classifications of themselves using a seven point scale from )3 to +3. 9 Finally, a scale comprised of answers to ten policy questions serves as a measure of general policy liberalism (Stimson, 1999; Goren, 2002). Each was coded )1 (conservative) to +1 (liberal), and then respondents answers were summed, which produced a scale ranging from )10 (all conservative policy preferences) to +10 (all liberal policy preferences). 10 In order to investigate voter heterogeneity, indicators of information and motivation are also required. To empirically separate their effects, independent measures of each must be included in the analysis. Following the work of others (e.g., Gronke, 2000; Jacobson, 2001; Kahn and Kenney, 1999; Krasno, 1994), I measure information about the candidates based on voters ability to (a) recall the candidates names (b) recognize their names (c) place them on a general ideology scale, and (d) provide reasons for liking/disliking them. The result is a 12 point scale that I coded to range from 0 to 1. Distinguishing the abortion issue public is less straightforward. The SES did not include questions to elicit directly how important or concerned respondents were with specific issues like abortion. However, Krosnick (1990, p. 65) found a positive relationship between issue importance and issue mentions in response to open-ended questions about presidential candidates. Following his lead, I classified voters based on whether or not they mentioned abortion in response to a set of open-ended questions about the Senate candidates. 11 Another indirect way to tap abortion salience involves identifying demographic groups whose interests are more closely linked to the policies in the issue area and are more likely to be affected (Krosnick, 1990; p. 72 75). With regard to abortion, then, women are a demographic group more likely to be motivated by the abortion issue. This notion is consistent with, for example, Luker s (1984) study of abortion activists where she found that over 80% of activists in both the pro-choice and pro-life movements in California were women (p. 194). RESULTS To examine the relationship between election context and abortion policy voting, Table 1 presents the results from a logit model of vote choice where the structure of vote choice is allowed to vary by whether there was contrast in the

188 HIGHTON TABLE 1. Parameter Estimates of Democratic Vote Choice in Senate Elections Variable Candidate contrast = 0 Candidate contrast = 1 Difference Abortion policy.028 (.051).213 (.046).185 (.067) preferences Policy liberalism.062 (.027).093 (.026).032 (.037) Ideological identification.271 (.073).275 (.065).005 (.098) Party identification.899 (.072).620 (.062) ).279 (.095) Economic assessments ).122 (.102) ).310 (.093) ).188 (.138) Incumbency 1.288 (.104).860 (.086) ).428 (.135) Age ).050 (.079).043 (.074).093 (.108) Education ).052 (.091).006 (.083).057 (.123) Family income ).027 (.089) ).109 (.078) ).082 (.118) Black.325 (.452).638 (.362).313 (.579) Female.122 (.156) ).207 (.147) ).329 (.214) Constant.103 (.421).087 (.390) ).016 (.574) p<.10; p<.05 Number of observations: 2,456 Log-likelihood: )1, 134 Pseudo R 2 :.33 Note: Standard errors in parentheses. candidates abortion positions. The model is fully interactive; there is an interaction between every independent variable and candidate contrast. For each independent variable, there are three estimates. The first is the parameter estimate for elections where the candidates held similar positions (candidate contrast = 0). The second coefficient reports the estimated effect in elections where the candidates held different abortion positions (candidate contrast = 1). The third reports the difference in the estimates across election types. The results show that the relationship between abortion policy preferences and vote choice is strengthened where there is candidate contrast. In elections where the candidates positions were similar, the estimated effect of abortion policy preferences was small (b =.028), and the null hypothesis of no relationship is plausible (p =.58). In those elections where there the candidates positions were different, the estimate is larger and significant (b =.213, p <.01). In addition, the possibility of no difference in the effects of abortion policy preferences across electoral context can be rejected with confidence (p <.01). 12 Substantively, the difference in the estimated effects is sizable. Figure 1 depicts the relationship between voters abortion policy opinions and the probability of voting for the Democratic candidate in both electoral contexts. 13 In elections where the candidates positions were similar, the observed relationship is slight. The difference in the estimated probability of voting Democratic between those with the most liberal and conservative abortion

