Emotion and Political Judgment

Affective Intelligence Theory (AIT)

The hypothesis that political judgment depends on citizens’ informational incentives can be traced to Downs (1957). More recently, Marcus and colleagues have provided both a powerful empirical demonstration of the primacy of motivation as well as a rigorous intellectual framework to understand it (MacKuen et al. 2010; Marcus 1988; Marcus and MacKuen 1993; Marcus, Neuman, and MacKuen 2000). Relying on insights from neuroscience (e.g., Damasio 1994; Gray 1981; LeDoux 1996), they argue that the nature of political reasoning lies in the experience of emotion. Specifically, Marcus and his colleagues developed a theory of affective intelligence, in which the key emotions of enthusiasm, anger, and anxiety—viewed as the products of functionally distinct bio-behavioral regulatory systems—determine the strategies citizens use to construct their political judgments. This view of emotion contrasts with popular and classic notions that sharply divide “reason” from “passion” (see Marcus 2002 for an extended discussion). In contrast, AIT posits that the experience of emotion is integral to the nature and quality of human judgment (see also Damasio 1994).

AIT posits that each of these three emotions corresponds with a distinct appraisal of the strategic character of a given environment. Enthusiasm is the result of appraisals that the current environment is “safe and rewarding,” and that familiar habits or modes of thought are likely to be useful. Anger (aversion) is a signal that the current environment is threatening but familiar, and that previously learned habits and solutions will be successful in dealing with the threat. Finally, anxiety results when the current environment is appraised as uncertain and thus potentially threatening. The novelty of this latter case suggests that habits, heuristics, and familiar solutions are unlikely to be useful. Anxiety thus promotes increased thoughtfulness and attention, and greater motivation for learning.

Applied to the realm of mass political judgment, both enthusiasm and anger are expected to motivate political participation and to increase voters’ reliance on their political habits (e.g., party identification). Anxiety, by contrast, is expected to lead voters to turn away from habitual responses and devote more attention to contemporary information in the environment. In line with AIT, Marcus, Neuman, and MacKuen (2000) show that anxious voters (1) pay greater attention to the political environment and acquire more information about candidates’ policy stands, (2) rely less on partisanship and more on policy preferences and assessments of candidate character in forming candidate evaluations, and (3) defect at higher rates in presidential elections (see also Marcus and MacKuen 1993). Brader (2006) provides powerful evidence of the role of anxiety by manipulating it experimentally in the context of political advertising. He finds that compared with an emotionally pallid ad, a threatening ad—which was intended to produce anxiety—led to greater attentiveness and more message-based persuasion. Finally, anxiety—though not anger or enthusiasm—has been found to increase both the quantity and quality of information processing, including the desire for balance and compromise (MacKuen et al. 2010; Valentino et al. 2008). Taken together, this work suggests that anxiety leads to political decisions that are more informed by contemporary information and less by reflexive partisan loyalty. As Marcus (2002: 116) forcefully states, “anxiety is the central emotion on which reason and democratic politics rest.”

Expectancy Violation Theory (EVT)

The affective intelligence model has gained wide acceptance in political science. However, reflective deliberation may arise through pathways other than anxiety. Opportunities for citizens to take a sober second look may derive from a broader set of dynamics. A large body of research on persuasion and social judgment in psychology indicates that people are motivated to attain a certain degree of confidence in the validity of their decisions, and that they will seek out (and deliberate about) information until actual confidence matches or exceeds a desired level (Chaiken, Liberman, and Eagly 1989; Groenendyk 2013; Lavine, Johnston, and Steenbergen 2012; Payne, Bettman, and Johnson 1993). The desire for confident knowledge reflects a well-established psychological preference for cognitive clarity and an aversion to ambiguity, uncertainty, and doubt (Kruglanski and Webster 1996; Moskowitz, Skurnik, and Galinsky 1999). At the same time, however, people are economy-minded and prefer to make their judgments without expending too much cognitive energy (for a review, see Kahneman 2011). This may be especially likely in domains such as politics, which for most people most of the time is not a compelling spectacle. The general goal, then, is to strike a balance between minimizing cognitive effort and maximizing decision confidence (Chaiken, Liberman, and Eagly 1989; Downs 1957).

In the context of mass politics (as in other judgment domains), this trade-off between efficiency and accuracy is thought to be resolved through reliance on information shortcuts (or heuristic cues; Downs 1957; Lau and Redlawsk 2001; Lupia 1994; Popkin 1994). In theory, these simplifying heuristics allow citizens to glean enough information from the environment to make reasonably informed decisions without immersing themselves in difficult-to-process (or uninteresting) details (for critical reviews, see Kuklinski and Quirk 2001; Lupia and McCubbins 1998; Mondak 1994). For example, as Lau and Redlawsk (2006, 232) write,

if the salient characteristics of a particular politician are consistent with the prototypic Republican, say, then voters may readily infer that she is for a strong defense, low taxes, against government intervention in the economy, against abortion, and so on; and will probably have a readily-available affective response (what Fiske and Pavelchak, 1986, call a schema-based affective response) to the party label.

In other words, party cues and other cognitive shortcuts often allow voters to go beyond the information given in making valid political inferences.

