Policy Confidence and Electoral Punishment: A New Dimension for Understanding Electoral Accountability Alan S. Gerber Yale University Department of Political Science Institution for Social and Policy Studies 77 Prospect Street, PO Box 208209 New Haven, CT 06520-8209 [email protected] 203.432.5232 (voice) Gregory A. Huber Yale University Department of Political Science Institution for Social and Policy Studies 77 Prospect Street, PO Box 208209 New Haven, CT 06520-8209 [email protected] 203.432.5731 (voice) David Doherty Yale University Institution for Social and Policy Studies 77 Prospect Street, PO Box 208209 New Haven, CT 06520-8209 [email protected] 203.432.5057 (voice) Conor M. Dowling Yale University Institution for Social and Policy Studies 77 Prospect Street, PO Box 208209 New Haven, CT 06520-8209 [email protected] 203.432.4811 (voice)
Citizens’ Policy Confidence and Electoral Punishment: A Neglected Dimension of Electoral Accountability ABSTRACT: If voters punish elected officials who adopt policy positions that they disagree with, then representatives should tend to adopt positions close to those of the electorate in anticipation of electoral sanction. Yet, scholars have noted gaps between citizen preferences and the behavior of elected officials. We argue that one important source of the gap between citizen preferences and elected officials’ behavior is that individual citizens understand they are sometimes not well qualified to evaluate policy. Our analysis of a series of experiments shows that citizens’ stated confidence in their own ability to evaluate a policy proposal substantially affects their willingness to punish a representative for their votes on that policy. Our results hold both across individuals (within policy areas) and within individuals (across policy areas) and suggest that, rather than a failure of representation, gaps between citizen preferences and elected officials’ behavior may reflect citizen deference to “expert” decision makers.
1
A guiding assumption of models of electoral government is that voters punish elected officials who adopt policy positions that they disagree with. By this logic, elected officials should tend to adopt positions close to those of the electorate in anticipation of the threat of electoral sanction (e.g., Downs 1957; Mayhew 1974). At the same time, scholars have also noted important gaps between citizen preferences and the behavior of elected officials. For example, citizens are often opposed to trade liberalization efforts, which nonetheless enjoy wide support in legislatures (Kaltenthaler, Gelleny, and Ceccoli 2004). Additionally, there is evidence that the nature of the representation relationship varies across citizens and issue areas. For instance, the opinions of wealthy citizens appear to affect legislator behavior more than the preferences of less wealthy citizens (Bartels 2008), while certain issue domains (particularly social issues) are characterized by greater concordance between citizens’ preferences and the behavior of elected officials. What explains these patterns? In this paper, we argue that one important source of the gap between citizen preferences and elected officials’ behavior is that individual citizens understand they are sometimes not well qualified to evaluate policy. Prior research has identified numerous dimensions to public opinion that are relevant to political accountability: a citizen’s preferred policy (i.e., their “attitude” on it, often operationalized as a preferred point on a line), the importance of the policy dimension to the citizen (for a review, see Miller and Krosnick 2001), the accessibility of the attitude to the citizen (see, for example, Goren 1997; Iyengar and Kinder 1987; Lau 1989; Zaller 1992; Zaller and Feldman 1992), and the citizen’s ambivalence on a given policy (see, for example, Alvarez and Brehm 1995, 1997, 1998; Lavine 2001). There is another component to citizen attitudes about policy evaluations, however, that receives less attention in the literature: how confident (or, certain) the citizen is about his or her own policy preferences.1 We argue that citizens sometimes recognize that, apart from, for example, how important something is, there are many policy areas where ultimately they cannot distinguish good from bad policy proposals. In these 1
We focus on citizens’ confidence/(un)certainty about their own policy positions/ability to evaluate policy in a policy area. This is distinct from much of the previous theoretical and empirical work on attitude certainty and electoral accountability that focuses on citizens’ uncertainty regarding candidates’ positions, and not their own evaluative confidence (for a review of this work, see Miller and Peterson 2004). We discuss this distinction in greater detail in the next section.
2
cases, citizens may be less likely to punish elected officials who take positions contrary to their own because they are uncertain that their policy preferences will lead to their desired outcomes. Therefore, we suggest that this variation in self-identified “evaluative confidence” across citizens and policy domains can help to explain incongruence between public preferences and representatives’ behavior. Rather than a failure of representation, this slippage may reflect citizen deference to “expert” decision makers. To provide a concrete but perhaps humorous example of such deference, consider recent debates about the status of the planet Pluto. In light of the discovery of other planet-like objects larger than Pluto, in 2006 the International Astronomical Union revised its definition of a planet so as to downgrade Pluto from “planet” to “dwarf planet” status. Chaos ensued, and “dissident” astronomers, including the NASA official overseeing a forthcoming mission to Pluto, attacked the decision on both policy and procedural grounds. In Illinois, the birthplace of Pluto’s discoverer, the State Senate even enacted a resolution restoring Pluto’s “full planetary status.” And yet, despite this vocal reaction, nothing has changed, and one would be hard pressed to find any evidence that electoral officials feel any political pressure to take any particular position on this matter. We contend that a simple explanation for this pattern is that even though Americans have opinions about almost everything, from the proper conduct of monetary policy by the Federal Reserve to whether an asteroid is really a planet, those opinions may not hold much weight with elected officials precisely because both voters and officials recognize that these are not issues where policy should necessarily respond to public sentiment. Below we report findings from a series of experiments embedded in a national survey. The analysis shows that citizens’ stated confidence in their own ability to evaluate a policy proposal substantially affects their willingness to reward or punish a representative for their votes on that policy. Notably, these conditioning effects are independent of the conditioning effects of the importance people assign to the policy area. The findings provide important insights into the contours of human decision making and the workings of representative democracy. The fact that individuals are able to understand the limits of their own abilities when making decisions challenges the notion that the mass public is unaware of its own ignorance. Furthermore, such strategic adaption by voters may explain the relative willingness
3
of citizens to choose non-policy decision-making heuristics as a substitute for their own naivety. More broadly, our findings challenge the assumption that mass participation will subvert the achievement of desirable public policy outcomes because voters are often poorly informed or hold opinions contrary to best scientific evidence (Caplan 2007).
1. Confidence in Policy Evaluations and Electoral Accountability To motivate our theorizing about the role of confidence in policy evaluations on electoral accountability, we present a simple model of electoral decision making. The key insight of this model is that when voters are less certain about their policy preferences, their evaluation of a candidate for office will be less responsive to the concordance between the official’s policy position and their own. Confidence, in this model, therefore decreases the uncertainty associated with choosing a particular candidate and causes voters to increase the weight they give to a candidate’s policy positions when evaluating the candidate. Before proceeding to the model, we note that prior scholarship has considered the question of uncertainty in electoral choice more generally. For example, the logic of retrospective voting builds on the notion that promises about future behavior are less credible than is evidence of past behavior. Therefore, retrospection can provide more precise estimates of incumbent rather than challenger behavior (Fiorina 1981; Key 1966). Building on this foundation, more recent work experimentally examines how voters respond to deliberate ambiguity about issue positions by candidates (Tomz and Van Houweling 2009). This scholarship focuses on uncertainty about candidate issue positions, rather than on uncertainty in individual-level evaluations of the same issues (also see Bartels 1986; Enelow and Hinich 1981; Franklin 1991; Page 1978; Palmer and Garner 2010; Peterson 2004; Shepsle 1972). Scholars have also examined the role of belief conflict as a source of uncertainty about one’s own positions (Alvarez and Brehm 1995; Craig and Martinez 2005; Craig, Martinez, Kane, and Gainous 2005; Feldman and Zaller 1992; McGraw, Hasecke, and Conger 2003; Lavine 2001). Not surprisingly, voters who view issues as ambiguous (i.e., not easily resolved due to competing values) have less stable policy opinions. That
4
scholarship has demarcated this form of uncertainty as ambivalence, because it originates not in lack of information about what policy choice is best, but instead in value competition. Our inquiry focuses on the role of individual’s confidence in their own policy positions. In this way, our work is most similar to work in political and social psychology that measures (attitude) certainty by using a “meta-attitudinal” measure that asks respondents how certain they are about their own issue attitudes (see, for example, Alvarez and Franklin 1994; Bassili 1996; Krosnick, Boninger, Chuang, Berent, and Carnot 1993; Krosnick and Schuman 1988; Visser, Krosnick, and Simmons 2003). Our work is distinct from this line of work because we experimentally manipulate the positions taken by an elected official to be evaluated by the respondent. This allows us to isolate the extent to which the effect of (exogenously introduced) policy agreement on candidate evaluations is moderated by individual-level policy confidence. Overall then, we combine the experimental approach employed by Peterson (2004) in his examination of the effect of the certainty (and accessibility) of candidate positions on candidate evaluations with the measurement of individual-level certainty in (and importance of) one’s own issue positions (like Visser, Krosnick, and Simmons 2003).2 1.1 A Model of Electoral Decision Making To focus on the role of confidence in shaping evaluations, we consider an electoral context in which the voter is evaluating an incumbent official. The incumbent I has known characteristics C and takes a policy position X in {A,B}. The voter V is of policy type T={A,B}, and receives utility α>0 if the politician enacts X=T and 0 otherwise. The voter may not know her own policy type, however, but instead observes signal S={A,B}, which is a potentially imperfect indicator of T. This signal is accurate with probability 1>=Q>=.5 (B=T with probability Q). Higher levels of Q therefore indicate greater confidence in one’s policy preferences—holding S fixed, a voter with higher levels of confidence is more likely to actually be of type T=S than one with lower levels of Q. Assuming risk neutrality, the voter’s expected utility associated with the incumbent is therefore 2
Importantly, Peterson (2004) defines “certainty” in much the same way as we do here (“how sure a person is about the attitude they hold,” 513). Peterson’s measure of attitude certainty, however, focuses on certainty about candidate positions.
