Diagnosed Anxiety and Political Judgment, WORD COUNT: 2,750 Diagnosed Anxiety Disorder and Political Judgment: An Empirical Test of Affective Intelligence Theory Christopher D. Johnston Department of Political Science Duke University Durham, NC 7708 christopher.johnston@duke.edu Howard Lavine Department of Political Science University of Minnesota Minneapolis, MN 55455 lavine@umn.edu 1 Diagnosed Anxiety and Political Judgment, 2 Abstract Political reasoning is contingent; sometimes voters rely primarily on their partisan habits, while at other times they take in more (or better) information and engage in more effortful reflection. Affective intelligence theory (AIT) – a theory of the impact of emotion on political judgment strategies – has risen to prominence as an account of the underlying mechanisms by which this heterogeneity occurs. However, past tests of the theory suffer from: (1) endogeneity (candidate evaluations causing emotions); and/or (2) confounds in the measurement of emotion. In this research, we overcome these limitations by employing a unique aspect of the 2008 ANES panel study: an item in the March wave in which respondents were asked whether they had been diagnosed with an anxiety disorder in the previous twelve months. Using this measure of anxiety (which is appropriately separated from candidate evaluation) and a statistical model capable of isolating causal order, we find strong support for the central dynamics of AIT. Diagnosed Anxiety and Political Judgment, 3 Contemporary models of political behavior posit that citizens are frugal with their cognitive resources, preferring low effort strategies when these offer a reasonable degree of judgment confidence. It is only when such strategies fail to reduce uncertainty that citizens will be motivated to employ more effortful modes of deliberation (Basinger and Lavine 2005; Hillygus and Shields 2008; Lavine, Johnston and Steenbergen 2012; Marcus, Neumann and MacKuen 2000). Within this general dual-process framework, affective intelligence theory (hereafter, AIT; Marcus et al. 2000) has risen to prominence as an account of the underlying mechanisms by which this process-related heterogeneity occurs. Relying on insights from neuroscience (e.g., Damasio 1994; Grey 1981; LeDoux 1996), Marcus and colleagues argue that the nature of political reasoning lies in the experience of emotion. Specifically, two key emotions, enthusiasm and anxiety – viewed as the products of functionally distinct biobehavioral regulatory systems – influence how citizens construct their political judgments. The “disposition system,” which mediates the emotion of enthusiasm, responds to positive incentives by directly initiating habit-based behavior. A second system, referred to by Marcus et al. (2000) as the “surveillance system,” detects threat and danger in the environment. Activation of the surveillance system results in the emotion of anxiety, which interrupts ongoing habitual action and promotes increased thoughtfulness and greater motivation for learning. Applied to the realm of mass political judgment, enthusiasm is expected to motivate political participation and to increase voters’ reliance on their political habits. Anxiety, by contrast, is expected to lead voters to turn away from habitual responses and devote more attention to diagnostic information in the environment. In line with AIT, Marcus et al. (2000) have shown that anxious voters (a) pay greater attention to the political environment and acquire more information about candidates’ policy stands; (b) rely less on partisanship and more on Diagnosed Anxiety and Political Judgment, 4 policy preferences and assessments of candidate character in forming candidate evaluations; and (c) defect at higher rates in presidential elections (see also Marcus and MacKuen 1993; Brader 2006). Finally, anxiety – though not anger or enthusiasm – has been found to increase the quantity and quality of information processing, including the desire for balance and compromise (MacKuen et al. 2010; Valentino, Hutchings, Banks and Davis 2008). Taken together, this work suggests that anxiety leads to political decisions that are more informed by contemporary information and less by partisan loyalty. In the language of dual-process theories of cognition, anxiety reduces heuristic processing and stimulates systematic processing. Despite its appeal, most empirical tests of AIT suffer from two important shortcomings: endogeneity and confounds in the measurement of emotion. First, nearly all AIT-inspired research relies on cross-sectional data, so that the variables assessing emotion and judgment are measured at the same time. It is therefore possible that