However, what if salient features of the object are not consistent with, or representative of, the prototype? For example, what if a candidate’s salient characteristics are disconfirming? In 1992, voters confronted a Democratic presidential candidate who favored the death penalty, free trade, and welfare reform, and who claimed “the era of big government is over.” Similarly, in 2000, George W. Bush’s stance as a “compassionate conservative” was out of step with perceptions of the Republican Party. When perceptions of the political world are not structured in a stereotypic manner, will citizens continue to rely on heuristic-based expectations (even though they may not apply), or might this prompt them to engage in deeper, more careful thinking? Psychological research indicates that when expectancies are disconfirmed, people engage in “sense-making activity aimed at explaining the discrepancy between what is and what was expected” (Roese and Sherman 2007, 91). Specifically, research indicates that disconfirming information is more salient, more memorable, more attention-grabbing, and instigates more processing time and more resource-demanding appraisal strategies than information that conforms to “top-down” cognitive expectations (e.g., Bargh and Thein 1985; Hastie and Kumar 1979; Jonas, Diehl, and Bromer 1997). Thus, when the “fit” between expectancies and contemporary information is weak, individuals may substitute deliberative for heuristic thinking to reach their judgment confidence thresholds.

There is evidence from several lines of research in political psychology—in addition to that pertaining to AIT—consistent with the expectancy violation perspective. First, Lavine and colleagues’ (Basinger and Lavine 2005; Lavine, Johnston, and Steenbergen 2012) research on partisan ambivalence indicates that the depth and objectivity of political information processing hinges on the degree to which short-term party evaluations disconfirm long-standing partisan identities. According to this work, ambivalence undermines the perceived reliability of partisan cues and motivates citizens to acquire more costly information to reach their judgments. Similarly, Groenendyk (2013) provides direct evidence that citizens engage in deliberative thought to come to terms with conflict between their own issue positions and those of their preferred party, and that this process can result in weakened partisan attachments under certain conditions. Related work by Hillygus and Shields (2008) indicates that partisan defection in presidential elections occurs when voters learn that their party’s candidate holds a disagreeable position on a personally important policy issue. This result can be understood as an instance of expectancy violation (“my party’s candidate should agree with me on issues I care about”), and Hillygus and Shields show that it creates openness to the possibility that the other party’s candidate may provide a better fit with the voter’s substantive preferences. Finally, in recent experimental work, Sniderman and Stiglitz (2012) find that voters rely strongly on party identification in forming their policy and candidate judgments, but only when party-identified candidates take stereotypical (i.e., expectancy confirming) policy positions (i.e., when the Democrat lines up to the left of the Republican in policy space).

Implications and Crucial Tests

Can a general EVT explain findings that have been attributed specifically to anxiety? To find out, we examine inconsistency between citizens’ partisan identities and the emotions engendered by the parties’ main representatives, that is, presidential candidates. According to AIT, the motivation to engage in reflection and reevaluation hinges specifically on activation of the surveillance system, that is, on the experience of anxiety. However, feeling anxious about the candidate of one’s own party (i.e., in-party anxiety) is only one route by which expectancy violation may occur. Such violations may also result from experiencing positive emotions toward the other party’s (i.e., the out-party) candidate. For example, in presiding over the economic boom of the late 1990s, Bill Clinton’s approval rating increased among Republicans (Lebo and Cassino 2007). In the context of AIT, enthusiasm toward the out-party’s candidate entails the activation of the disposition system rather than the surveillance system. Its influence in stimulating deliberative thought would thus seem difficult to square with AIT. That is, if the experience of enthusiasm toward the out-party candidate alters the bases of political judgments in the same way as anxiety toward the in-party candidate, then (1) AIT is in need of theoretical revision, and/or (2) AIT is incomplete and adaptive political behavior may arise from other common circumstances, independent of the experience of anxiety.

The predictions of AIT and EVT in the context of candidate evaluation are depicted in Figure 1. ³ According to AIT (see top panel), feelings of anxiety heighten the voter’s reliance on issue preferences and depress the voter’s reliance on party identification, whereas feelings of enthusiasm and anger motivate habit-based participation. ⁴ According to EVT, the political effects of anger, enthusiasm, and anxiety depend entirely on whether they constitute expectancy-violating or expectancy-confirming emotions, and thus crucially depend on the object with which the emotion is associated. Anxiety and anger experienced in relation to the candidate of one’s own party or, equally, enthusiasm for the candidate of the opposing party—hereafter “in-candidate anxiety and anger” and “out-candidate enthusiasm”—should undermine judgment confidence on the basis of partisan cues and thus prompt more issue-based evaluation, as both emotions (so directed) constitute violations of general partisan expectancies. By contrast, both enthusiasm toward the candidate of one’s own party and anxiety and anger toward the candidate of the other party—hereafter “in-candidate enthusiasm” and “out-candidate anxiety and anger”—reinforce partisan expectations, and should thus heighten citizens’ confidence in standing partisan decisions (making it unnecessary to scrutinize the candidates along the more cognitively difficult metric of issue agreement). These expectancy-confirming emotions should increase party-based voting and decrease issue-based voting.

OPEN IN VIEWER

Figure 1.

Graphical depictions of theoretical models.

To summarize, AIT posits that particular discrete emotions are associated with particular political judgment styles: anxiety increases reflection while anger and enthusiasm promote heuristic styles of judgment. In contrast, EVT posits that discrete emotions have different consequences depending on the object with which the emotion is associated. When positive emotions (enthusiasm) are associated with in-party objects and negative emotions (anxiety, anger) are associated with out-party objects, heuristic judgment is more likely. When negative emotions are associated with in-party objects, and positive emotions with out-party objects, deliberation is more likely. Thus, for example, feelings of anger toward the in-party candidate are a signal that the political environment is nonstereotypical, and deliberation is likely to be useful. EVT thus agrees with AIT that anxiety will sometimes promote deliberation and enthusiasm and anger will sometimes promote reflexive partisanship. Where EVT diverges from AIT is in the claim that each discrete emotion is necessarily related to one specific judgment style. Instead, it is the interaction between that emotion and the relationship of the eliciting object to the citizen that matters. In this sense, EVT is a more general theory because EVT can account for all of the findings from the survey data supporting AIT, while also providing an account of patterns that lie outside the predictions of AIT. We return to these issues in the conclusion.