5
U(C,X,T,S,Q)= F(C) + (X=S)*Q*α + (X≠S)*(1-Q)*α, where F(C) is an unspecified function that maps the incumbent’s (non-policy related) characteristics to expected future behavior. In this equation the term (X=S)*Q*α is the expected reward of following one’s signal, which turns out to be the appropriate policy choice with probability Q. The (X≠S)*(1-Q)*α term is the expected reward for not following her signal, which will turn out to have been the correct decision (because the signal was in error) with probability (1-Q). To isolate the effect of changes in uncertainty, we consider the expected effect of changes in incumbent policy positions on expected voter welfare as Q changes. In particular, the expected utility associated with a(n apparently) congruent incumbent is U(C,X=S,Q)= F(C) + Q*α, while the utility associated with a non-congruent incumbent is U(C,X≠S,Q)= F(C) + (1-Q)*α. Subtracting the first from the second, the surplus utility associated with the congruent candidate is α(2Q-1), which is greater than 0 given that α>0 and Q>=1/2. Thus, the model predicts that a congruent incumbent is evaluated more favorably than a non-congruent incumbent. Taking the derivative with respect to Q yields 2α, which is also positive, indicating that the reward for being congruent (and the punishment for being non-congruent) is larger the more confident the voter is in which policy is better for her. In other words, voter’s who are more certain about which policy benefits them give more reward (punishment) to a candidate who adopts (opposes) that policy. Thus, the central hypothesis to be tested in this paper is that an individual’s policy confidence affects the relationship between policy congruence (between an individual and incumbent) and evaluations of the incumbent. Greater confidence (or, certainty) should result in more positive (negative) evaluations of a congruent (non-congruent) incumbent. This model also illuminates the way in which other factors are likely to affect the evaluation of elected officials. Here, we highlight two such considerations. The first is policy importance, which is represented by α in our model. Given that candidates take positions on many issues, but voters may care
6
about some issues more than others, the importance of issue congruence in a given domain is likely to vary with the weight a voter gives to policy positions in that domain (Krosnick 1988; Krosnick and Abelson 1992). This importance can be thought of as “an individual's subjective sense of the concern, caring, and significance he or she attaches to an attitude” (Boninger, Krosnick, Berent, and Fabrigar 1995, 160). More concretely, holding constant a voter’s certainty, a voter who cares a great deal about abortion policy, for example, will reward (punish) an official who takes her preferred (dispreferred) position regarding abortion more than a citizen who does not believe that abortion policy is as important. In our model, when α is larger, the voter places more weight on an incumbent being congruent in this particular domain than when α is smaller. Of course, in some cases the importance people attach to a given issue and their certainty about their ability to evaluate policies related to that issue may be highly correlated, but analytically the two remain distinct. The model can also accommodate voter decision making on the basis of characteristics other than policy positions. Most prominent is the possibility that voters are inclined to support members of their own party, but not members of other parties. In this case, including measures of partisan congruence in F(C) accounts for such partisan cueing. A more subtle possibility, however, is that partisanship may serve as a substitute for substantive policy knowledge. If voters recognize their own uncertainty about policy and as a result discount their own policy positions when making judgments, they may, as a consequence, give more weight to alternative metrics of decision making, including partisan congruence. While the equation articulated above does not account for such a possibility, in the empirical analysis below it motivates our inclusion of the interaction between certainty and both policy congruence and alternative factors that affect evaluations of candidates in certain model specifications.
2. Experimental Design In order to assess the validity of our model empirically we turned to an experimental setting in which we manipulate candidate positions and other characteristics (detailed below). This approach allows us to isolate the effect of differences in certainty, subject to the caveats we discuss below about the use of
7
observed (rather than experimentally manipulated) variation in certainty. In particular, we implemented a survey experiment embedded in the 2009 Cooperative Congressional Election Study (CCES: Ansolabehere 2010). The CCES is an Internet-based survey that uses a combination of sampling and matching techniques to account for the fact that opt-in Internet survey respondents may differ from the general population on factors such as political interest. This process is designed to approximate a random digit dialing sample.3 Details of the experiment are summarized in Table 1. In addition to gathering background information about respondents’ individual characteristics, the survey included both a pre- and post-stimulus series of survey questions that we employ in our analysis. Summary statistics, exact question wording, and coding rules for all model variables appear in Table A1 in the Appendix. <
> Near the beginning of our survey and prior to the stimulus,4 respondents were asked (in this order) (1) how important they thought each of four policy areas was, (2) whether they were inclined to support or oppose a particular policy proposal in each area, and (3) how confident they were that they could design effective policies in that domain (measured using two items). The four policy areas were teen pregnancy, same sex marriage, job creation, and banking regulation. The order of these four policy areas was randomized once for each respondent, with the same order maintained throughout their survey. Later, toward the end of our survey, respondents were presented with the experimental manipulations embedded in four vignettes. We introduced the vignettes with this text: “Now we would 3
The final survey sample is constructed by drawing a target population sample that is representative of the general population on a variety of characteristics (e.g., gender, age, race, income, education, state of residence, party identification) based on the 2005-2007 American Community Study, November 2008 Current Population Survey Supplement, and the 2007 Pew Religious Life Survey. After administering the CCES survey to more respondents than is required, YouGov/Polimetrix used nearest neighbor matching to select cases to match to the target sample. Weights were then calculated using propensity scores to adjust the final sample to reflect the national public on these demographic and other characteristics. Because our randomization takes places within the selected sample, our analysis is unweighted. For more detailed information on this type of survey and sampling technique see Vavreck and Rivers (2008). 4 The CCES is a collaborative effort among researchers at a number of universities to create a large, national survey. Pooling their resources, researchers contribute to “common content,” or a set of questions that all respondents answer, followed by the “private content,” another set of questions specific to each individual team. Each segment takes about 10 minutes for the respondent to complete, with the common content appearing first. Our survey experiment appeared on our universities private content.
8
like to ask you about votes cast this term by congressional representatives from districts elsewhere in the country. The representatives’ names have been altered.” For each of the four policy areas, we then presented this text, with the Fictional Last Name, Party, and Vote independently and randomly assigned5: [Fictional Last Name (Miller, Williams, Moore, Taylor, or Jones)] is a [Party (Democrat or Republican)] running for reelection to the U.S. House of Representatives in 2010. During this past term, he voted [Vote (for or against)] [Policy proposal]. If you lived in Representative [Fictional Last Name]’s district, would his vote on this bill make you more inclined to vote for him or less inclined to vote for him? Responses to each evaluation question were scored using a horizontal slider that linearly scaled responses from 0 to 100, with the 0 point labeled “much less inclined to vote for him” and the 100 point labeled “much more inclined to vote for him.” The primary dependent variable in our analysis is responses to this question, which we label Support for Representative. 2.1 Manipulation Verification Our first task is to verify that respondents reacted to the experimentally manipulated policy and party positions in the vignettes. For the policy manipulation, we do so by comparing support for the representative when he took a position in agreement with the respondent relative to when his position was in disagreement with the respondent’s. (Agreement is coded when the respondent has a strong or weak policy position in the same direction as the representative.) This analysis is restricted to cases where a respondent took a position in favor of or against a particular position.) This difference in average support is reported in the first row of Table 2. Across all four policy areas we find that representatives randomly assigned to a position in agreement with the respondent are evaluated more favorably than a respondent assigned to a position in disagreement with the respondent. The magnitudes of these effects are also quite large—from 27 points (on a 100 point scale) in the case of banking regulation to 50 points in the case of same sex marriage. All differences are statistically significant at p<.01. <
> 5
Last name is Miller, Williams, Moore, Taylor, or Jones with equal probability. (We did not find any differential effects of the last names. Results available upon request.) Party is Democrat or Republican with equal probability. Vote is for or against with equal probability.
9
By contrast, the effect of party agreement is more modest. Here we report differences in average support when the representative is randomly assigned to share the respondent’s partisanship relative to when he is of a different party than the respondent. (We code party agreement as existing when a respondent is a strong or weak partisan of the same party as the representative and party disagreement as existing when a respondent is a strong or weak partisan of the opposite party; independents and partisan “leaners” are excluded in this analysis). Shared partisanship is worth between 5 and 9 points on the 100 point scale. While these differences are statistically significant at p<.01, they are substantively much smaller than the differences associated with variation in policy agreement. This manipulation verification demonstrates that people paid attention to our vignettes and responded to party and policy position cues.