POLICY VOTING IN SENATE ELECTIONS 189 Predicted Probability of Voting Democratic.70.65.60.55.50.45.40-3 -2-1 0 1 2 3 conservative moderate liberal Abortion Opinion Candidates' Abortion Positions similar different FIG. 1. Estimated Effects of Abortion Policy Preferences on Vote Choice. policy preferences is just.02. But, when the candidates positions diverged, the effect of voters abortion policy preferences increases to.21. Clearly, there is a strong relationship between electoral context and the apparent level of policy voting. An examination of the other interaction terms in Table 1 is important to assess an alternative explanation that could explain the larger effects of abortion policy preferences in elections where the candidates positions diverge. If these elections were more intense and competitive, then the richer informational environment could explain the larger effect of abortion policy preferences. 14 However, if this was the case, then the effects of the other political attitudes (party identification, ideological identification, and policy liberalism) would also be larger. But the estimates reported in Table 1 show that the effect of party identification is diminished, the effect of ideological identification is almost unchanged, and although the effect of policy liberalism is larger, the possibility of no difference cannot be rejected with much confidence (p =.40). 15 The next set of hypotheses relate to voter heterogeneity: how information and motivation influence the level of policy voting. Because the relationship between abortion policy preferences and vote choice is insignificant in elections where the candidates held similar abortion positions, the analysis is limited to those contests where the candidates abortion policy positions were different (candidate contrast = 1). The information hypothesis predicts that the level of abortion policy voting depends on how much information voters

190 HIGHTON have about the candidates. Thus the model includes an interaction between abortion policy preferences and information. The motivation hypothesis specifies that voters in the abortion issue public will place greater weight on their abortion policy preferences. This implies interactions between abortion policy preferences and the two measures of salience (gender and mentioning abortion in response to the open-ended questions.) The logit parameter estimates are presented in Table 2. Both the information and motivation hypotheses receive support. The interaction between information and abortion policy preferences is positive (b =.518, p =.05) as are the interactions between abortion policy preferences and gender (b=.187, p=.04) and mentioning abortion (b =.266, p =.05). And, a test of the joint significance of the three interactions strongly suggests rejecting the null hypothesis of no interaction effects (p =.003). Figure 2 illustrates the effect of abortion policy preferences for those with low, medium, and high levels of information. 16 Among the least informed, the probability of voting Democratic increases.06 as abortion policy preferences vary from conservative to liberal. Among those in the middle, the estimated effect is.19, while for those with high information the estimated effect is.29. TABLE 2. Parameter Estimates of Democratic Vote Choice (Elections where Candidates held Contrasting Abortion Positions) Baseline Effects Logit Estimate Standard Error Constant ).505.485 Age.029.075 Education ).020.086 Family income ).142.079 Black.773.369 Female ).132.152 Abortion mention ).510.251 Candidate information 1.063.433 Incumbency.891.087 Economic assessments ).283.094 Party identification.627.063 Ideological identification.265.066 Policy liberalism.091.026 Abortion policy preferences ).332.217 Interactive effects: abortion policy preferences Candidate information.518.264 Female.187.093 Abortion mention.266.134 p<.10; p<.05 Number of observations: 1,283 Log-likelihood: )600 Pseudo R 2 :.33.