the former is endogenous to the latter. Ladd and Lenz (2008) proposed such an endogenous affect mechanism to account for the indirect effect of anxiety in a pooled cross-sectional analysis using ANES data. They argued that the stronger influence of policy agreement on candidate evaluation among anxious (vs. non-anxious) voters could occur in the absence of anxiety playing a causal role. If endogenous affect is responsible for the interactions (between anxiety and partisanship and anxiety and issue orientation), they should disappear when anxiety is measured at a lagged interval. This is exactly what Ladd and Lenz (2008) found. A second, related problem is that emotion is almost always measured with respect to a specific candidate. For example, the traditional ANES emotion battery reads: “Has [candidate] – because of the kind of person he is, or because of something he has done – made you feel [emotion term]?” This object-centered measurement approach generates responses that are better Diagnosed Anxiety and Political Judgment, 5 characterized as object-evaluations – i.e., attitudes – than as discrete emotions. That is, when answering a question about whether Barack Obama has ever “made you feel anxious,” voters are likely to rely on their overall attitudes toward Obama, rather than attempt to recall whether he has engendered a specific emotion such as anxiety, anger, or sadness. It isn’t clear, then, whether the motives underlying variation in judgment effort are due to the experience of anxiety per se (i.e., a discrete emotion) or because of attitudinal ambivalence, which has been shown to exert similar effects on political judgment (e.g., Basinger & Lavine, 2005; Lavine et al. 2012). Given the prominence of AIT in contemporary studies of political judgment, it is important to conduct tests of the theory in which the key variable of anxiety is operationalized independent of candidate evaluation (i.e., the DV), and with statistical models that can better isolate causal mechanisms. We further these goals by utilizing a unique aspect of the 2008 ANES panel study. The (panel) structure of the data allows us to rule out endogeneity as an alternative explanation for the observed empirical patterns. Moreover, we leverage a happy coincidence of the inclusion (for reasons unrelated to AIT) of information regarding respondents’ status as diagnosed (or not) with an anxiety disorder. This allows us to operationalize AIT’s key variable in a way that is uncontaminated by attitudes toward the political outcome variables of interest. The combination of these two features allows for a particularly appealing and unique test of this prominent theory. Hypotheses According to AIT, anxiety initiates a shift from reliance on standing partisan habits to a more systematic mode of information processing in which citizens think more carefully about contemporary information in the political environment. Thus, with respect to candidate Diagnosed Anxiety and Political Judgment, 6 evaluation, we expect anxious voters to rely less on partisan cues and more on policy preferences than non-anxious voters. In the realm of preference formation over issues, we expect updating among anxious citizens to occur more on the basis of policy-relevant considerations (e.g. material interests), and less on the basis of simple partisan cues, and that non-anxious citizens will show the reverse pattern (i.e., privileging partisan identity). We test these general hypotheses by modeling candidate evaluations, tax preferences, and health care preferences across the 2008 campaign season. Methods Data To test our hypotheses, we rely on data from the 2008 ANES Panel Study. Respondents participated in one survey per month throughout 2008. Of these, six (January, February, June, September, October, and November) were focused on political content related to the 2008 elections. The “off-wave” surveys consisted of non-political content selected by groups helping to finance the study. We take advantage of three characteristics of this study. First, evaluations of the two major party presidential candidates (McCain and Obama) were included in both the February and October waves. By controlling for prior candidate evaluations, we can examine how voters’ electoral preferences changed over the course of the campaign, and thus better isolate the impact of different considerations on evaluations. Second, two policy issues which were particularly salient during this election – health care reform and taxes – were included in both the January and October waves of the study. As with the modeling of candidate evaluation, this allows us to examine how respondents updated their policy preferences