3. Results: The Effect of Variation in Confidence on Evaluations of Representatives The preceding analysis demonstrates that survey respondents appear to respond meaningfully to variation in the policy position and party membership manipulations embedded in our experiment. We next assess whether the degree of this responsiveness varies, as we predicted, with variation in a respondent’s selfassessed confidence to construct and evaluate policy in a given domain. To foreshadow, we find that both across individuals and for a given individual, across policy domains, greater confidence in one’s ability to evaluate policy is associated with a greater willingness to reward (punish) representatives who take (in)congruent policy positions. 3.1 Bivariate Analysis In Table 3, we begin by displaying average policy confidence and importance across policy domains. Respondents appear to be most confident in the areas of job creation and same sex marriage policy, less confident in the area of teen pregnancy policy, and least confident in their ability to assess desirable policy in the area of banking regulation. By contrast, respondents believe job creation is by far the most important policy area, and then rank the remaining areas in descending importance as banking regulation, teen pregnancy, and same sex marriage. Thus, the average effects of policy- and party-disagreement shown in Table 2a may reasonably reflect variation in either or both of these evaluations—some
10
respondents may be confident but not care about a particular policy, others may care a great deal but lack confidence, and so on. <
> To isolate the effects of variation in confidence and importance, we also present the effect of differences in policy agreement (the same difference reported in Table 2) by levels of confidence and importance. Respondents are scored as high confidence (importance) if their score is above the midpoint of the confidence (importance) scale and low confidence (importance) if it is below it. These are therefore difference-in-difference estimates. (The effect of variation in confidence and importance on the effect of party-agreement is reported in Table A2 in the Appendix.) In all cases, the effect of policy agreement is larger for more confident respondents. For example, in the case of teen pregnancy, policy congruence increases the representative’s evaluation by 35 points among the more confident, but only by 25 points among the less confident. This difference, about 10 points, is statistically significant at p<.01. For same sex marriage the magnitude of the difference is larger (19 points) and it is also statistically significant at p<.01. In the case of job creation the difference is more modest—only 5 points—but it remains statistically significant at p<.10. Finally, for banking regulation, the difference between the more and less confident is only 3 points and is not statistically significant. In order to provide an assessment of the relative magnitude of these differences, parallel results for variation in importance are also presented. For two policy areas—same sex marriage and banking regulation—we find relatively large (16 and 12 points, respectively) and clearly statistically significant differences between those who regard an area as important relative to those who do not. In those cases the effect of policy agreement is larger among those who think a policy area is important. This is also true in the area of job creation (the difference-in-differences estimate is 9, but the p-value is only .19), but in the case of teen pregnancy the estimate is near zero and negative. One interesting pattern that emerges from this comparison of the effect of policy confidence and importance is that the moderating effects of these factors appear to vary across policy areas. In the case of same sex marriage, confidence and importance both appear material, but for teen pregnancy, confidence
11
is far more material, whereas in the case of banking regulation, importance matters more. Given that confidence and importance may be systematically related, we next consider the relative effect of each in a multivariate context. 3.2 Across-individual (within policy area) analysis Our first analysis strategy exploits variation across respondents in policy confidence. In particular, building on our above theoretical exposition, we present models where we allow the effect of policy and party agreement on evaluation of the representative to vary with self-assessments of policy confidence and importance. To make the subsequent presentation more compact, we begin by defining two vectors: D and P. D is a vector of respondent demographic characteristics that we believe might independently alter evaluations of representatives. In particular, D includes respondent age (and age squared, to allow for non-linearity), race (omitted category is White, indicators for Black, Hispanic, and other non-White), education (scaled 0-1), and income (scaled 0-1, along with a separate indicator for income refused). P is a vector of policy-relevant characteristics that deserves more careful exposition. We posit that representatives may be evaluated differently on the basis of the respondent’s policy positions (coded from -1 [strongly oppose] to 1 [strongly support]) and partisanship (coded from -1 [strong Democrat] to 1 [strong Republican])6, the representative’s (experimentally manipulated) policy positions and partisanship, and the concordance between the respondent and the representative’s policy positions and partisanship. Thus, P is constructed as: P = Policy Position + Party + Rep’s Policy Position + Rep’s Party + Policy Position*Rep’s Policy Position + Party*Rep’s Party. We refer to the term Policy Position*Rep’s Policy Position as “policy agreement” (or, interchangeably, “policy congruence”) and it is scaled from -1 to 1, with 1 indicating the respondent has a strong policy position in agreement with the representative’s and -1 indicates the respondent has a strong policy position in opposition to the representative’s. Similarly, we refer to the term Party*Rep’s Party as “party agreement,” and it is scaled from -1 to 1, with 1 indicating the respondent has a strong party identification 6
Both weak and leaning Democrats (Republicans) are scored -.5 (.5). Independents are scored 0.
12
shared by the representative and -1 indicates the respondent has a strong party identification in opposition to the representative’s. A regression of the form (1) Evaluation of Representative = B0 + D + P + e is shown in appendix Table A3 and confirms our earlier manipulation verification (see Table 2) that representatives are rewarded for having the same issue positions and the same partisanship as a respondent. Our focus is on the effect of confidence on the relative weight given to those policy-relevant factors listed in P. In particular, we are most interested in whether confidence increases the weight respondents give to the congruence between a respondent’s policy position and those of the representative when evaluating him. Thus, we begin by estimating this equation: (1’) Evaluation of Representative = B0 + D + P + Confidence + Importance + Confidence*P + Importance*P + e. Equation (1’) allows there to be a direct effect of both confidence and importance on the evaluation of the representative. More importantly, however, it also allows the weight given to all policy-relevant characteristics to vary independently with both confidence and importance. For each policy area, estimates based on equation (1’) appear in Table 4. Before discussing those results, we note that while we are most interested in the interaction between Confidence and Policy Position*Rep’s Policy Position, this specification is flexible enough to allow there to be other (not theoretically specified by us) effects of confidence on other covariates.7 Including those additional interactions with confidence is a conservative strategy that ensures our results are not driven by some other mechanism associated with variation in
7
We have also tested for a three-way interaction between confidence, importance, and the measures (including policy agreement) that appear in vector P. Including these additional interactions does not substantively alter the results for the interaction between confidence and policy agreement that we focus on here. Thus, for reasons of presentational simplicity, we confine our presentation to models with twoway interaction terms.
13
confidence on another variable that happens to be correlated with Policy Position*Rep’s Policy Position (e.g., Policy Agreement).8 <
> Examining Table 4, we find support in three of the four policy domains for the claim that selfassessed variation in confidence affects the relationship between policy agreement and the evaluation of the representative. (In the other policy area—banking regulation—the effect of confidence is to increase the weight placed on policy agreement, but this effect is relatively small and is not statistically significant.) In the area of teen pregnancy, for example, among respondents with very low levels of confidence (0 on the confidence scale), a representative whose policy positions are congruent with those of the respondent is graded 37 point higher (on the 100 point scale) than a representative whose policy positions are incongruent (policy congruence ranges from -1 to 1). Among the most confident (1 on the confidence scale), the difference is instead 48 points, or more than 30% larger than among the least confident.9 These estimates, along with those for the other three policy areas, are displayed graphically in Figure 1. The p-value of the coefficient on the interaction between confidence and policy agreement is less than .05 for a two-tailed test (two-tailed tests are reported, despite our directional hypothesis). Similarly, comparing low (confidence=0) and high (confidence=1) confidence individuals in the other policy areas, the effects of policy congruence are 34 and 58 points (p<.01, an increase of 70%), respectively, for same sex marriage, 24 and 34 points (p<.10, 44%) for job creation, and 17 and 19 points (p>.10, 14%) for banking regulation. Banking stands out relative to the other areas because the interaction between confidence and policy congruence is statistically insignificant. <
> Two other findings in the table are also worth highlighting. The first is that the effect of importance on the weight given to policy congruence is statistically significant only in a single area: same 8
While we randomly assign the representative’s policy position, we do not assign the respondent’s policy position, and so we have to be sensitive to the possibility of naturally occurring covariation between the respondent’s policy position and other variables. 9 This is calculated as (1-(-1))*18.5 + (1-(-1))*5.7 = 48.4.
