Jump to ContentJump to Main Navigation
The Affect EffectDynamics of Emotion in Political Thinking and Behavior$

George E. Marcus, W. Russell Neuman, and Michael MacKuen

Print publication date: 2007

Print ISBN-13: 9780226574417

Published to Chicago Scholarship Online: March 2013

DOI: 10.7208/chicago/9780226574431.001.0001

Show Summary Details
Page of

PRINTED FROM CHICAGO SCHOLARSHIP ONLINE (www.chicago.universitypressscholarship.com). (c) Copyright University of Chicago Press, 2021. All Rights Reserved. An individual user may print out a PDF of a single chapter of a monograph in CHSO for personal use. Subscriber: null; date: 26 February 2021

Testing Some Implications of Affective Intelligence Theory at the Aggregate Level

Testing Some Implications of Affective Intelligence Theory at the Aggregate Level

(p.316) Chapter Thirteen Testing Some Implications of Affective Intelligence Theory at the Aggregate Level
The Affect Effect

Peter F. Nardulli

James H. Kuklinski

University of Chicago Press

Abstract and Keywords

This chapter investigates the electoral politics of the past thirty years to see how political dynamics have been moved by threats to economic prosperity, individual safety, and collective physical security. The conception of democratic governance emphasizes change and dynamics; affective intelligence predicts only that surveillance will occur under high but not low anxiety. The increased anxiety evoked by bad or worsening conditions does not produce irrational, unthinking reactions. The propositions were derived with minimal guidance from affective intelligence theory. Anger decreased estimates of risk and thus promoted support of the war; anxiety worked in the opposite way. Fear is another plausible substitute for anxiety. Adopting fear or anger would work just as effectively as adopting anxiety. Without attention to emotions, political scientists cannot fully understand politics.

Keywords:   electoral politics, political dynamics, individual safety, economic prosperity, physical security, democratic governance, affective intelligence theory, anxiety, anger, fear

Psychologists study emotions for their own sake. Political scientists care about emotions only if they further understanding of politics. Accordingly, this chapter begins with a conception of democratic governance and only then asks how incorporating emotions might enhance it. Three core desires—the desires for economic prosperity, individual physical safety, and collective physical security—serve as the conception's foundation. The conception predicts that when threats to these core desires arise and politicians fail to address them, an electorate that normally might not be watching voices its discontent at the ballot box.

Initially, we borrow from affective intelligence theory devised by Marcus and colleagues (Marcus, Neuman, and MacKuen 2000; MacKuen, Marcus, Neuman, and Keele, chapter 6 in this volume) and assume anxiety to be the triggering mechanism: threats to core desires increase anxiety, in turn evoking intense surveillance and departures from routine voting behavior.1 We derive aggregate-level implications from this assumption and find some empirical support for them. When deriving the implications, however, we find affective intelligence theory to be lacking. It is silent as to the long-term dynamics between threats and anxiety or between (p.317) anxiety and political decision making. The propositions thus represent our best guesses, but they are guesses, and their confirmation hardly represents indisputable support for the theory.

We then turn to two topics that in one form or another pervade many of the chapters in this volume: affective intelligence's singular focus on anxiety and the value gained from including emotions in the study of decision making. Whereas other authors address them from the perspective of individual-level mental processes, our aggregate-level perspective puts considerable distance between the workings of the human mind and the broad contours of (American) politics. From this perspective, some of the finer distinctions that motivate political psychology (between types of emotions, for example) look unimportant, at least given the current and still immature state of research in neuroscience and psychology.

A Conception of Representative Democracy, Anxiety Included

Most studies of democratic governance adopt one of two perspectives. Beginning with Miller and Stokes (1963) and continuing through the 1970s, scholars used (typically cross-sectional) survey data to consider whether the actions of political elites, primarily legislators, correspond to the ideological leanings and policy preferences of their constituents. This bottom-up perspective, which views legislators' responsiveness as the key to effective democratic representation, found little to modest evidence of it, despite the strong assumption that causal influence flows completely from constituencies' preferences to legislators' roll-call behavior. Later, scholars turned to survey and laboratory experiments to demonstrate that elites shape the ways in which citizens think about issues. This work showed, for example, that people express more support for affirmative action when encouraged to view it as a remedy for past discrimination against African Americans than when encouraged to view it as reverse discrimination (Kinder and Sanders 1990; but see Druckman 2001; Druckman and Nelson 2003; Sniderman and Theriault 2004). In the vein of Schattschneider (1960), this top-down perspective places political elites in the driver's seat. Taken together, the two research traditions offer a minimal role for ordinary citizens; if politicians aren't ignoring them, they are telling them how and what to think.