POLICY VOTING IN SENATE ELECTIONS 191 Predicted Probability of Voting Democratic.70.60.50.40.30.20-3 -2-1 0 1 2 3 conservative moderate liberal Abortion Opinion Information low medium high FIG. 2. Voter Information and Abortion Policy Voting. In short, greater information about the candidates considerably strengthens the relationship between abortion policy preferences and vote choice. The estimated effects of the two motivation indicators are displayed in Figures 3 and 4. Among men, the estimated effect is just.08, substantially smaller than the estimated effect of.26 for women. For voters who did not mention abortion in response to the open-ended questions the estimated effect of abortion policy preferences is smaller (.15) than for the group for whom abortion was more salient (.38). So, like the effects of information, motivation appears to moderate the observed effects of abortion policy preferences, too. Taken together, the results in Table 2, which are depicted in Figures 2 4 reveal substantial heterogeneity within the voting public. DISCUSSION The results provide important insights into a number of aspects of voting behavior. First, consider the trivial relationship between abortion policy preferences and vote choice in elections where both Senate candidates were either pro-choice or pro-life. In these elections, the candidates presented voters with no alternative regarding abortion policy. But, if voters either attribute national party positions to senatorial candidates or hold candidates responsible for their parties positions, then one would expect to observe at least some abortion policy voting in elections where the candidates positions were similar. The absence of such a finding suggests that neither phenomenon

192 HIGHTON Predicted Probability of Voting Democratic.70.60.50.40.30.20-3 -2-1 0 1 2 3 conservative moderate liberal Abortion Opinion men women FIG. 3. Salience (Gender) and Abortion Policy Voting. occurs. Thus to the extent that they matter, the role of national party abortion positions on vote choice in Senate elections is indirect, operating through party identification. If people adjust their partisan affiliations to bring them in Predicted Probability of Voting Democratic.70.60.50.40.30.20-3 -2-1 0 1 2 3 conservative moderate liberal Abortion Opinion no mention mention FIG. 4. Salience (Mentioning Abortion) and Abortion Policy Voting.

POLICY VOTING IN SENATE ELECTIONS 193 line with their abortion preferences, then part of the effect of party identification apparent across all the vote choice models should be interpreted as reflecting the long-term impact of national party abortion positions on vote choice. 17 The sizable relationship between abortion policy preferences and vote choice evident in elections where the candidates held contrasting positions is also important. Candidates holding positions that put them on the wrong side of the issue in their respective electorates (e.g., pro-choice candidates in Mississippi and pro-life candidates in Rhode Island) who face opponents with the stronger position have strong incentives to obfuscate and employ strategies of ambiguity in order to minimize the relationship between abortion policy preferences and vote choice. If they were completely successful, then there would be no apparent relationship between abortion preferences and vote choice. 18 Therefore, the substantial relationship suggests that either voters are not confused by the strategic behavior of obfuscating candidates or that the countervailing incentives for candidates who hold the stronger position offset the effects of this behavior. Either way, when the candidates positions diverge, voters have the opportunity to vote on the basis of policy, and the results reported here indicate that they do. The finding that candidate divergence on abortion is associated with the magnitude of the relationship between abortion policy preferences and vote choice is consistent with previous work on candidate ideological contrast (Abramowitz, 1981; Westlye, 1991; Wright and Berkman, 1986). These studies show that in Senate elections where ideological differences between candidates are greater, the relationship between ideological identification and vote choice is stronger. The present study complements this research by showing that the relationship between candidate divergence and vote choice extends from general ideological contrast to the specific issue of abortion. However, there is an unresolved question of causality. Do voters actually respond to candidate contrast (whether at the general ideological level or specific issue level) or is candidate contrast endogenous, perhaps caused by the nature of voters ideological/policy preferences across the states? Some insight into this question is possible because the survey data cover two election years (1990 and 1992) thereby providing data on two elections from some states. As it turns out there were 17 states with two contested Senate elections in 1990 and 1992. If candidates positions are determined by fixed state characteristics, then there should be a close mapping between similarity/contrast in the 17 pairs of elections. However, 8 of the states do not display this pattern. Instead, one of the elections had candidates with similar positions and the other had candidates with contrasting positions. In these states, then, there is variation in candidates positions that is independent of other fixed (at least in the short run) state factors, like the distribution of voters abortion policy preferences. Replicating the model from Table 1 and limiting the analysis to