Data and Variables

To test these competing hypotheses, we rely on the ANES cumulative file between 1980 (when the emotion terms debuted) to 2004 (we examine the 2008 election separately below). To maximize consistency with Marcus, Neuman, and MacKuen’s (2000) analysis, we measure presidential vote as respondents’ self-reported vote intention in the preelection study. This variable is a three-point ordinal measure (intend to vote Democratic = −1, undecided = 0, or Republican = 1). To measure emotional reactions toward the two major-party candidates, respondents in each election study were asked the following question for each of four emotion terms (afraid, angry, hopeful, and proud): “Has [candidate]—because of the kind of person he is, or because of something he has done—made you feel [emotion term]?” Consistent with Marcus, Neuman, and MacKuen, we combined the proud and hopeful items (separately for each candidate) as a measure of candidate-specific enthusiasm, used the angry item to measure anger, and used the afraid item as a measure of candidate-specific anxiety. The emotions items were then recoded to reflect whether their referent was the candidate of the in-party or the candidate of the out-party (e.g., enthusiasm toward George Bush in 2004 is coded as in-candidate enthusiasm for Republicans, and as out-candidate enthusiasm for Democrats). Partisan identity was measured with the traditional ANES seven-point item (including leaners), and coded so that higher values indicate stronger Republican identification (“true” Independents, approximately 10% of the sample across election years, are excluded from the analyses, and leaners are treated as partisans for the purposes of coding referents for the emotions items).

Following Marcus, Neuman, and MacKuen (2000), issue voting was measured via comparative candidate issue distance. Specifically, we calculated the absolute distance between the respondent’s position on a given issue and his or her perception of each candidate’s position. For each issue, we then subtracted the absolute distance from the Republican candidate from the absolute distance from the Democratic candidate, and averaged across issues (higher values thus indicate greater average proximity to the Republican candidate). The issues available for this analysis are defense spending, foreign policy orientation toward the USSR, women’s role in society, government aid to black Americans, government versus private health insurance, government spending and services, and government guaranteed jobs and income. We use the average for all available issues in a given year for each respondent. Prior to analysis, all variables were recoded to a zero to one range, except for comparative issue distance, which was recoded to have a mean of zero and a standard deviation of one. ⁶ The coefficients for this variable represent the expected change in the dependent variable for a one standard deviation change in comparative issue distance. Finally, we controlled for age, gender, race, education, and income.

Vote Intention

To examine the bases of variation in voting strategy, we regressed vote intention on partisanship, comparative issue distance, the six emotion variables (in- and out-candidate enthusiasm, anxiety and anger), the interactions of partisanship and comparative issue distance with all six emotion variables, demographic controls, and fixed effects for survey-year. ⁷ As the coefficients indicate, the results conform better to EVT than AIT: the expectancy-violating emotions (in-candidate anxiety and anger, and out-candidate enthusiasm) exert statistically significant and substantively strong moderating effects in decreasing the marginal impact of partisanship, whereas the expectancy-confirming emotions (out-candidate anxiety and anger, and in-candidate enthusiasm) exert statistically significant and substantively strong moderating effects in increasing its marginal effect. The coefficients for comparative issue distance generally support EVT over AIT: the expectancy-violating emotion of out-candidate enthusiasm strongly increases respondents’ electoral reliance on issues, whereas two of the three expectancy-confirming emotions decrease the impact of issues. Out-candidate anger—an expectancy-confirming emotion—is associated with increased use of issues, however, an effect that is contrary to the predictions of both AIT and EVT.

To get a sense of the magnitude of these effects, in Figure 2 we report changes in the predicted probabilities of voting Republican as a function of the emotion terms. The y-axis in the figure represents the expected change in the marginal effect on voting Republican of either partisanship (left plot) or comparative issue distance (right plot), as each emotion term moves from its minimum to maximum value, holding all other variables at their central tendencies (the thick vertical lines are 68% confidence intervals, and the thin vertical lines are 95% confidence intervals). In other words, the dots in these figures represent the change in the effect of partisanship and issues on voting in the presence of each emotion shown on the x-axis.

OPEN IN VIEWER

Figure 2.

Effects of partisanship and issue proximity, by emotions, 1980–2004 ANES.

As the left plot shows, the marginal effect of partisanship decreases 17 percentage points in the presence of the expectancy-violating emotion of in-candidate anxiety. The moderating effects of the other expectancy-violating emotions (i.e., in-candidate anger and out-candidate enthusiasm) are even larger: here, the marginal effect of partisanship decreases by 18 and 82 percentage points, respectively. In line with EVT, the expectancy-confirming emotions have the opposite set of effects: the influence of party identification is 6 percentage points larger among respondents reporting anxiety about the other party’s candidate, 7 points larger in the presence of out-candidate anger, and 41 points larger among those reporting enthusiasm toward their own party’s candidate.