14
sex marriage. Those who think same sex marriage is highly important (importance=1) reward policy congruence by 15 points more than those who do not (importance=0, effect of congruence is 34), an increase of about 44%. The second finding we note is that for banking regulation, the interaction between importance and party agreement is anomalous relative to all other policy areas. Those who think banking regulation is important rely more on party cues than those who do not. In this particular policy area, however, 85% of respondents report banking regulation is either “somewhat important” or “one of the most important issues,” with 49% believing it is “one of the most important issues.” Additionally, only 11% of the sample opposes the particular policy we presented, and only 33% of the sample is above the midpoint of the policy confidence scale in this area. Putting this together, we have a policy area where most respondents support a policy, but are not sure about their ability to evaluate it despite the fact that they think the problem area is very important. In this instance, they rely on party cues. While we are reticent to over-interpret this particular finding ex post, it is consistent with the general idea that party or other cues may substitute for substantive policy knowledge, not because citizens do not have policy opinions, but because they know enough not to trust them (Lupia 1994). When citizens are unwilling to trust their own policy instincts, the cue provided by partisanship is a relatively low cost and on-average relatively unbiased estimate of one’s own preferences. 3.2.1 Robustness of across-individual analysis We examine the robustness of these results to different model specifications and sample restrictions in Table 5. For each policy area, we first repeat the Table 4 specification after adding interactions between each of the demographic measures in D and policy agreement. This specification, which is shown in columns (R1), (R4), (R7), and (R10), allows us to ascertain whether it is confidence per se, rather than some other correlate of confidence, that explains the greater influence of policy congruence on evaluations among the more confident. For teen pregnancy, the coefficient on Confidence*Policy Position*Rep’s Policy Position drops in size and is no longer statistically significant. In the case of same sex marriage and job creation, the interaction coefficient and its precision is essentially unchanged. Finally, in the case of banking regulation, the interaction remains statistically
15
insignificant and becomes even smaller in absolute magnitude. Thus, the results from this model specification suggest that it is confidence, and not some other factor that we control for, that affects the relationship between policy agreement and evaluations of representatives. <
> The next extension we consider is whether our results are influenced by the potential correlation between confidence and position-taking. In particular, suppose that only confident respondents were willing to take clear positions on the different policy proposals. In this case, we might be misinterpreting the importance of confidence by ignoring the possibility that confidence could affect one’s policy opinions rather than the weight given to those opinions. For this reason, in columns (R2), (R5), (R8), and (R11) we repeat the Table 4 specification after restricting our sample to those respondents who took a clear position in favor of or against (1 or -1) a given policy. Except for banking regulation, restricting our attention to those taking strong positions increases the size of the coefficient for the interaction between confidence and policy congruence, which suggests that confidence is important in explaining variation in the weight respondent’s give to policy congruence even after accounting for whether a respondent is willing to take a strong policy position. Finally, the last extension we consider is whether our results are somehow misleading because of the correlation between policy confidence and policy importance. To isolate the role of importance relative to confidence, we again repeat the Table 4 specifications, this time restricting our sample to those respondents who thought the policy area was very important (importance=1). This restriction, reported in columns (R3), (R6), (R9), and (R12), removes all variation in importance, and so we drop those terms from the estimated models. Once again, we find that even among those who view a given policy domain as highly important, the more confident punish policy incongruence more than the less confident. For same sex marriage and job creation the results are highly similar to those reported in Table 4, while for teen pregnancy the effect of the interaction between confidence and policy congruence grows by about 100%. Finally, for banking regulation, we continue to observe null results.
16
This set of robustness checks leaves us reasonably certain that across individuals (within a policy area) confidence in one’s own ability to evaluate policy in a given are alters the relationship between policy agreement and evaluations of representatives, with greater policy confidence resulting in more negative evaluations of representatives who oppose the respondent’s policy position. 3.3 Across-policy area (within individual) analysis One potential limitation of the preceding analysis is that it does not account for the possibility that individuals differ in their pattern of survey responses for reasons unspecified in our model. For example, some individuals may simply express more confidence across all policy domains because of the way they interpret the confidence questions and not because they are actually more willing to punish on the basis of their opinions across domains. Alternatively, average confidence could be correlated with some (unobserved) factor that inclines individuals toward a greater willingness to punish across policy domains. In this case, we would mistakenly attribute to variation in confidence the role of some other covariate that is omitted from our model. For these reasons, in this section we present a series of models where we attempt to account for individual-level variation in average confidence levels. We do so in three ways. First, we begin by analyzing the effect of confidence on the importance of policy congruence by comparing, within a given individual, the importance of policy congruence on the evaluation of the representative for their most and least confident policy domain. To conduct this analysis, we first discard all individuals who do not have unique maximum and minimum confidence policy areas (e.g., individuals who have relatively low variation in their confidence across policy domains). We then estimate the following equation: (2) Evaluation of Representative = B0 + Policy Area + Policy Agreement*Policy Area + Policy Agreement*Most Confident Policy + Most Confident Policy + Individual Fixed Effect. In this specification, we have two observations for each individual (for their most and least confident policy domains). We then account for all static individual characteristics of the individual through the incorporation of individual-level fixed effects. Additionally, we allow there to be average differences in
17
evaluations and average differences in the effects of policy congruence across policy domains. Our theory predicts a positive coefficient on the interaction between confidence and congruence. Results employing this specification appear in column (1) of Table 6. (Note that to save space we do not report the coefficients for the policy area identifiers or the policy area x policy agreement interactions. Suppressed coefficients are listed in the bottom rows of the table.) We find that policy agreement had a larger effect in an individual’s most confident policy domain than in her least confident domain, and this effect is statistically significant. Note that because our specification includes individualspecific fixed effects, policy area fixed effects, and the interaction between policy area and congruence, we can rule out the possibility that our earlier results are caused by some individuals, on average, rewarding congruence more than others or some policy domains intrinsically being seen as more confident. It is also interesting that there is substantial disagreement across individuals in terms of which policy domains they are most confident about. For example, for the 376 respondents included in the column (1) specification (who have unique most and least confident policy areas), 17% are most confident about teen pregnancy, 40% about same sex marriage, 34% about job creation, and 10% about banking regulation. Even among those individuals who agree about what policy area they are most confident about, they differ in which policy area they are least confident about. Of those most confident about jobs, for example, 27% are least confident about banking, 59% about same sex marriage, and 32% about teen pregnancy. <
> A limitation of the specification presented in column (1) is that we discard a great deal of data because we examine only the most and least confident policy areas for individuals who have unique maximum and minimum confidence levels. In columns (2) through (7), by contrast, we retain our original dataset but pool all four policy areas together. To account for individual-level differences in average confidence (and importance), in all columns we mean center confidence (and importance) for each individual. In other words, we scale confidence so that higher scores are not those where an individual is more confident than another person, but for domains where she is more confident than she is in the other
18
domains. We label this rescaled measure “relative” confidence. Additionally, as with column (1), in all specifications we include policy-area specific fixed effects to account for average differences in support across policy domains. This specification is therefore nearly identical to that shown in equation (1’) above, but with confidence and importance rescaled and policy domain indicators added. In column (2), we find that the coefficient on relative confidence * policy agreement is positive, statistically significant, and large. A .5 unit increase in relative confidence (e.g., from .25 units below the individual’s average confidence to .25 units above it) is associated with a 10 point increase in the relative importance of policy agreement, and this effect is highly statistically significant.10 Allowing the effect of the demographic variables in D to vary across policy domains in column (3) has almost no effect on the magnitude or precision of this estimate. In column (4) we go one step further and replace the core position, confidence, importance, representative’s position and party, and party and policy agreement terms with those variables interacted with indicators for each of the four policy domains (a total of 21 additional terms). This reduces by about 60% the estimated effect of the relative confidence * policy agreement term, but does not effect the precision of the estimate, which remains positive and statistically significant. Finally, in columns (5) through (7), we repeat with slight modifications the specifications from columns (2) through (4) after adding individual-level fixed effects to the analysis. The pattern of coefficients in these three columns nearly mirrors those found in the columns without fixed effects, and so we defer discussing them other than to note that they provide further evidence that our results are robust to specifications that examine variation in confidence across policy domains for a given individual.
4. Discussion and Conclusion The foregoing analysis provides, to our knowledge, the first evidence that although citizens have policy opinions in many areas, they recognize their relative lack of expertise when deciding whether to act on those opinions to evaluate elected officials. Our experimental evidence shows that individuals who are 10
This is calculated as .5 * 2 * 10.2 = 10.