194 HIGHTON TABLE 3. Parameter Estimates of Democratic Vote Choice (States with Two Elections and Variation in Candidate Contrast) Variable Candidate Contrast = 0 Candidate Contrast = 1 Difference Abortion policy.115 (.109).305 (.095).191 (.145) preferences Policy liberalism.056 (.055).072 (.055).016 (.078) Ideological identification.386 (.154).399 (.135).012 (.205) Party identification 1.070 (.167).589 (.130) ).481 (.213) Economic assessments ).179 (.237) ).441 (.196) ).261 (.307) Incumbency 2.113 (.263) 1.217 (.240) ).897 (.356) Age.059 (.169).067 (.151).008 (.227) Education.116 (.192) ).082 (.173) ).198 (.258) Family income.303 (.188) ).192 (.163) ).496 (.249) Black.163 (.742) 1.103 (.718).940 (1.032) Female ).095 (.328) ).913 (.317) ).818 (.457) Constant )1.085 (.906) 1.194 (.864) 2.278 (1.252) p<.10; p<.05 Number of observations: 650 Log-likelihood: )275 Pseudo R 2 :.39. Note: Standard errors in parentheses. respondents in the eight-states serves as a natural experiment that can provide leverage on the question of causality by keeping state characteristics constant while allowing candidate contrast to vary. Table 3 reports the parameter estimates. Several results support the proposition that voters do respond to the candidates positions: (i) the effect of abortion policy preferences where there was no candidate contrast cannot confidently be distinguished from zero (p =.29) while there is a substantial effect of abortion policy preferences in the elections where the candidates held different positions (b =.305, p <.01), and (ii) the difference in the two effects (.191) nearly matches the difference reported in Table 1 based on all the elections (.185). Providing a reason for reservation, though, is the fact that the difference between the two estimates of the effect of abortion policy preferences in the eight-state subsample does not meet conventional standards of statistical significance (p =.19). This might be due to the smaller sample size (n = 650) because of the restricted number of states, but knowing for sure awaits additional data and further study. Turning from the relationship between candidates positions and abortion policy voting to the heterogeneity among voters, the results are also revealing. First, the relationship between abortion opinion and vote choice varies substantially with the amount of information voters have about the candidates. This finding that the structure of vote choice is different for the less and better

POLICY VOTING IN SENATE ELECTIONS 195 informed complements those from previous studies of presidential voting. Another question about information effects relates to whether voters ballot choices would be different if they were better informed. Some research suggests that even though the structure of their decisions is different, poorly informed voters make the same choices that they would if they were fully informed (e.g., Lupia 1994; Popkin, 1994). Other studies are less optimistic (Althaus, 2001; Bartels, 1996). To investigate this question, I used the logit estimates in Table 2 to calculate two predicted probabilities of voting Democratic for every voter. One is based on their actual information levels. The second is computed after setting information to its highest level. Thus for every voter it is possible to estimate how much the probability of voting Democratic changes with full (compared to actual) information. If poorly informed voters arrive at the same decisions they would make if they were more informed, then the change will be negligible. However, the results of this exercise indicate that the average movement among those with low information is 7.8 percentage points. Moreover, these changes do not cancel out in the aggregate; the estimated change in the aggregate preferences of the least informed voters is 6.6 percentage points toward the Democrats. 