over the course of the campaign, and to determine the extent to which such changes were driven by partisanship Diagnosed Anxiety and Political Judgment, 7 versus more substantive considerations. Third, in the March wave of the study, respondents were asked whether they had been diagnosed with an anxiety disorder in the previous twelve months. We utilize this dichotomous indicator to operationalize the key variable of anxiety.1 Anxiety. Anxiety was assessed with the following single item (with follow-up) “Have you ever been told by a doctor or other health professional that you had an anxiety disorder?” If yes, a follow-up question was asked: “During the past twelve months, have you had an anxiety disorder?” Anxiety was coded one if the respondent reported being diagnosed with an anxiety disorder in the previous twelve months (10% of the sample), and zero otherwise. Substantively, this measure can be understood as increasing the probability of experiencing anxiety at any given time during the campaign, and thus as an instrument for the actual experience of anxiety during the process of political judgment. To alleviate concerns that diagnosed anxiety is simply a proxy for cognitive ability or political engagement (or other factors), in Figure A1 of the Appendix we provide correlations between anxiety diagnosis and (1) education, (2) political knowledge, (3) income, and (4) partisan strength. All of these correlations turn out to be negative in direction (i.e., anxiety is linked to less education, income, weaker party ID, etc.). These correlations (which we control for in any case) thus work against our hypotheses. Comparative Candidate Evaluation. Evaluations of Obama and McCain – each measured on 7-point scales – were solicited in the February and October waves of the panel study. In each wave, we subtracted the Obama evaluation from the McCain evaluation, and recoded these to range from zero to one. 1 For item wordings, see Appendix. Diagnosed Anxiety and Political Judgment, 8 Issue Orientation. Eight policy issues were included in the January wave of the survey: gay marriage, tax increases on wealthy Americans, prescription drugs for senior citizens, universal health care, suspension of habeas corpus for suspected terrorists, warrantless wiretapping, temporary work visas for non-citizens, and an eventual path to citizenship for illegal immigrants. Consistent with Marcus et al. (2000), we averaged responses to these issues to create an overall scale representing the respondent’s issue orientation (alpha = .61 for the January scale and .69 for the October scale). Income and Perceptions of Health Care. We assessed respondents’ reliance on substantive considerations in the updating of preferences over taxes and health care with two measures. For the taxes model we utilize household income (measured prior to the start of the panel study). For the healthcare model we use the respondent’s reported perception of the change in healthcare quality from 2001 to 2008. We expect respondents who perceive a negative change in quality to prefer liberal reforms to the health care system. Partisanship. Respondents’ party identification was measured with the traditional sevenpoint scale in the January wave. Controls. We control for several variables, including: age, gender (1=male), race (two dummy variables, one for black and one for Hispanic respondents), education, income, and political sophistication. All variables were recoded to range from zero to one. Diagnosed Anxiety and Political Judgment, 9 Results Candidate Evaluations Based on AIT, we predicted that anxiety would decrease the influence of partisanship and increase the influence of issue orientation on candidate evaluations. Using our measure of diagnosed anxiety, we find an empirical pattern consistent with AIT. As shown in the first three columns of Table 1, the interaction of anxiety with partisanship is negative and statistically significant (B=-.11, p<.05), indicating that anxiety tempers the influence of partisanship on comparative candidate evaluation over the course of the campaign.2 We also find support for AIT’s compensatory dynamic, that anxiety significantly increases voters’ electoral reliance on issue orientation (B=.18, p<.05). Thus, in a context in which (a) there is little concern for reverse causality, and (b) anxiety is operationalized separately from candidate evaluation itself, we find strong support for the central dynamics of AIT. To better interpret the substance of these findings, in Figure 1 we plot the marginal effects of partisanship and issue orientation across the two values of anxiety. For non-diagnosed respondents, the expected change in relative