19
more confident in their policy opinions are more inclined to punish officials for taking incongruent policy opinions, and a given individual is more inclined to punish in this manner in policy domains where she is more confident than in those where she is less confident. In this section we discuss the implications of this finding for models of citizen decision making and patterns of electoral competition. We also discuss the potential limitations of our analysis. The first implication of our model is to better illuminate the contours of mass opinion. Heretofore, scholars have documented that citizens vary in their opinions and in the relative importance they give to different policy opinions (Grynaviski 2003). However, a limitation of that research is that it treats opinions as something that citizens themselves uniformly respect, while our research suggests that in fact citizens recognize that those opinions are sometimes not to be trusted or acted upon. This dimension to public opinion has been largely neglected (but see Visser, Krosnick, and Simmons 2003). The capacity for introspection is a quality overlooked in prior accounts of the sources and consequences of citizen opinions. Citizens may on average be ill-informed, but to assume that a lack of knowledge undercuts democratic performance or generates undesirable policy outcomes follows only if citizens are willing to act on those opinions, an assumption left untested in prior research. Additionally, we find that citizens hold opinions despite their acknowledged lack of confidence in those opinions. One possibility is that individuals who are aware of their lack of knowledge turn to alternative sources—elite shortcuts, for example, to form opinions when they are unsure of their own preferred policies. By contrast, we find that citizens continue to maintain their own opinions—and express them in surveys—even when they acknowledge their lack of expertise and are unwilling to act on them. (We do find that in the area of banking regulation, the policy area where respondents are on average least confident, citizens do appear to rely more heavily on partisan cues to evaluate representatives. However, at the individual-level, lower confidence is not associated with giving greater weight to those partisan cues.) This willingness to express and maintain opinions despite a lack of confidence in those opinions also has implications for electoral competition. In particular, if elected officials must anticipate the
20
consequences of taking positions and their behaviors in office, our data helps explain why elected officials may sometimes be willing to buck mass opinion. In short, elected officials need not respond to expressed public opinion if citizens are not willing to act on those opinions. In this account, the apparent importance of interest groups and issue publics may not originate so much in intensity of interests, but instead in a greater willingness to act on underlying beliefs. Arnold (1990), for example, notes that legislators need to anticipate the activities of “potential publics” when their unpopular actions will be revealed to previously uninformed citizens. Such an account is accurate, however, only if citizens will then act on their newfound knowledge. But, according to our results, many do not. More generally, considering policy confidence and importance as separate dimensions helps to illuminate the ways in which politicians think about the electorate when considering the political consequences of their decisions. For example, we show in Table 7 that, in our sample, more confident citizens are both more “extreme” and more liberal than the citizenry as a whole in the areas of teen pregnancy and same sex marriage. This comparison holds true even among the subset of citizens who regard these policy domains as important. For teen pregnancy, average support in the entire sample is .1 on a scale that ranges from -1 to 1 and about 44% of the entire sample takes a position either strongly supporting or opposing the policy. Among those who are more confident, the average support rises to .25 and the proportion holding extreme opinions rises to 62%. Similarly, for same sex marriage, the electorate as a whole is on average opposed to the policy, but among those who regard the policy area as important, the public is in fact on average supportive. Focusing on those who regard same sex marriage as both important and who are confident in this policy area, we see that support grows even more and fully 96% of the public holds opinions strongly favoring or opposing it. <
> As Fiorina and others have noted, Americans are on average relatively moderate (e.g., Fiorina, Abrams, and Pope 2004). But in many salient policy domains, this average moderation may mask electorally relevant polarization. Beyond the focus of our study, for example, it would not surprise us to learn that among those who regard abortion policy as important and are confident in their opinions,
21
attitudes are more polarized than among the electorate at large. Such polarization among electorally relevant voters has obvious consequences because it contravenes the general tendency to move toward the ideological center: Doing so makes sense only if voters will reward such moderation, but we show that the voters most confident in their opinions will likely view such moderation skeptically while less confident ones will not necessarily punish non-moderation. Politicians therefore have stronger incentives to establish electoral coalitions built from coalitions of citizens who both care deeply about particular policy domains and also act on their expressed opinions. Of course, discussing variation in confidence raises an important subsequent question: Who is confident, and in what policy domains? Does confidence overlap with beliefs about policy importance? Where does confidence come from? While detailed analysis is beyond the scope of the current paper, we find that certain factors (income, education, and being male) are associated with being more confident across all four policy domains, while the effects of other demographic measures like age and race varies substantially across the different policy areas.11 The finding that income and education are associated with greater confidence is generally consistent with the notion that better educated and wealthier individuals have the resources to more fully engage politics (Verba, Schlozman, and Brady 1995), although the mechanism appears to be more than simply mobilization. (Additionally, because of our experimental design, we can also rule out the possibility that those factors affect the knowledge the respondent has of the representative’s behavior, a factor that may be more relevant outside of the lab- or survey-context.) We close by noting three potential limitations of our analysis. The first concerns the domain of our analysis. We consider only four policy areas, and then only at a single point in time. Whether results for other domains or in other contexts would be similar is a matter worthy of additional study. Second, we must consider issues of external validity: Participants in our survey evaluate fictionalized representatives taking policy positions on issues where we have previously elicited the respondent’s own policy opinions. The dynamics we observe may not be as visible in settings where respondents choose whether to become
11
Statistical analysis is presented in Appendix Table A4.
22
informed about legislator behavior and where other informational cues beyond party and position are available. Finally, we acknowledge that while we are primarily interested in the importance of confidence in one’s policy positions (or, evaluative confidence), we do not manipulate it in our experiment. It is possible that, although we examine the data within as well as across subjects, confidence levels are a proxy for some neglected alternative dimension of respondent attitudes. That said, the analysis we conduct is only as persuasive as our efforts to address alternative explanations for the patterns we observe. We went to some lengths to consider the robustness of our results to alternative model specifications and within- and across-person analysis. We cannot by design, however, rule out every feasible alternative explanation for our finding that policy confidence increases the weight given to policy congruence when evaluating elected officials. These concerns aside, our more general effort has been to demonstrate that citizens appear to understand and act on the basis of their own confidence in making policy judgments. Such introspection has important consequences for interpreting observed patterns of citizen opinions, the role of mass opinion in guiding electoral behavior, and models of citizen decision making.
23
References Alvarez, R. Michael, and John Brehm. 1995. “American Ambivalence Towards Abortion Policy: Development of a Heteroskedastic Probit Model of Competing Values.” American Journal of Political Science 39: 1055-1082. Alvarez, R. Michael, and John Brehm. 1997. “Are Americans Ambivalent Towards Racial Policies.” American Journal of Political Science 41: 345-374. Alvarez, R. Michael, and John Brehm. 1998. “Speaking in Two Voices: American Equivocation about the Internal Revenue Service.” American Journal of Political Science 42: 418-452. Alvarez, R. Michael, and Charles H. Franklin. 1994. “Uncertainty and Political Perceptions.” Journal of Politics 56: 671-688. Ansolabehere, Stephen, Cooperative Congressional Election Study, 2009: COMMON CONTENT. [Computer File] Release 2: March 2, 2010. Cambridge, MA: M.I.T. [producer] Arnold, R. Douglas. 1990. The Logic of Congressional Action. New Haven, CT: Yale University Press. Bartels, Larry, M. 1986. “Issue Voting Under Uncertainty: An Empirical Test.” American Journal of Political Science 30: 709-728. Bartels, Larry, M. 2008. Unequal Democracy: The Political Economy of the New Gilded Age. Princeton, NJ: Princeton University Press. Bassili, John N. 1996. “Meta-judgmental versus Operative Indexes of Psychological Attributes: The Case of Measures of Attitude Strength.” Journal of Personality and Social Psychology 71: 637-653. Boninger, David S., Jon A. Krosnick, Matthew K. Berent, and Leandre R. Fabrigar 1995. “The Causes and Consequences of Attitude Importance.” In R. E. Petty and J. A. Krosnick, eds., Attitude Strength: Antecedents and Consequences, 159-190. Hillsdale, NJ: Erlbaum. Caplan, Bryan. 2007. The Myth of the Rational Voter: Why Democracies Choose Bad Policies. Princeton, NJ: Princeton University Press. Craig, Stephen C., and Michael D. Martinez, eds. 2005. Ambivalence and the Structure of Political Opinion. New York: Palgrave. Craig, Stephen C., Michael D. Martinez, James G. Kane, and Jason Gainous. 2005. “Core Values, Value Conflict, and Citizens' Ambivalence about Gay Rights.” Political Research Quarterly 58: 5-17. Downs, Anthony. 1957. An Economic Theory of Democracy. New York: Harper and Row. Enelow, James, and Melvin J. Hinich. 1981. “A New Approach to Voter Uncertainty in the Downsian Spatial Model.” American Journal of Political Science 25: 483-493. Feldman, Stanley, and John Zaller. 1992. “The Political Culture of Ambivalence: Ideological Responses to the Welfare State.” American Journal of Political Science 36: 268-307.