The right plot of Figure 2 displays the expected changes in the marginal effect of comparative issue distance on the probability of voting Republican. Although the moderating impact of in-candidate anxiety (AIT’s signature variable) is not significant, the expectancy-violating emotion of out-candidate enthusiasm produces an increase of 49 percentage points in the marginal impact of comparative issue distance. This means that the expected difference in voting Republican as a function of comparative issue distance is 49 percentage points larger among those who experienced pride and hope about the out-party candidate, all else equal. Also in line with EVT, out-candidate anxiety decreases the marginal effect of comparative issue distance by 16 percentage points, whereas in-candidate enthusiasm decreases this effect by 9 percentage points. Finally, in conflict with both AIT and EVT, out-candidate anger significantly increased the use of issues by 10 percentage points.

In sum, the results of our first analysis are more consistent with an expectancy violation framework than with AIT. Deliberative thinking—reflected in judging presidential candidates in terms of issue proximity rather than partisanship—is not limited to the experience of anxiety. In fact, experiencing enthusiasm toward the other party’s candidate is a much stronger predictor of whether respondents engage in policy voting. Furthermore, the experience of anxiety toward the other party’s candidate decreases policy voting, which is in line with EVT but not predicted by current versions of AIT. Thus, it is not the experience of anxiety per se that produces variation in judgment strategies, but whether an emotion confirms or violates one’s partisan expectations. This is further supported by the moderating role of anger on the use of partisanship, which is positive when experienced toward the out-party candidate and negative when experienced toward the in-party candidate. That is, anger can promote either reflexive or reflective thinking depending on what object the anger is directed toward.

When the contemporary political environment (as represented by the major-party candidates) violates partisan expectations, citizens eschew partisanship as an easy-to-use judgment cue, and devote more thought to which candidate’s policy preferences dovetail with their own. Conversely, when the contemporary environment conforms to partisan expectations (e.g., when the in-party candidate engenders positive emotions and the out-party candidate engenders negative emotions), citizens are less likely to engage in cognitive costly (i.e., spatial) forms of voting. Instead, they are content to make their decisions largely on the basis of partisan loyalty. This result, using discrete emotions as the primary moderators, is very similar to that of other recent work that examines other forms of partisan expectancy violations, such as salient mismatches on important issues (Hillygus and Shields 2008), Democrats and Republicans lining up on the “wrong sides” of a unidimensional ideological continuum (Sniderman and Stiglitz 2012), or experimental manipulations that force respondents to think about the things they dislike about their own party (Lavine, Johnston, and Steenbergen 2012). An expectancy violation perspective captures all of these findings within a parsimonious framework.

Time of Vote Decision

By its very nature, reflective deliberation entails a consideration of one’s political options beyond what is implied by partisan habits. In particular, it implies a willingness to be influenced or “enlightened” by the campaign (e.g., Gelman and King 1993). Conversely, reflexive partisanship entails early decision foreclosure, perhaps even before the party’s nominee is known (Lavine, Johnston, and Steenbergen 2012). Why do some voters remain open into the fall campaign, whereas others “knew all along” that they would vote for one candidate (or party) or the other? According to AIT, anxiety should lead voters to reflect on and reevaluate their options, leading to later time-to-decision, whereas enthusiasm should lead to early decision foreclosure. According to EVT, decision foreclosure should be inhibited by the experience of expectancy-violating emotions and facilitated by the experience of expectancy-confirming emotions.

To empirically contrast the two perspectives (in presidential elections between 1980 and 2004), we examine an ANES item that asks voters when they came to their voting decisions. The item (administered in the postelection surveys) reads, “How long before the election did you decide that you were going to vote the way you did?” Six response options are provided, ranging from “knew all along,” “when candidate announced,” “during conventions,” “postconvention period,” “last two weeks of campaign,” and “on Election Day.” We regressed this variable on the six emotion variables, political interest, partisan strength, and the demographic controls. The results of this analysis, estimated via multinomial logit, are shown in Figure 3, and support EVT. ⁸ Specifically, all three of the expectancy-violating emotions (in-candidate anxiety and anger, and out-candidate enthusiasm) significantly increase the time until decision, and all three of the expectancy-confirming emotions (out-candidate anxiety and anger, and in-candidate enthusiasm) significantly decrease the time to decision. Each point in Figure 3 represents the predicted probability that the respondent made up his or her mind sometime in the postconvention period (i.e., during the campaign until Election Day). Once again, the largest emotional effect is for out-candidate enthusiasm, which increases the probability of deciding during the heart of the campaign by 12 percentage points. The effects of in-candidate anxiety and anger are also positive, as expected by EVT, though somewhat smaller (7 and 3 points, respectively). Finally, all three expectancy-confirming emotions—out-candidate anxiety and anger, and in-candidate enthusiasm—decrease the probability of deciding during the campaign season (by 2, 4, and 7 points, respectively).

OPEN IN VIEWER

Figure 3.

Timing of vote decision, 1980–2004 ANES.

These findings strongly converge with those above. Specifically, expectancy-violating emotions interrupt habit-based judgment, and thus open citizens’ minds to other options, reflected in both the greater use of issues in voting behavior and the correspondingly longer time to decision that such deliberation entails. Expectancy-confirming emotions, by contrast, entail heavy reliance on partisan habit and early decision foreclosure. In sum, it is not anxiety, anger, or enthusiasm per se that matters, but rather the object to which such emotions are attached, and the relationship of that object to citizens’ political predispositions.