preference for McCain, moving from strong Democrat to strong Republican, is .28, or 28% of the range of the scale. This is a substantial change, as it controls for initial candidate preferences in February. For diagnosed respondents, however, this effect drops by 10 percentage points, to .18. With respect to issue orientation, the effect among non-diagnosed respondents is .26, while for diagnosed respondents it increases by a substantial 18 percentage points, to .44. In sum, among non-anxious individuals, partisanship and issue orientation exert approximately equally strong effects on candidate evaluation; by contrast, among anxious individuals, the effect for issues is about 2.5 times that of partisanship. 2 As we had strong priors (derived from AIT), we use one-tailed tests for our key hypotheses. However, we report the two-tailed p-values for all variables in Table 1. Diagnosed Anxiety and Political Judgment, 10 Policy Preference Updating We turn now to the results for policy preferences in October, controlling for preferences in January. These results are shown in the middle and right columns of Table 1. For taxes, we find a pattern consistent with AIT, although the results are somewhat weaker than for candidate evaluations. First, as expected, anxiety significantly decreases the impact of partisanship on preference updating over the course of the campaign (B=-.17, p<.05). Moreover, anxiety increases the influence of income on tax preferences, but this effect does not quite reach statistical significance (B=.19, p<.15). The overall pattern, however – graphed in Figure 2a – is highly consistent AIT’s predictions. For non-anxious respondents, the predicted marginal effect of partisanship (.24) is much larger than that of income (.04). As the theory predicts, this pattern is reversed for anxious respondents, with predicted marginal effects of .24 and .07 for income and partisanship, respectively. Finally, we turn to the results for healthcare preferences, found in the last three columns of Table 1, and plotted in Figure 2b. Although the relevant interactions fail to reach conventional levels of statistical significance, the pattern of effects is in line with AIT. Specifically, for nonanxious respondents, the predicted marginal effect of partisanship (.22) is much larger than that for healthcare perceptions (.08). For anxious respondents, however, the impact of perceptions doubles (to .16) and the impact of partisanship decreases slightly (to .18). Conclusion An important thrust of contemporary work on mass politics is that political reasoning is contingent: sometimes voters rely primarily on their partisan habits, ignoring more diagnostic information (even if they are capable of acquiring it), while at other times they take in more (or better) information and engage in more effortful reflection. The purpose of our investigation was Diagnosed Anxiety and Political Judgment, 11 to provide an improved test of the leading theory of process-related heterogeneity, affective intelligence theory (Marcus et al. 2000). Given the prominence of the theory in political science, as well as the increasing interest in emotion among social scientists generally, we believe it is critical to test AIT within a more auspicious context, one that overcomes the twin problems of endogeneity and confounds of measurement. In three tests, we find meaningful support for AIT. In the core model predicting candidate evaluations, we find strong support for AIT’s signature dynamics: anxiety decreases voters’ reliance on partisan habits, and increases their reliance on the more normatively pleasing (but cognitively demanding) dimension of issue preferences. We conceptually replicated this dynamic in our analysis of preference updating on salient policy issues. Over the course of the 2008 campaign, changes in preferences on taxes and healthcare were driven more by partisanship among non-anxious individuals, and driven more by either material interest or perceptions of the quality the policy status quo (healthcare) among anxious individuals. Taken together, this pattern of findings – protected as they are from both endogeneity and concerns about the object-related measurement of anxiety – provides strong evidence for one of the leading theories in the discipline. Our analyses are not, however, without their own limitations. First, in utilizing a panel design, we are better able to isolate the causal effects of key variables, but this comes at the expense of statistical power. By controlling for lagged values of key dependent variables, especially during time spans of only several months, these tests are conservative. Second, although our unique operationalization