24
Fiorina, Morris P. 1981. Retrospective Voting in American Elections. New Haven, CT: Yale University Press. Fiorina, Morris P., Samuel J. Abrams, and Jeremy C. Pope. 2004. Culture War? The Myth of a Polarized America. New York: Pearson Longman. Franklin, Charles H. 1991. “Eschewing Obfuscation? Campaigns and the Perception of U.S. Senate Incumbents.” American Political Science Review 85: 1194-1214. Goren, Paul. 1997. “Political Expertise and Issue Voting in Presidential Elections.” Political Research Quarterly 50: 387-412. Grynaviski, Jeffrey. 2003. “Do Issue Publics Exist? An Application of Bayesian Mixture Models.” Presented at the American Political Science Association meetings, Philadelphia, PA. Iyengar, Shanto, and Donald R. Kinder. 1987. News that Matters. Chicago: University of Chicago Press. Kaltenthaler, Karl C., Ronald D. Gelleny, and Stephen J. Ceccoli. 2004. “ Explaining Citizen Support for Trade Liberalization.” International Studies Quarterly 48: 829-852. Key, V. O. 1966. The Responsible Electorate: Rationality in Presidential Voting, 1936-1960. Cambridge, MA: Harvard University Press. Krosnick, Jon A. 1988. “The Role of Attitude Importance in Social Evaluation: A Study of Policy Preferences, Presidential Candidate Evaluations, and Voting Behavior.” Journal of Personality and Social Psychology 55: 196-210. Krosnick, Jon A., David S. Boninger, Yao C. Chuang, Matthew K. Berent, and Catherine G. Carnot. 1993. “Attitude Strength: One Construct or Many Related Constructs.” Journal of Personality and Social Psychology 65: 1132-1151. Krosnick, Jon A., and Robert P. Abelson. 1992. “The Case for Measuring Attitude Strength in Surveys.” In J. M. Tanur, ed., Questions About Questions: Inquiries into the Cognitive Bases of Surveys, 177-203. New York: Russell Sage. Krosnick, Jon A., and Howard Schuman. 1988. “Attitude Intensity, Importance, and Certainty and Susceptibility to Response Effects.” Journal of Personality and Social Psychology 65: 11321151. Lavine, Howard. 2001. “The Electoral Consequences of Ambivalence Toward Presidential Candidates.” American Journal of Political Science 45: 915-929. Lau, Richard R. 1989. “Construct Accessibility and Electoral Choice.” Political Behavior 11: 5-32. Mayhew, David R. 1974. The Electoral Connection. New Haven, CT: Yale University Press. McGraw, Kathleen M., Edward Hasecke, and Kimberly Conger. 2003. “Ambivalence, Uncertainty, and Processes of Candidate Evaluation.” Political Psychology 24: 421-448.
25
Miller, Joanne M., and Jon A. Krosnick. 2001. “The Origins of Policy Issue Salience: Sociotropic Importance for the Nation or Personal Importance for the Citizen?” Unpublished Manuscript. The Ohio State University. Miller, Joanne M., and David A. M. Peterson. 2004. “Theoretical and Empirical Implications of Attitude Strength.” Journal of Politics 66: 847-867. Page, Benjamin I. 1978. Choices and Echoes in Presidential Elections. Chicago: University of Chicago Press. Palmer, Harvey D., and Andrew D. Garner. 2010. “Accounting for Candidate Obfuscation and Electoral Context when Modeling Issue Voting Under Uncertainty.” Presented at the Midwest Political Science Association meetings, Chicago, Il. Peterson, David A. M. 2004. “Certainty or Accessibility: Attitude Strength in Candidate Evaluations.” American Journal of Political Science 48: 513-520. Shepsle, Kenneth A. 1972. “The Strategy of Ambiguity: Uncertainty and Electoral Competition.” American Political Science Review 66: 555-568. Tomz, Michael, and Robert P. Van Houweling. 2009. “The Electoral Implications of Candidate Ambiguity.” American Political Science Review 103: 83-98. Vavreck, Lynn, and Douglas Rivers. 2008. “The 2006 Cooperative Congressional Election Study.” Journal of Elections, Public Opinion and Parties 18: 355-366. Visser, Penny S., Jon A. Krosnick, and Joseph P. Simmons. 2003. “Distinguishing the Cognitive and Behavioral Consequences of Attitude Importance and Certainty: A New Approach to Testing the Common-factor Hypothesis.” Journal of Experimental Social Psychology 39: 118-141. Verba, Sidney, Kay Lehman Schlozman, and Henry E. Brady. 1995. Voice and Equality: Civic Voluntarism in American Politics. Cambridge: Harvard University Press. Zaller, John. 1992. The Nature and Origins of Mass Opinion. New York: Cambridge University Press. Zaller, John, and Stanley Feldman. 1992. “A Simple Theory of the Survey Response: Answering Questions versus Revealing Preferences.” American Journal of Political Science 36: 579-616.
26
Table 1. Experimental Design and Question Wording Demographics (age, race, gender, education, income) and party affiliation. Pre-stimulus (see below for policy areas and descriptions) Importance Regardless of what you think should or should not be done in each of the following areas, how important (relative to all other issues – not just those listed here) are the following issues to you? [policy area] Position To what extent would you support or oppose a policy that would [policy description]? Confidence 1 How confident are you that you would be able to design effective policy in the area of [policy area] Confidence 2 How confident are you that you have the knowledge and expertise to evaluate policy in each of the following areas? [policy area] Stimulus / Outcome [Name] is a [Party] running for reelection to the U.S. House of Representatives in 2010. During this past term, he voted [Position] a bill that would [policy description]. If you lived in Representative [Name]’s district, would his vote on this bill make you more inclined to vote for him or less inclined to vote for him? Name: [Miller/Williams/Moore/Taylor/Jones] Party: [Democrat/Republican] Position: [for/against] Policy Areas Policy Descriptions Teen Pregnancy provide federal funding to teach high school students about contraception use as a way to prevent teen pregnancy Same Sex Marriage legalize same sex marriage Job Creation fund public works programs–such as building highways and bridges–as a way to create jobs Banking Regulation require those trading in asset-backed securities – investment instruments that pool together a large set of smaller investments – to report their trades to a regulatory authority
Table 2. Manipulation Verification: Effect of Policy and Party Agreement on Representative Evaluation Difference in Mean Evaluation Teen Pregnancy Same Sex Marriage Job Creation Banking Regulation Policy Agreement - Policy Disagreement 29.4 50.2 35.3 27.0 Party Agreement - Party Disagreement 4.9 5.4 6.5 8.8 Note: All differences are statistically significant (pairwise t-tests, unequal variance) at p<.001. Unweighted analysis. Analysis excludes respondents who did not take a policy position. Policy Agreement is if the representative supports (opposes) and respondent strongly or somewhat supports (opposes) the policy. Policy Disagreement is if the representative supports (opposes) and respondent strongly or somewhat opposes (supports) the policy. Party Agreement is if the representative and respondent share partisanship, where respondent is strong or weak partisan. Independents and leaners excluded. Party Disagreement is if the representative and respondent do not share partisanship, where respondent is strong or weak partisan. Independents and leaners excluded. Source: 2009 CCES.
Table 3. Bivariate Analysis: Effect of Policy Agreement on Representative Evaluation by Policy Confidence and Importance Average Confidence Importance
Teen Pregnancy 0.50 0.62
Same Sex Marriage 0.54 0.45
Job Creation 0.55 0.92
Banking Regulation 0.42 0.76
Difference in Mean Evaluation, Policy Agreement - Policy Disagreement High Confidence Low Confidence Difference in Difference p-value (two-tailed)
Teen Pregnancy 35.2 25.3 9.9 0.00
Same Sex Marriage 58.3 38.9 19.4 0.00
Job Creation 36.6 31.5 5.1 0.07
Banking Regulation 29.5 26.4 3.0 0.34
28.8 30.8 -2.0 0.51
58.4 42.1 16.3 0.00
35.5 26.7 8.8 0.19
28.6 17.0 11.7 0.01
High Importance Low Importance Difference in Difference p-value (two-tailed)
Note: Unweighted analysis. Analysis excludes respondents who did not take a policy position. Policy Agreement is if the representative supports (opposes) and respondent strongly or somewhat supports (opposes) the policy. Policy Disagreement is if the representative supports (opposes) and respondent strongly or somewhat opposes (supports) the policy. High confidence is top two categories; low confidence is bottom two categories. High importance is top two categories; low importance is bottom two categories. See Appendix Table A1 for categories. Source: 2009 CCES.