Defection

If citizens are open to persuasion during campaigns and willing to judge the candidates on the basis of policy agreement, the probability of partisan defection should increase. Defection constitutes a simple way to summarize the overall influence of emotions on voting behavior. AIT suggests that anxiety will increase defection rates, all else equal, whereas anger and enthusiasm will decrease defection rates. EVT suggests that defection will increase as a function of expectancy-violating emotions and decrease as a function of expectancy-confirming emotions. Figure 4 plots the changes in the probability of defection as a function of the experience of each of the six emotions. ⁹ Once again we confirm the basic EVT pattern: expectancy-confirming emotions significantly decrease defection rates, whereas expectancy-violating emotions increase defection rates. Anxiety, anger, and enthusiasm all influence defection from partisan habits, but they do so in exactly opposite ways conditional on the object with which the emotion is associated (i.e., the candidate of their own party or the other party’s candidate).

OPEN IN VIEWER

Figure 4.

Effects of emotions on defection rates, 1980–2004 ANES.

Data and Variables

Vote intention was not available in the 2008 ANES panel study; we therefore used a measure of comparative candidate evaluation, assessed in both the June and October waves of the study. The questions read, “Do you like [Barack Obama/John McCain], dislike him, or neither like nor dislike him?” Responses ranged from one (“dislike a great deal”) to seven (“like a great deal”), with values of four corresponding to “neither like nor dislike.” We calculated the difference between respondents’ attitudes toward McCain and Obama, with higher values indicating greater relative support for McCain. Self-reported emotional responses to the candidates were asked in the February wave of the study. The questions read, “When you think about (Barack Obama/John McCain), how [emotion term] does he make you feel?” Responses ranged from one (“not at all”) to five (“extremely”). In these data, in-candidate anxiety and anger and out-candidate anxiety and anger were correlated at.74 and .75, respectively. The inclusion of both emotions in the same model would thus produce high levels of multicollinearity; essentially, these items are measuring the same construct (see also Lodge and Taber 2013). We thus focus on anxiety only in the 2008 panel, as this is the primary moderating variable in AIT. Analyses examining anger (excluding anxiety) produced very similar estimates and substantive conclusions, which makes sense given their high correlation. ¹⁰

Partisanship was measured using the traditional 7-point branching format as above and was administered in the January wave of the study. Respondents’ issue preferences were also assessed in January, and their perceptions of the candidates’ preferences were assessed in June. As we examine the change in candidate evaluations from June to October, we restrict our analysis to those issues that were salient during the campaign (e.g., those that received attention in the media and in the candidates’ speeches and advertisements). By focusing on prominent issues (i.e., those likely to change during the election), we seek to temper the probability of a Type II error while maintaining the causal leverage of the panel design (see Lenz 2012). The ANES included two issues that were salient during the 2008 campaign and on which the candidates took distinctly different positions: raising taxes on those making more than US$200,000 per year and government-provided health insurance. The two items read as follows: “Do you favor, oppose, or neither favor nor oppose raising federal income taxes for people who make more than US$200,000 per year?” and “Do you favor, oppose, or neither favor nor oppose the U.S. government paying for all necessary medical care for all Americans?” Comparative issue distance was calculated in the same manner as in the cross-sectional study, with higher values corresponding to greater proximity to McCain.

As in the previous analysis, we control for age, gender, race, education, and income. All variables were recoded to range from zero to one, except comparative issue distance, which (as before) was coded to have a mean of zero and a standard deviation of one.

Party- and Issue-Based Change in Candidate Evaluation

To examine how patterns of emotional response altered the bases of change in candidate evaluation from June to October, we regressed comparative candidate evaluations in October on lagged evaluations in June, partisanship in January, comparative issue distance (with policy preferences measured in January and perceptions of the candidates’ stands measured in June), and all controls. We also included the four emotion items and their interactions with both partisanship and comparative issue distance.

The results of this analysis are shown in Figure 5. Figure 5 can be interpreted identically to Figure 2, except that the estimates refer to the expected changes in relative candidate evaluation across the range of the relevant independent variable. With respect to party identification, for example, the estimates can be interpreted as the additional expected gap in candidate evaluations in October between Republicans and Democrats above and beyond initial differences observed in June. As expected, the panel results are weaker than those generated with cross-sectional data; nevertheless, the substantive pattern of the interactions favors EVT.

OPEN IN VIEWER

Figure 5.

Effects of partisanship and issue proximity, by emotions, 2008 ANES panel.

In particular, as Figure 5 shows, both expectancy-confirming emotions (out-candidate anxiety and in-candidate enthusiasm) heighten the role of partisanship on changes in candidate judgments, and one of the two expectancy-violating emotions (out-candidate enthusiasm) sharply reduces it. Emotion-induced changes in the impact of issue proximity are also more consistent with EVT than AIT. In particular, the moderating impact of in-candidate anxiety does not approach statistical significance. By contrast, out-candidate enthusiasm sharply increases the impact of policy on changes in candidate evaluation. All else equal, the marginal effect of comparative issue distance on campaign-related change in candidate evaluation among Democrats who express enthusiasm for McCain and Republicans who express enthusiasm toward Obama is larger than for citizens who do not experience this expectancy-violating emotion. Again, for both party- and issue-based candidate evaluation, it is not the discrete emotion of anxiety that produces changes, but whether an emotion—either anxiety or enthusiasm—confirms or violates one’s partisan expectations.