of diagnosed anxiety does not suffer from problems of object association, it is more diffuse than candidate-centered measures. Specifically, we assume that diagnosed anxiety is a proxy – albeit a rather inefficient one – for the probability of Diagnosed Anxiety and Political Judgment, 12 experiencing anxiety at any given time point; including during a context in which political judgments are made in a campaign. These caveats should not be taken as apologies for the limitations of the study, but simply as an attempt to place bounds on the interpretation of our results. In conclusion, this study, for its own limitations, provides a needed empirical reexamination of AIT. Our analysis, shorn of concerns with endogeneity and confounded measurement, indicates that anxiety does indeed move citizens to eschew toeing the party line, and to consult more substantive considerations in making their judgments. In a more general vein, the concept of democracy implies the constraint of representatives by citizens. In a political world where partisan cues are utilized unflinchingly, citizens may become the effective agents of elite interests, as the latter learn that they may take actions (e.g. make policy statements) with impunity. The surveillance system, and its manifestation in the form of the experience of anxiety, as Marcus and his colleagues (1988, 1993, 2000, 2010) have long argued, represents a bulwark against this form of tyranny. Diagnosed Anxiety and Political Judgment, 13 REFERENCES Basinger, Scott, and Howard Lavine. 2005. “Ambivalence, Information and Electoral Choice.” American Political Science Review 99(2): 169-184. Brader, Ted. 2006. Campaigning for Hearts and Minds: How Emotional Appeals in Political Ads Work. Chicago: The University of Chicago Press. Chaiken, Shelly, Akiva Liberman, and Alice Eagly. 1989. Heuristic and Systematic Processing Within and Beyond the Persuasion Context. In Unintended thought: Limits of awareness, intention, and control, eds. Jim Uleman and John Bargh. New York: Guilford, 212-252. Damasio, Antonio. 1994. Descartes Error: Emotion, Reason and the Human Brain. New York: G.P. Putnams Sons. Gray, Jeffrey A. 1981."The Psycho-physiology of Anxiety." In Dimensions of Personality, ed. Richard Lynn. New York: Pergamon, 233-252. Hillygus, D. Sunshine, and Todd G. Shields. 2008. The Persuadable Voter: Wedge Issues in Presidential Campaigns. Princeton: Princeton University Press. Ladd, Jonathan, and Gabriel Lenz. 2008. “Reassessing the Role of Anxiety in Vote Choice.” Political Psychology 29(2): 275-296. Lavine, Howard, Christopher Johnston, and Marco Steenbergen. 2012. The Ambivalent Partisan: How Critical Loyalty Promotes Democracy. New York: Oxford University Press. Ledoux, Joseph. 1996. The Emotional Brain: The Mysterious Underpinnings of Emotional Life. New York: Simon and Schuster. MacKuen, Michael, Jennifer Wolak, Luke Keele, and George E. Marcus. 2010. “Civic Engagements: Resolute Partisanship or Reflective Deliberation.” American Journal of Political Science 54 (2): 440-458. Marcus, George. 1988. “The Structure of Emotional Response: 1984 Presidential Candidates.” American Political Science Review 82(3): 737-761. Marcus, George and Michael MacKuen. 1993. “Anxiety, Enthusiasm and the Vote: The Emotional Underpinnings of Learning and Involvement During Presidential Campaigns.” American Political Science Review 87(3): 672-685. Marcus, George, W. Russell Neuman, and Michael Mackuen. 2000. Affective Intelligence and Political Judgment. Chicago: The University of Chicago Press. Valentino, Nicholas, Vincent Hutchings, Antoine Banks, and Anne Davis. 2008. “Is a Worried Citizen a Good Citizen? Emotions, Political Information Seeking, and Learning Via the Internet.” Political Psychology 29(2): 243-273. Diagnosed Anxiety and Political Judgment, 14 APPENDIX A. QUESTION WORDINGS FOR SELECTED MODEL VARIABLES Candidate Evaluations Do you like Barack Obama (John McCain), dislike him, or neither like nor dislike him? Do you like/dislike him: a great deal, a moderate amount, or a little? Policy Issues (All measured on a seven-point scale indicating degree of support or opposition) Do you favor, oppose, or neither favor nor oppose an amendment to the U.S. Constitution banning marriage between two people who are the same sex? Do you favor, oppose, or neither favor nor oppose raising federal income taxes for people who make more than $200,000 per year? Do you favor, oppose, or neither favor nor oppose the U.S. government paying for all of the cost of prescription drugs for senior citizens who are living on very little income? Do you favor, oppose, or neither favor nor oppose the U.S. government paying for all necessary