Table 4. Effect of Party and Policy Agreement on Representative Evaluation by Variation in Confidence and Importance, OLS Regression (1) Teen Pregnancy
(2) (3) (4) Same Sex Banking Marriage Job Creation Regulation Evaluation of Representative (0-100) Position (+1=Supp.,-1=Opp.) 4.773 6.460 -0.113 1.910 [2.445]* [1.918]*** [5.526] [3.543] Confidence in pol. choice (0-1) -0.865 -0.815 1.376 2.789 [2.070] [2.072] [2.242] [2.599] Importance (0-1) 3.840 3.011 2.124 -0.561 [2.254]* [1.948] [3.242] [2.829] Rep's position (1=Supp, -1=Opp.) 1.572 -2.901 -2.473 0.288 [1.506] [1.240]** [2.986] [1.911] Rep's party (1=R, -1=D) -1.449 1.339 -0.434 2.787 [1.558] [1.175] [2.994] [1.824] Policy agreement (-1=Dis. to 1=Agg.) 18.510 17.020 12.001 8.484 [2.350]*** [1.748]*** [5.405]** [3.483]** Party agreement (-1 to 1) 3.422 4.684 0.282 -1.959 [2.265] [1.742]*** [4.650] [2.609] Conf. X PID (-1 to 1) 0.334 4.369 4.012 3.134 [3.045] [3.200] [2.939] [2.734] Conf. X Position (-1 to 1) -1.094 0.487 2.620 -1.609 [3.155] [2.753] [3.480] [3.836] Conf. X Rep's position (0-1) 0.744 2.509 -4.917 -2.813 [1.974] [1.971] [2.103]** [2.503] Conf. X Rep's party (-1 to 1) 0.456 2.348 1.322 0.820 [1.975] [1.915] [1.758] [1.898] Conf. X Policy Agreement (-1=Dis. to 1=Agg.) 5.693 11.928 5.243 1.193 [2.883]** [2.364]*** [3.047]* [3.644] Conf. X Party agreement (-1 to 1) 0.037 -1.247 3.942 -0.618 [2.788] [2.749] [2.517] [2.535] Impt. X PID (-1 to 1) 0.465 -0.217 -2.908 -3.686 [3.571] [3.343] [4.900] [3.562] Impt. X Position (-1 to 1) 1.152 -3.923 2.798 4.392 [3.478] [2.834] [5.672] [4.404] Impt. X Rep's position (0-1) -3.789 -0.242 7.976 5.575 [2.171]* [1.853] [3.177]** [2.746]** Impt. X Rep's party (-1 to 1) 1.847 -3.979 0.808 -2.256 [2.201] [1.763]** [3.198] [2.344] Impt. X Policy Agreement (-1=Dis. to 1=Agg.) -2.511 7.619 7.159 6.770 [3.165] [2.093]*** [5.460] [4.233] Impt. X Party agreement (-1 to 1) 1.266 -0.133 2.603 11.480 [3.234] [2.421] [4.940] [3.298]*** Age (in years) 0.068 0.146 -0.152 -0.133 [0.228] [0.239] [0.204] [0.215] Age-squared/100 -0.123 -0.147 0.058 0.140 [0.235] [0.243] [0.212] [0.224] Black -0.726 2.242 -0.964 -2.017 [2.038] [2.339] [1.954] [2.049] Hispanic -2.663 -1.181 -0.948 -2.052 [2.271] [2.189] [2.309] [2.072] Other Race 1.611 1.149 1.476 0.123 [2.350] [2.261] [2.685] [2.245] Female=1 -1.808 -1.233 2.051 -0.081 [1.216] [1.245] [1.125]* [1.205] 5 Pt Party ID (-1=Dem; 1=Rep) -0.187 -3.047 2.314 1.263 [2.365] [1.997] [4.566] [2.728] Education (0 no HS; 1 postgrad) 3.311 -2.691 3.031 2.660 [2.175] [2.011] [2.072] [2.083] Income (0=<$10k; .93=>150k; 1=RF) -3.439 4.591 -1.184 -5.242 [2.717] [2.680]* [2.431] [2.557]** Income Missing/RF -3.199 -2.757 -2.299 0.056 [2.571] [2.642] [2.359] [2.676] Constant 47.947 43.396 50.381 51.382 [5.582]*** [5.839]*** [5.271]*** [5.139]*** Observations 1500 1492 1487 1482 R-squared 0.320 0.580 0.440 0.320 Note: Unweighted analysis. OLS Coefficients with robust standard errors in brackets. * significant at 10%; ** significant at 5%; *** significant at 1%, two-tailed tests. Source: 2009 CCES.
Table 5. Robustiness of Effect of Party and Policy Agreement on Representative Evaluation by Variation in Confidence and Importance, OLS Regression (R1)
(R2)
(R3)
(R4)
Teen Pregnancy
Additional Policy Agreement Interactions Position (+1=Supp.,-1=Opp.) Confidence in pol. choice (0-1) Importance (0-1) Rep's position (1=Supp, -1=Opp.) Rep's party (1=R, -1=D) Policy agreement (-1=Dis. to 1=Agg.) Party agreement (-1 to 1) Conf. X PID (-1 to 1) Conf. X Position (-1 to 1) Conf. X Rep's position (0-1) Conf. X Rep's party (-1 to 1) Conf. X Policy Agreement (-1=Dis. to 1=Agg.) Conf. X Party agreement (-1 to 1) Impt. X PID (-1 to 1) Impt. X Position (-1 to 1) Impt. X Rep's position (0-1) Impt. X Rep's party (-1 to 1) Impt. X Policy Agreement (-1=Dis. to 1=Agg.) Impt. X Party agreement (-1 to 1) Constant
Among those Among those who believe holding strong policy opinions policy is very important (Position = 1 or (Importance=1) 1) 6.091 [3.374]* -0.364 [3.457] 5.452 [3.780] 6.869 [2.693]** -1.101 [2.741] 20.036 [2.698]*** 6.595 [3.683]* -5.046 [5.399] -2.603 [4.032] -1.081 [3.293] -1.723 [3.290] 5.295 [3.312] 0.814 [4.476] -1.621 [6.279] -1.354 [4.630] -8.225 [3.445]** 3.752 [3.516] -3.884 [3.554] -0.650 [4.797] 45.356 [9.672]***
Includes Age, Race, Gender, Education and Income? Yes Yes Includes Age, Race, Gender, Education and Income X Policy Agreement Interactions? Yes No Observations 1500 664 R-squared 0.330 0.410 Note: Unweighted analysis. OLS Coefficients with robust standard errors in brackets. * significant at 10%; ** significant at 5%; *** significant at 1%, two-tailed tests. Source: 2009 CCES.
Table 6. Within person across policy areas, effect of Policy Agreement on Representative Evaluation by Variation in Confidence, OLS Regression (1)
Confidence relative to individual's average confidence Importance relative to individual's average confidence Rep's position (1=Supp, -1=Opp.) Rep's party (1=R, -1=D) Policy agreement (-1=Dis. to 1=Agg.) Party agreement (-1 to 1) Conf. X PID (-1 to 1) Conf. X Position (-1 to 1) Conf. X Rep's position (0-1) Conf. X Rep's party (-1 to 1) Conf. X Policy Agreement (-1=Dis. to 1=Agg.) Conf. X Party agreement (-1 to 1) Impt. X PID (-1 to 1) Impt. X Position (-1 to 1) Impt. X Rep's position (0-1) Impt. X Rep's party (-1 to 1) Impt. X Policy Agreement (-1=Dis. to 1=Agg.) Impt. X Party agreement (-1 to 1) Constant Includes Policy Area indicators? Includes Individual-level fixed effects? Includes Age, Race, Gender, Education and Income? Includes Age, Race, Gender, Education and Income x Policy Area Interactions?
(4)
(5)
(6)
(7)
No
Yes
Yes
Yes
No No Yes Yes Includes Individual's Position, Confidence, and Yes (Just policy Importance, Rep.'s position and party, and Policy agreement x and Party Agreement x Policy Area Interactions? Policy Area) No No Yes Observations 752 5624 5624 5624 R-squared 0.480 0.410 0.410 0.430 Note: Unweighted analysis. OLS Coefficients with robust standard errors clustered at the individual level in brackets. * significant at 10%; ** significant at 5%; *** significant at 1%, two-tailed tests. Source: 2009 CCES.
Included for each policy area seperately
Position (+1=Supp.,-1=Opp.)
(3)
Included for each policy area seperately
Policy agreement (-1=Dis. to 1=Agg.) x Most Confident Policy Area Most Confident Policy Area
Table 7. Teen Pregnancy and Same Sex Marriage Opinions for Entire Sample and by Policy Confidence and Importance Teen Pregnancy Opinions Same Sex Marriage Opinions Mean Opinion (+1=Support, Proportion Extreme (1 or -1) Mean Opinion (+1=Support, Proportion Extreme (1 or -1) 1=Oppose) Opinions 1=Oppose) Opinions Entire sample 0.098 0.440 -0.202 0.700 High Confidence 0.254 0.620 -0.086 0.910 High Importance 0.155 0.440 0.042 0.850 High Importance and High Confidence 0.311 0.600 0.108 0.960 Note: Unweighted analysis. High confidence is top two categories; High Importance is top two categories. See Appendix Table A1 for categories. Source: 2009 CCES.
Figure 1: Effect of Policy Agreement on Evaluation of Representative by Confidence Levels 60
Difference in Evaluation of Representative, Policy Agreement - Policy Disagreement
50
40
30 Lowest Confidence (0) Highest Confidence (1)
20
10
0 Teen Pregnancy
Same Sex Marriage
Job Creation
Policy Area
Note: Results based on marginal effects for regression models shown in Table 4. Source: 2009 CCES.