19 Thus it is not only the case that the underlying determinants of vote choice differ across voters, but that information influences the decisions voters make, too. The second dimension of voter heterogeneity concerns motivation, or salience. Because previous attempts to estimate the relationship between salience and policy voting do not simultaneously account for the relationship between information and policy voting, interpreting their results is problematic. Here, with the information/policy voting interaction also included, the salience effects cannot be attributed to information. Thus, the findings indicate that salience, independent of information, is an important factor that moderates the observed level of abortion policy voting. The effects of gender and mentioning abortion are significant, not only because they help to provide a more accurate understanding of abortion policy voting. In addition, they imply that despite the fact that there is only a small... minority of the public that pays great attention to politics and is well informed about it (Zaller, 1992, p. 16), many voters may nevertheless base their choices on policy considerations because Krosnick (1990) has shown that there is not a significant degree of overlap among issue publics. Given that Senate candidates often hold different positions on a number of issues, the findings presented here suggest that there is far more policy voting than one would expect based on the mass public s generally low level of political information. Finally, it is important to consider the generality of the results reported here. One aspect relates to whether the voting patterns revealed in Senate elections exist in other types of elections. At the presidential level, the answer

196 HIGHTON is straightforward. Previous research shows that as the parties pulled apart on abortion, voters in presidential elections followed (Adams, 1997). And, Abramowitz (1995) demonstrates the relevance of salience in the presidential election year (1992) covered here. In House elections, extant research suggests a caveat about generalization. Many House elections (Gronke 2000; Krasno, 1994) barely receive any media attention and inspire little interest among voters. As a result, the differences between elections with and without candidate divergence (Table 1 and Figure 1) would probably be muted, and might be imperceptible, because although the candidates present alternatives on abortion, voters would be less likely to become aware of them. A second facet of generality concerns whether the findings about abortion policy voting apply to other issues. Abortion has become a high profile and easy (Carmines and Stimson, 1980) issue about which many Americans have fixed, or nearly so, views. These characteristics are not common and therefore imply that for many other issues the pattern of results reported in this paper would not be evident. But, as Abramowitz (1995, p. 185) notes, abortion is not alone; it is one of a number of divisive social issues [like gay rights and sexual equality] that have emerged on the political agenda in recent years. Gun control and the death penalty are two other prominent issues that could be added to such a list. An important avenue for future research is to analyze policy voting on issues like these. CONCLUSION One of the main incentives for elected officials to represent and respond to the preferences of their constituents is the prospect of being punished by voters at the ballot box if they do not. Consequently, investigations like this one that attempt to understand the contours of policy voting in elections are not merely about understanding the determinants of vote choice. They speak to questions that are fundamental to theories of representative democracy. The findings indicate that in Senate elections voters divorce candidates from their national parties abortion positions, but do align their ballot choices with the positions of the candidates. At the same time, the evidence of salience effects that exist independently of information effects implies that across issues, politicians are held accountable by different segments of the mass public. Overall, then, the results in this paper indicate that early formulations of the policy voting question ( Do voters cast ballots on the basis of their policy preferences ) implying simple answers of yes or no are inadequate. A variety of factors influence the level of policy voting. Important considerations relate to opportunity, information, and motivation, which mean that future investigations of the representation process would benefit from considering the role of these factors in connecting members of the mass public to their elected officials.

POLICY VOTING IN SENATE ELECTIONS 197 APPENDIX: VARIABLE CODING The individual-level data used for this paper are from the 1990 and 1992 SES (Miller et al., 1999). The primary independent variable used in the analysis is a measure of voters abortion policy preferences. This measure is based on respondents answers to three questions about abortion policy. One asked about the circumstances under which abortion should be legal (all, some, never). A second asked whether individuals favored or opposed parental consent laws for those under the age of 18. The third asked about the use of government funds to cover the costs of abortion for those who cannot afford them. Responses to each of these questions were scaled to range from )1 (least supportive of liberal abortion policy) to +1 (most supportive of liberal abortion policy). A simple additive scale was constructed (alpha reliability =.62) producing a measure ranging from )3 to +3. The policy liberalism scale was constructed in a similar fashion. It is based on the answers to ten questions that asked about the following: the environment, public schools, food stamps, fighting the disease AIDS, child care, defense spending, medical care, government assistance for the homeless and blacks, and the death penalty. Each was coded on a )1 (least liberal) to +1 (most liberal) scale and then an additive scale was constructed (alpha reliability =.68) producing a measure ranging from )10 to +10. Coding the other political disposition variables (party and ideological identification) was straightforward and is described in the text. NOTES 1. Niemi and Weisberg (2001) provide a useful summary of this vast literature. 