Party Cue-Taking and Policy Persuasion

Thus far, our cross-sectional and panel analyses indicate that issue voting is abetted by the experience of expectancy-violating emotions (especially those who experience enthusiasm toward the out-party candidate). In the panel analysis, which protects against reverse causation, the coefficient capturing the relationship between comparative issue distance and October candidate preferences (controlling for preferences in June) indicates that respondents who experience expectancy-violating emotions relied less on partisan habits and more on their policy preferences in updating their evaluations of the candidates. In our final analysis, we leverage the 2008 panel study to examine whether those who experience expectancy-confirming emotions engage in the reverse process: whether they rely on partisanship to update their policy preferences. Lenz (2009, 2012) finds that campaigns often lead voters to adopt the preferred party’s (or candidate’s) positions as their own. That is, rather than engaging in issue voting, voters are persuaded to adopt party-approved positions or those of their preferred candidates (often after simply learning those positions).

The question we raise here is whether this process of partisan-based policy persuasion is limited to voters holding expectancy-confirming emotions. If such emotions bolster confidence in the reliability of partisan cues as a sufficient basis for candidate choice, citizens may be especially likely to take on party- (or candidate-) approved positions as their own (see Lavine, Johnston, and Steenbergen 2012). If so, this suggests that the process responsible for producing a cross-sectional relationship between policy distance and candidate evaluation is heterogeneous. Among those who have some doubt about the reliability of partisan cues (i.e., those experiencing expectancy-violating emotions), the relationship between issue distance and candidate evaluation likely reflects the influence of the former on the latter. However, among those who derive sufficient electoral confidence on the basis of party identification (i.e., those who experience expectancy-confirming emotions), the relationship between issue distance and candidate evaluation may reflect the effect of the latter on the former.

In the 2008 ANES panel study, respondents’ positions on tax increases and health care reform were measured in both January and October. This allows for an examination of whether partisanship (measured in January) led to changes in policy preference on these two salient campaign issues from January to October. We regressed preferences over each issue (separately) in October on lagged preferences from January, partisanship measured in January, and the interaction of partisanship with the four emotion items. Issue preferences were coded so that higher values indicated opposition to both tax hikes and universal government health care.

The results are in line with EVT. Both expectancy-confirming emotions—in-candidate enthusiasm and out-candidate anxiety—significantly increased the impact of partisanship on the change in tax policy preference (from January to October; left panel of Figure 6), and in-candidate enthusiasm had the same impact on changes in health insurance preferences (right panel of Figure 6). In each case, the magnitude of the effect was about 20 percentage points. The results for the expectancy-violating emotions were in the (expected) opposite direction—indicating a decrease in the marginal effect of partisanship on changes in tax and health insurance preferences—but inefficiently estimated. However, citizens who experienced both expectancy-violating emotions show a statistically significant and substantively large decline in the effect of partisanship (B = −.29, p < .10, for taxes; B = −.46, p < .01, for health insurance).

OPEN IN VIEWER

Figure 6.

Party cue-taking and changes in issue preferences, 2008 ANES panel.

The findings from the panel study provide additional support for EVT. Although the stringent nature of these tests produces noisier findings than the cross-sectional data, we can place greater confidence in their causal direction. Taken as a whole, we find a pattern of reflexive and reflective citizenship more consistent with EVT than AIT. As we have noted, the political judgment strategies of voters seem to hinge less on the experience of any particular emotion (e.g., anxiety), and more on whether the overall pattern of emotional response to the candidates confirms or disconfirms party-based expectations. When expectations are confirmed, political judgments are driven mostly by party identification; when expectations are violated, voters engage in more deliberative thinking by considering the question of issue agreement.

Summary of Findings

AIT and EVT make distinct predictions about when and why citizens will act like reflexive partisans or reflective deliberators in making political judgments about candidates and issues. AIT posits that the switch to reflective judgment hinges specifically on the experience of anxiety. According to Marcus and MacKuen (1993: 677),

. . . anxiety and enthusiasm play importantly different roles in the voting decision. In particular, the data indicate that enthusiasm directly affects voting preference (reflecting something very close to the voting decision itself), while anxiety has practically no direct impact on choice. Equally important for our point of view, anxiety appears to give voters pause—to get voters to base their decision on candidate characteristics or campaign information rather than merely stick with their “standing choice.

Thus, according to AIT, anxiety affects political judgment in a functionally distinct manner from enthusiasm and anger. EVT, by contrast, posits that judgment style depends on whether an emotional response represents a confirmation or a violation of one’s partisan expectancies. The former signals that standing habits may provide for sufficient judgmental confidence, whereas the latter signals that additional cognitive effort may be required (see Lavine, Johnston, and Steenbergen 2012). Both AIT and EVT hold that the experience of anxiety toward the in-party candidate may produce a shift from reflexive to reflective judgment. The two perspectives may also agree that partisan habits provide a sufficient basis for political judgment when enthusiasm toward the in-party candidate is high. In other cases, however, the two perspectives diverge. According to EVT, anxiety toward the in-party candidate is simply one route through which effortful political judgment might arise; any information (whether emotional or otherwise) that disconfirms partisan expectations should produce the same form of changes as in-candidate anxiety.

With respect to “crucial tests”—major points of divergence between EVT and AIT—we find that (1) enthusiasm toward the out-party’s candidate—an expectancy-violating emotion—functioned in a way that AIT reserves exclusively for anxiety (e.g., increasing issue voting and decreasing party voting); (2) anxiety toward the out-party candidate—an expectancy-confirming emotion—functioned in a way that is similar to enthusiasm toward the in-party candidate (e.g., increasing party voting and decreasing issue voting); and (3) anger toward the in-party candidate—an expectancy-violating emotion—functioned in a way that is similar to anxiety toward the in-party candidate. None of these three findings are predicted by AIT, but all fit comfortably within the EVT paradigm.