Banking Regulation
Table A1. Summary Statistics for Model Variables and Coding Rules Variable Support for Rep. (0-100)
Teen Pregnancy 48.130
Same Sex Marriage
Job Creation
47.615
50.500
49.468
[26.7831]
[34.2301]
[27.6438]
[25.8746]
0.107
-0.202
0.410
0.424
[.7223]
[.8399]
[.6081]
[.5664]
0.506
0.540
0.548
0.419
[.3125]
[.3538]
[.3243]
[.327]
Position (+1=Supp.,-1=Opp.) Confidence in pol. choice (0-1) Importance (0-1) Age (in years) Age-squared/100 Black=1 Hispanic=1 Other Race=1 Female=1 5 Pt Party ID (-1=Dem; 1=Rep) Education (0 no HS; 1 postgrad) Income (0=<$10k; .93=>150k; 1=RF) Income Missing/RF Observations
Banking Regulation
0.621
0.449
0.920
0.765
[.2978]
[.3828]
[.1871]
[.2717]
48.867
48.867
48.827
48.744
[15.4774]
[15.4455]
[15.4381]
[15.4091]
26.273
26.264
26.222
26.132
[15.1459]
[15.1091]
[15.0799]
[15.0514]
0.110
0.109
0.110
0.110
[.313]
[.3112]
[.3134]
[.313]
0.100
0.098
0.098
0.100
[.3001]
[.2972]
[.2968]
[.2999] 0.056
0.057
0.056
0.057
[.2313]
[.2306]
[.2309]
[.23]
0.517
0.519
0.517
0.520
[.4999]
[.4998]
[.4999]
[.4998]
-0.062
-0.061
-0.058
-0.066
[.7209]
[.7183]
[.7206]
[.721]
0.455
0.454
0.457
0.457
[.2883]
[.287]
[.288]
[.2884]
0.535
0.532
0.534
0.534
[.2762]
[.2766]
[.2766]
[.275] 0.086
0.087
0.086
0.087
[.2824]
[.2801]
[.2816]
[.28]
1500
1492
1487
1482
Note: Means with standard deviations in brackets. Summary statistics are reported by policy area for Table 4 sample. Coding rules of selected variables Position (See Table 1 for question wording) -1=Strongly oppose, -.5=Somewhat oppose, 0=Neither support or oppose, .5=Somewhat support, 1=Strongly support Importance (See Table 1 for question wording) 0=Not at all important, .33=Of little importance, .66=Somewhat important, 1=One of the most important issues Confidence (Two items, average of two items, see Table 1 for question wording) Both items scored: 0=Not at all confident, .33=A little confident, .66=Somewhat confident, 1=Very confident Education 0=No high school degree, .2=High school degree, .4=Some college, .6=2 year college, .8=4 year college, 1=Post-graduate degree Income 0.00=less than $10,000, 0.07=$10,000 - $14,999, 0.14=$15,000 - $19,999, 0.21=$20,000 - $24,999, 0.29=$25,000 - $29,999, 0.36=$30,000 - $39,999, 0.43=$40,000 - $49,999, 0.50=$50,000 - $59,999, 0.57=$60,000 - $69,999, 0.64=$70,000 - $79,999, 0.71=$80,000 - $99,999, 0.79=$100,000 - $119,999, 0.86=$120,000 - $149,999, 0.93=$150,000 or more, 1.00=prefer not to say Source: 2009 CCES.
Table A2. Effect of Party Agreement on Representative Evaluation by Policy Confidence and Importance Difference in Mean Evaluation, Party Agreement - Party Disagreement High Confidence Low Confidence Difference in Difference p-value (two-tailed)
Teen Pregnancy 3.6 6.4 -2.8 0.47
Same Sex Marriage 6.7 2.3 4.5 0.35
Job Creation 8.5 7.9 0.6 0.88
Banking Regulation 10.8 9.0 1.8 0.63
High Importance 5.5 6.4 6.7 11.0 Low Importance 4.0 4.3 -0.3 -2.6 Difference in Difference 1.5 2.1 6.9 13.6 p-value (two-tailed) 0.70 0.62 0.51 0.00 Note: Unweighted analysis. Party Agreement is if the representative and respondent share partisanship, where respondent is strong or weak partisan. Independents and leaners excluded. Party Disagreement is if the representative and respondent do not share partisanship, where respondent is strong or weak partisan. Independents and leaners excluded. High confidence is top two categories; low confidence is bottom two categories. High importance is top two categories; low importance is bottom two categories. See Appendix Table A1 for categories. Source: 2009 CCES.
Table A3. Effect of Party and Policy Agreement on Representative Evaluation, OLS Regression (1) Teen Pregnancy
(2) (3) Same Sex Marriage Job Creation Evaluation of Representative (0-100) Position (+1=Supp.,-1=Opp.) 5.279 4.527 3.937 [0.982]*** [0.987]*** [1.118]*** Rep's position (1=Supp, -1=Opp.) -0.313 0.011 1.800 [0.584] [0.599] [0.685]*** Rep's party (1=R, -1=D) -0.090 0.930 1.060 [0.583] [0.607] [0.553]* Policy agreement (-1=Dis. to 1=Agg.) 19.791 28.614 22.321 [0.894]*** [0.741]*** [1.007]*** Party agreement (-1 to 1) 4.148 4.292 4.822 [0.828]*** [0.867]*** [0.792]*** Age (in years) 0.025 0.071 -0.124 [0.228] [0.243] [0.205] Age-squared/100 -0.064 -0.087 0.028 [0.234] [0.247] [0.213] Black -0.112 2.481 -0.714 [2.045] [2.396] [1.922] Hispanic -2.352 -1.120 -0.747 [2.228] [2.259] [2.292] Other Race 1.563 1.647 1.892 [2.338] [2.302] [2.675] Female=1 -1.345 -1.495 1.307 [1.205] [1.210] [1.105] 5 Pt Party ID (-1=Dem; 1=Rep) 0.441 -0.310 1.711 [0.909] [1.124] [0.886]* Education (0 no HS; 1 postgrad) 2.946 -0.982 3.699 [2.156] [2.056] [2.072]* Income (0=<$10k; .93=>150k; 1=RF) -3.861 4.428 -0.773 [2.673] [2.758] [2.460] Income Missing/RF -2.766 -2.587 -2.406 [2.570] [2.670] [2.349] Constant 50.510 45.520 52.336 [5.345]*** [5.738]*** [4.593]*** Observations 1500 1492 1487 R-squared 0.310 0.550 0.430 Note: Unweighted analysis. OLS Coefficients with robust standard errors in brackets. * significant at 10%; ** significant at 5%; *** significant at 1%, two-tailed tests. Source: 2009 CCES.
Table A4. Predicting Policy Confidence, OLS Regression (1) (2) Teen Teen Pregnancy Pregnancy Baseline Model With Ideology Age (in years) Female=1 Education (0 no HS; 1 postgrad) Income (0=<$10k; .93=>150k; 1=RF) Income Missing/RF Black Hispanic Other Race Very Liberal Liberal Conservative Very Conservative Ideology=Not Sure Strong Dem
Policy Confidence and Electoral Punishment: A New ...
Department of Political Science ...... of the specification presented in column (1) is that we discard a great deal of data ..... Chicago: University of Chicago Press.
Nov 13, 2011 - proportional to his bargaining power or vote share. ..... distinct platforms in such a way that there's a room for bargaining (i.e. pA < pB so that.
proaches. It is either assumed that the central bank sets the interest rate according ... rate. These facts will serve as a benchmark with which we want to compare.
rameters; the degree of forward-looking behavior in the determination of infla- tion and output; and the variances of inflation and output shocksâto match.
... Classmates com for Free Reconnect with high school friends browse the biggest database of online yearbooks and or plan high school reunions The latest ...
Mar 24, 2015 - Our model identifies two opposing forces that drive variation in turnout in response to an increase ..... and working back to the media outlets' choice of reporting strategies. ..... Henceforth, we focus on the interior solution (Ïi(Î
Jun 17, 2016 - are both office and policy motivated, to campaign by announcing a ...... the lemma says that against any strategy for candidate âi, any best ...
Boise State University, USA. Abstract. ..... cal significance, this model also indicates that how a country's government is structured is .... Available online at: www.
... Andrew Reynolds, David Samuels,. Richard Snyder, Richard Soudriette, R. Kent Weaver. This publication consists of detailed information of , as well as other about . Our solutions was launched with a wish to work as. a complete on the internet ele
Oct 8, 2005 - porting free trade, which is efficient but whose benefits are not easily targetable ...... Entry barriers can also be influenced by campaign finance laws and ease of access to the media as well as by some intrinsic political forces that
Jun 21, 2018 - In our study, peers also play an important role since they determine ..... time, in federal prisons because they tend to commit more serious ...
Dec 15, 2011 - State-specific effects thus relax the assumption that two states with similar number ... Specifically, Donohue and Wolfers relax the assumptions.
limit how far to the right R can push policy. R's voter-optimal platform is thus an interval from αR(1), the lowest policy R would ever take, up to a ceiling. Calculus determines the optimal ceiling to be a0, defined in Equation 4. The reasoning is
Altruistic punishment is a behaviour in which individuals punish others at a cost to themselves in order to provide a public good. Fehr and Gächter1 present.
Other typologies exist (Martin 1997; Reynolds and Reilly 1997). We do not ...... states of Georgia and Louisiana, which elect a total of 18 members (out of 435). 7.
![[PDF BOOK] A New Model: What Confidence, Beauty ...](https://p.pdfkul.com/img/300x300/pdf-book-a-new-model-what-confidence-beauty-_5a2628cc1723dd567af5403b.jpg)