2. At the same time, the candidate who holds the stronger position has incentives to keep this from happening. 3. Rivers (1988) argues that within a given election context there is a diversity of voting behavior (p. 753). Sniderman (1993) contends that it is systematically misleading to characterize the political reasoning of mass publics en bloc (p. 225). 4. The importance of examining multiple elections, with variation in candidates positions becomes evident when the alternative of analyzing a single election is considered: An issue question may simply identify a subset of voters with a particular political bias; for example, other things being equal, people who favor the ERA may also favor liberal Democrats like Alan Cranston and might have supported him just as consistently had Gann [his opponent] endorsed the ERA, killing it as a campaign issue. Just because an issue position is related to the vote does not mean that the issue itself influenced the vote (Jacobson and Wolfinger 1989, p. 526). Without variation in context, the cause of policy voting remains obscure 5. Although a variable more nuanced than the dichotomy employed in this paper might be desirable, reliably and validly constructing one was not feasible. The analysis is based on the 1990 and 1992 elections because the necessary individual-level survey data was available for these election years. (The survey data covers 1988, too, but I could not complete the codings for these elections likely due to the fact that the Abortion Report was not yet being published which led me to exclude them from my analysis.) Before combining the 1990 and 1992 surveys, I conducted preliminary analyses of the data from each year and found similar results. 6. Candidate contrast is most commonly observed when the Democratic candidate is pro-choice and the Republican is pro-life. Instances of contrast where the Democrat is pro-life and the Republican is pro-choice are rare. Only 2 of the 66 contested elections have this characteristic.

198 HIGHTON Analyzing the correlates of ballot choice for voters in these elections reveals that none of the independent variables has an effect that can confidently be distinguished from zero. Moreover, a likelihood ratio test provides no confidence that the variables together explain vote choice among these voters (p =.43). Drawing any conclusions about the meaning of these results is severely hampered by the small number of elections (n = 2) and voters (n = 95). Because these elections (and voters) are theoretically distinct but provide so few cases, I dropped them from the subsequent analysis. 7. The need to take into account factors related to policy preferences has not been ignored by other scholars. Indeed, most acknowledge its importance. But, data constraints, especially for the analysis of vote choice in subpresidential elections have often been a hindrance to addressing the problem. 8. Independent leaners and weak partisans are coded together at )1 (Republican) or +1 (Democrat). Pure independents and apoliticals are coded 0. 9. Respondents who initially did not place themselves on the scale but in a follow-up question did were coded either )1 (conservative) or +1 (liberal). Moderates and those who did not classify themselves were coded 0. 10. The questions asked about the following: the environment, public schools, food stamps, fighting the disease AIDS, child care, defense spending, medical care, government assistance for the homeless and blacks, and the death penalty. The appendix provides further details on the scale construction. 11. Members of the abortion issue public are more likely to mention abortion in response to the open-ended questions about the candidates. However, it is probably the case that people are more likely to mention abortion when the issue has greater importance in the campaign. Thus the abortion mention measure is unlikely to be a pure indicator of the abortion issue public. It may also tap salience in the campaign. 12. A likelihood ratio test comparing the overall fit of the model reported in Table 1 to one where the effects of all the independent variables are constrained to be equivalent in the two election contexts strongly supports rejecting the null hypothesis of no differences across election types (p =.002). 13. To do this, I varied the values of abortion opinion and candidate contrast while keeping the values of the other independent variables at their actual values. 