The reality appears to be that enthusiasm, anger, and anxiety can lead voters to reach their decisions through different means, depending on whether the emotion confirms or violates partisan expectations. Overall, the results of four sets of analyses, utilizing both cross-sectional and panel data, and examining candidate evaluation and defection, change in issue preferences over the course of campaigns, and time until vote decision, converge on the conclusion that an expectancy violation perspective offers a more general framework than AIT for understanding variability in political judgment strategies. AIT is consistent with some of our findings but cannot account for all patterns observed. The reason, we argue, is that AIT posits only unconditional effects of emotions, without regard to the object with which a given emotion is associated.

Why It Matters

Insofar as the expectancy violation model is centered on partisan expectations, it should be especially relevant to the polarized nature of contemporary American politics. Perhaps most important, as elites have become more homogeneous within parties and more divisive between them, mass partisanship has intensified, making party identification the prime determinant and organizer of a range of political attitudes (e.g., Hetherington 2001; Lavine, Johnston, and Steenbergen 2012). Moreover, as the parties have taken distinct positions on a variety of moral issues (e.g., abortion, gay rights, gun control), the notion of partisan identities as social identities—defining what Democrats and Republicans are stereotypically like as people (e.g., religious vs. secular )—has also intensified, leading the two partisan groups to hold increasingly negative feelings about each other and to sort into “echo chambers” where these feelings are reinforced (Iyengar, Sood, and Lelkes 2012; Mason 2015; Miller and Conover 2015). According to EVT, this “affective polarization” along party lines should facilitate expectancy-confirming feelings and beliefs, producing political judgments that are reflexive, shallow, and biased. The current polarized political environment is thus likely to favor directional goals over accuracy goals at the individual-level, contributing to greater political division.

EVT also provides insights into when accuracy goals may trump directional goals, potentially alleviating partisan polarization. In particular, economic performance can provide an important source of identity-conflicting considerations (thus making accuracy goals more salient). It is well documented that economic downturns damage the president’s party in presidential and congressional elections, and that economic expansions have the opposite effect (Lewis-Beck and Stegmaier 2000). From the perspective of EVT, economic swings should not only sway the parties’ electoral fortunes, they should lead some groups of partisans—namely, those for whom economic change defies partisan expectations—to think less like reflexive partisans and more like reflective deliberators. For example, after the economic meltdown in the fall of 2008, many Republicans evinced deeper, more evenhanded thinking (see Lavine, Johnston, and Steenbergen 2012). This research raises questions about the thresholds associated with open-mindedness in the face of expectancy-violating information: when does the switch to reflective deliberation occur (Redlawsk, Civettini, and Emmerson 2010)? It also raises questions regarding the types of people who are most susceptible to such effects. Although beyond the scope of the present paper, this is an important area for future research.

Finally, our work speaks to the large, contemporary literature on emotions in politics. This literature tends to view discrete emotions as having a consistent influence on information processing and judgment across various contexts. For example, anger increases the motivation to participate but decreases the motivation to deliberate as a general matter (MacKuen et al. 2010). Our paper suggests that emotional influences are not so straightforward. Specifically, if discrete emotions are elicited by an object rather than being diffuse and unattached—the latter being the case in much experimental work on emotions (e.g., Brader 2006; Lerner and Keltner 2001)—their influence will be conditional on the individual’s relationship to that object. This is an intuitive point, because emotions like anger, anxiety, and enthusiasm have evaluative implications in addition to any epistemic implications regarding uncertainty in one’s environment. It matters a great deal for one’s subsequent political behavior, we think, whether one is angry because a political enemy did something wrong or an ally did something wrong. Nonetheless, we are unaware of any work in political behavior that develops the importance of object-attachment in any depth when thinking about the political effects of discrete emotions. This would seem a ripe area for future research. It would be particularly interesting, for instance, to manipulate discrete emotions like anger and enthusiasm and the nature of the object eliciting these within a single experimental design. Such a study would provide a particularly powerful means of examining variation in the influence of emotions across contexts.

Where Does This Leave AIT?

What, then, are we to make of AIT in light of the evidence presented in this research? Most fundamentally, we believe AIT is best considered a special case of a more general expectancy violation perspective. This seems to also be, at least implicitly, the position taken by Marcus (2002). Consider this description of the affective intelligence model:

[The surveillance system] has access to the current plan of action and what we can normally expect in the environment . . . By matching the anticipated conditions of the environment with the most recent reports, the surveillance system can tell whether or not the environment is cooperating . . . it finds either a match (the environment is as expected) or a mismatch (something about the environment doesn’t fit the particular expectations of this course of action. (Marcus 2002, 100)

This core statement about the surveillance system says nothing about anxiety per se, unless anxiety is defined, tautologically, as the surveillance system itself. Indeed, as described, the surveillance system appears to function simply as a general detector of expectancy violation. Its purpose is to monitor the environment against theory-based expectations and provide feedback about whether standing habits will suffice in responding to the political landscape. As we see it, the experience of anxiety is simply one possible concomitant of a mismatch between theory-based expectations and the environment—but so is anger or even enthusiasm when these are directed at relevant objects. Anxiety is therefore not necessary to induce the types of dynamics posited by both theories. By placing anxiety at the center of the theory, AIT is (at present) too limited to account for the wide range of situations that may induce shifts in processing and judgment style. ¹¹ In sum, we believe that by equating the activation of the surveillance system with the experience of anxiety, AIT unnecessarily restricts its applicability.