medical care for all Americans? Imagine that the U.S. government suspects a person in the United States of being a terrorist. Do you favor, oppose, or neither favor nor oppose the government being able to put this person in prison for months without ever bringing the person to court and charging him or her with a crime? Do you favor, oppose, or neither favor nor oppose the U.S. government being required to get a court order before it can listen in on phone calls made by American citizens who are suspected of being terrorists? Citizens of other countries who have come to live in the United States without the permission of the U.S. government are called "illegal immigrants." Do you favor, oppose, or neither favor nor oppose allowing illegal immigrants to work in the United States for up to three years, after which they would have to go back to their home country? Do you favor, oppose, or neither favor nor oppose the U.S. government making it possible for illegal immigrants to become U.S. citizens? Diagnosed Anxiety and Political Judgment, 15 Health Care Perceptions Compared to 2001, would you say the following is now: much better, somewhat better, about the same, somewhat worse, or much worse? Health Care in the U.S. Political Knowledge Do you happen to know how many times an individual can be elected President of the United States under current laws? For how many years is a United States Senator elected – that is, how many years are there in one full term of office for a U.S. Senator? How many U.S. Senators are there from each state? For how many years is a member of the United States House of Representatives elected – that is, how many years are there in one full term of office for a U.S. House member? According to federal law, if the President of the United States dies, is no longer willing or able to serve, or is removed from office by Congress, the Vice President would become the President. If the Vice President were unable or unwilling to serve, who would be eligible to become president next? The Chief Justice of the Supreme Court, the Secretary of State, or the Speaker of the House of Representatives? What percentage vote of the House and the Senate is needed to override a Presidential veto? A bare majority, two-thirds, three-fourths, or ninety percent? Diagnosed Anxiety and Political Judgment, APPENDIX B. CORRELATIONS OF DIAGNOSED ANXIETY WITH POLITICAL MODERATORS -.3 -.2 -.1 0 .1 .2 .3 Figure A1. Correlations of Diagnosed Anxiety with Political Variables Education Sophistication Income Notes: Bars are estimated gammas for each political variable with diagnosed anxiety. Partisan Strength 16 Diagnosed Anxiety and Political Judgment, Figure 1. Effects of PID and Issues on Candidate Evaluations General Issue Preferences 0 .1 .2 .3 .4 .5 .6 Partisanship No Anxiety Notes: Bars represent predicted marginal effects. Spikes are 95% CIs. Diagnosed Anxiety 17 Diagnosed Anxiety and Political Judgment, Figure 2A. Effects of PID and Income on Tax Preferences Income -.2 0 .2 .4 .6 Partisanship No Anxiety Diagnosed Anxiety Notes: Bars represent predicted marginal effects. Spikes are 95% CIs. Figure 2B. Effects of PID and Perceptions on Health Care Preferences Health Care Perceptions No Anxiety Diagnosed Anxiety -.2 0 .2 .4 .6 Partisanship Notes: Bars represent marginal effects. Spikes are 95% CIs. 18 Diagnosed Anxiety and Political Judgment, 19 Table 1. Regression Estimates for All Models ______________________________________________________________________________ Evaluations Taxes Health Care Variables B SE p B SE p B SE p ______________________________________________________________________________ Partisanship .25 .05 .00 -.07 .10 .48 .20 .11 .06 PID X Anxiety -.11 .06 .07 -.17 .10 .08 -.06 .11 .59 Issues .09 .10 .35 Issues X Anxiety .18 .10 .08 Income -.02 .03 .49 .09 .15 .56 .15 .06 .02 Income X Anxiety .19 .17 .27 Health Perceptions .13 .15 .40 Health X Anxiety .10 .18 .58 Anxiety -.03 .04 .41 -.05 .12 .66 -.04 .07 .57 Lagged PID X Soph .56 .05 .03 .07 .00 .53 .59 .47 .03 .13 .00 .00 .55 .04 .03 .15 .00 .81 Ideo/Income/Health X Soph .25 .14 .07 -.06 .21 .76 -.07 .23 .74 Sophistication Age Male Black Hispanic Education Constant -.19 .00 -.01 -.07 -.04 -.06 .16 .05 .03 .01 .02 .03 .02 .05 .00 .95 .33 .00 .09 .01 .00 -.21 .05 .04 .03 -.01 .05 .08 .15 .05 .02 .05 .05 .04 .10 .15 .30 .07 .54 .80 .29 .44 .05 -.03 .03 -.05 -.08 .01 -.01 .10 .06 .02 .05 .06 .05 .08 .62 .63 .22 .36 .14 .82 .95 R2 .69 .58 .56 N 1026 531 529 ______________________________________________________________________________ Notes: Data from 2008 ANES Panel Study. Entries are OLS estimates and standard errors. All variables are coded on a zero to one scale.