14. Westlye (1991) and Kahn and Kenney (1999) show that the effect of what I call policy liberalism in this paper is greater in more intense campaigns. Kahn and Kenney (1999) also demonstrate that the effects of ideology and party identification are strengthened in intense elections. 15. This does not mean that campaign intensity is unimportant. Rather, the association between candidate contrast and the level of abortion policy voting cannot be explained by the effect of campaign intensity on the relationship between political attitudes and vote choice. 16. Low information is defined as being at the 10th percentile of the information scale; medium information is the 50th percentile; and high information is the 90th percentile. 17. Does party identification cause abortion opinions or do abortion opinions cause party identification or is there reciprocal causation? Without ample panel data at the individual level over an extended period of time, the question is nearly intractable. However, Adams (1997) suggests that causality flows from abortion positions to party identification. Even so, determining how much of the party identification effect should be attributed to abortion is beyond the scope of this paper. 18. I say completely successful because we do not know what the observed level of policy voting would be in the absence of strategic candidate behavior. Thus, it is possible that strategic obfuscation succeeds by lowering the apparent level of policy voting from a higher (unobserved) level to the one that is observed. 19. Because (a) the information scale is not extensive like those in other research (e.g., Althaus, 1998, 2001), and (b) this paper only examines the interaction between information and one variable

POLICY VOTING IN SENATE ELECTIONS 199 (abortion policy preferences), these estimates should be considered suggestive, showing that information matters, but not putting a precise estimate on how much it matters. Nevertheless, the finding that full information would benefit the Democratic party in Senate elections is consistent with the finding of partisan bias in Senate elections reported by Althaus (2001). REFERENCES Abortion Report. (1990). South Dakota: abortion issue in Senate debate. October 15. American Political Network. Abortion Report. (1992). IL Sen 92: Braun first campaign ad hits the airwaves. August 18. American Political Network. Abramowitz, Alan I. (1981). Choices and echoes in the 1978 U.S. Senate elections: a research note. American Journal of Political Science 25: 112 118. Abramowitz, Alan I. (1995). It s abortion, stupid: policy voting in the 1992 presidential election. Journal of Politics 57: 176 186. Adams, Greg D. (1997). Abortion: evidence of an issue evolution. American Journal of Political Science 41: 718 737. Althaus, Scott L. (1998). Information effects in collective preferences. American Political Science Review 92: 545 558. Althaus, Scott L. (2001). Who s voted in when the people tune out? information effects in congressional elections. In Roderick P. Hart and Daron R. Shaw (eds.), Communication in U.S. Elections: New Agendas, pp. 33 54. Lanham: Rowman & Littlefield. Alvarez, Michael R. (1998). Information and Elections: Revised to Include the 1996 Presidential Election. Ann Arbor: University of Michigan Press. Alvarez, R. Michael, and Jonathan Nagler (1995). Economics, issues and the Perot candidacy: voter choice in the 1992 presidential election. American Journal of Political Science 39: 714 744. Bartels, Larry M. (1996). Uninformed votes: information effects in presidential elections. American Journal of Political Science 40: 194 230. Berelson, Bernard R., Paul F. Lazarsfeld, and William N. McPhee (1954). Voting: A Study of Opinion Formation in a Presidential Campaign. Chicago: University of Chicago Press. Brody, Richard A., and Benjamin Page (1972). Comment: the assessment of policy voting. American Political Science Review 66: 450 458. Campbell, Angus, Philip E. Converse, Warren E. Miller, and Donald E. Stokes (1960). The American Voter. Chicago: University of Chicago Press. Carmines, Edward G., and James A. Stimson (1980). The two faces of issue voting. American Political Science Review 74: 78 91. Converse, Philip E. (1964). The nature of belief systems in mass publics. In David E. Aptor (ed.), Ideology and Discontent, pp. 206 261. New York: Free Press. Delli Carpini, Michael X., and Scott Keeter (1996). What Americans Know about Politics and Why it Matters. New Haven: Yale University Press. Franklin, Charles H. (1991). Eschewing obfuscation: campaigns and the perception of U.S. Senate incumbents. American Political Science Review 85: 1193 1214. Goren, Paul (2002). Policy mood and the American voter. Paper prepared for delivery at the Annual Meeting of the American Political Science Association. Boston. Gronke, Paul (2000). The Electorate, the Campaign, and the Office: A Unified Approach to Senate and House Elections. Ann Arbor: University of Michigan Press. Jacoby, William G. (1986). Levels of conceptualization and reliance on the liberalconservative Continuum. Journal of Politics 48: 423 432.