A counterargument from an AIT perspective might be that expectancy violation always produces anxiety, even if the self-reported emotion suggests otherwise. For example, one might argue that the experience of out-candidate enthusiasm is, by its very nature, anxiety-producing, in that one finds it unusual to feel enthusiastic about a candidate of the opposing party. This seems to make sense if we equate anxiety with decision uncertainty, but then it is unclear how AIT adds anything to an expectancy violation account. If anxiety simply is uncertainty within AIT, then it seems confusing to call it anxiety, because the term has an intuitive meaning that suggests something more than this (e.g., Brader’s 2006 experiments, in which anxiety is manipulated through threatening images and sound, or the anxiety items in the ANES that ask about being “afraid”). If anxiety is just uncertainty, then it is not clear what theoretical work is being done by appealing to discrete emotions.

However, perhaps epistemic uncertainty is discomforting in a way that resembles anxiety. In the present study, for example, respondents who report enthusiasm about the out-party candidate may experience internal conflict or dissonance as a result of their long-standing negative feelings toward the opposing party and their contemporaneous positive feelings about the out-party’s primary representative (Lavine, Johnston, and Steenbergen 2012). It is easy to imagine that these conflicting feelings can cause discomfort in a way that might be conceptually related to the emotion of anxiety. It is not obvious how to resolve this possibility, but even here it seems to us that anxiety plays only a secondary, mediating role, and is therefore not the proper focus for theory building. Indeed, even if anxiety mediates the change from reflexive to reflective judgment in this scenario, the prime mover of those changes is still expectancy violation—for example, the experience of enthusiasm toward the out-party candidate.

A different way to resolve this problem might be to reject ANES-based measures of emotions. For example, in asking “Has John McCain made you feel afraid?” we might be capturing generic good-bad evaluations rather than the discrete emotions of anger, anxiety, and enthusiasm. Some evidence of this can be seen in the high correlation between anxiety and anger in the 2008 panel study (~.75; see also Lodge and Taber 2013 for a similar perspective). From this perspective, we might once again suggest that respondents who report positive evaluations of the out-party candidate or negative evaluations of the in-candidate actually experience anxiety as a result, and our problem is one of measurement rather than theory. Even here, however, we see an expectancy violation perspective as more parsimonious, because it is not clear what work is being done by appealing to discrete emotions. That is, in this scenario, a full account of changes in judgment style requires only two pieces of information: (1) Is the evaluation of the object positive or negative? (2) Is the object associated with the in-party or the out-party? At no point do we need to introduce the additional discrete emotions of anxiety, anger, or enthusiasm—it is enough to know the valence of the evaluation and the nature of the object, which is another way of saying that it is enough to know whether the evaluation is expectancy-confirming or expectancy-violating.

We do agree, however, that the ANES emotions items are far from ideal, and that future research will need to find other ways of testing AIT against alternatives. For example, Brader (2006) finds support for AIT using experimental inductions of emotion. Thus, an alternative strategy for distinguishing the two perspectives would involve experimental methods that utilize anxiety-, enthusiasm-, or anger-based inductions of expectancy confirmation and violation. Should anxiety-based inductions outperform the others in producing changes in political judgment across the board then AIT would be bolstered. If, however, the pattern of changes were to stem from expectancy confirmation versus violation—rather than one discrete emotion or another—EVT would be bolstered.

Overall, our results suggest that it is worthwhile to seriously consider alternative theories, or to consider how AIT may be integrated with more general dual-process models of information processing and judgment. This is far from a settled matter, and new empirical approaches may yield important insights.

Expectancy Violation as a Meta-theoretical Perspective

Partisanship serves as a powerful device in simplifying the task of political judgment: by relying on knowledge, expectations, and beliefs about the two major political parties, partisan cues can help citizens make predictions about the political world (e.g., the preferences of partisan elites on policy issues). More often than not, partisan expectancies are confirmed, either because elites structure the environment along a single partisan-ideological dimension (Jackman and Sniderman 2002) or because citizens attempt to fit new evidence into existing theories (Druckman 2012; Lodge and Taber 2013). Political reality is not always so cooperative, however, and motivated reasoning is not unbounded (Hillygus and Shields 2008; Kunda 1990; Lavine, Johnston, and Steenbergen 2012). When the contemporary political world is strongly at odds with partisan expectations, citizens are likely to set aside their partisan lenses and turn to more effortful judgment strategies to reach their confidence thresholds. This general approach—in which expectancy violation provides the basis for a meta-theoretical conception of political judgment—views citizens as engaged in a constant negotiation among potentially competing judgment goals: making “good,” accurate, or justifiable judgments, making them relatively easily (with minimal cognitive effort), and preserving the validity of one’s prior beliefs and commitments.

The concept of expectancy violation ties together a number of recent theoretical approaches to political reasoning, including Hillygus and Shields’ (2008) “persuadable voter,” Lavine, Johnston, and Steenbergen’s (2012) “ambivalent partisan,” and Sniderman and Stiglitz’s (2012) “reputational premium” and “order rule.” In each of these accounts, citizens rely on standing partisan habits and eschew more deliberative forms of reasoning when the political environment is structured in a predictable manner. When the fit between expectancies and reality is weak, however, citizens step up their efforts to reach confident judgments. It is perhaps this tactical flexibility—and not the use of heuristic or systematic forms of reasoning per se—that is the embodiment of rationality and good citizenship.