The conception we offer immediately below differs from either of the two principal perspectives. Its key elements are a shared set of core political desires and the rise and fall of threats to those desires. It predicts that (p.318) the rise of such threats, especially when they become severe, will cause citizens to react politically.

McCubbins and Schwartz's distinction (1984) between how police and fire departments function captures the essence of our conception. Just as members of fire departments do not constantly monitor their communities for fires, so citizens do not constantly monitor politics for purposes of evaluating their political agents' performances. Rather, fire alarms—conditions that threaten core political desires—draw citizens' attention. Especially when severe, multiple, and enduring, these fire alarms generate collective electoral jolts that affect politicians' electoral fortunes.

Core Political Desires

Whereas most studies of democratic governance focus heavily on citizens' preferences regarding current issues, ranging from race and abortion to size of the military, we begin with something more fundamental: core political desires. In Western societies, these core desires include general yearnings—for liberty, democracy, and happiness—as well as a set of derivative and more specific yearnings—for safety and security, economic prosperity, and equal treatment under the law. This set of yearnings falls second on Maslow's hierarchy of needs (1968) after physiological needs such as sex, food, and sleep, and citizens in developed nations expect them to be met. Indeed, their fulfillment constitutes the very definition of well-being in such societies.

In the United States and elsewhere, ordinary citizens look to politicians and other democratic stewards to ensure safety, security, and economic prosperity. They expect policymakers to keep their streets and neighborhoods safe, to prevent and protect them from outside aggression, and to keep inflation and unemployment within reasonable bounds. That these themes dominate political rhetoric underlines the importance of core desires to democratic governance.

Threats to Core Values

Threats to core values constitute the other key component in our conception of democratic governance. We assume that most people recognize such threats when they exist. The threats, after all, take well-known forms, such as rising crime rates, rising unemployment levels, and looming prospects of war. Moreover, the media cover such threats widely and intensely, especially as their seriousness grows. Partisan politicians who stand to gain from the threats' existence dramatize the bad and worsening state (p.319) of affairs. For all these reasons, serious threats to core desires will appear on people's perceptual screens.

Threats come and go. If the country is at peace but the economy shows dramatic signs of faltering, the economy, not war, will grab most people's attention. In the absence of threats to any core value, that is, during times of peace, prosperity, and collective safety, partisan politics as usual will rule the day.

Note the importance of change, of ebb and flow, to the definition of a threat. A 10 percent unemployment rate represents a threat only because it is much higher than the 3 or 4 percent rate that preceded it. The identification of a condition as a threat, then, requires continuous comparisons to other conditions. Whether the current state of affairs represents a threat depends on how it stacks up vis-à-vis prior states, especially when citizens view relatively good conditions of the past as the norm.

But why do threats to core values, with their potentially damaging political consequences, occur at all? Why don't politicians always keep crime and unemployment rates at relatively low levels? One answer is that these phenomena ebb and flow independent of politics and policymaking. According to this view, politicians lack the capacity to preempt most potential threats. Alternatively, although politicians might possess the capacity to stop potential threats from becoming real, other demands take center court until potential turns to reality. Politicians, like citizens, respond to fire alarms. Because our interest is the citizenry's response to changing conditions, we do not adjudicate between these two explanations (Nardulli 2005 presents a fuller discussion). In any event, some combination of the two explanations probably comes closest to the truth.


According to the research program in affective intelligence (Marcus and MacKuen 1993; Marcus, Neuman, and MacKuen 2000), ominous-looking changes in the political environment initially gain citizens' attention. These out-of-the-ordinary changes activate the so-called surveillance system, which in turn produces anxiety. Heightened anxiety decreases reliance on habit and leads to a reconsideration of choices for purposes of changing the situation that evoked the emotion. Simply, and in the political context, anxious citizens begin to consider alternative political choices.2

(p.320) Although this core idea has great intuitive appeal, the limits of affective intelligence theory become immediately obvious on trying to derive implications about the dynamics between environmental changes and anxiety, on one hand, and anxiety and behavioral changes, on the other. For example, do higher levels of threat lead linearly to higher levels of anxiety, or does the relationship take a different form? Do sustained, long-lasting threats increase anxiety more than short-lived ones? Do two severe threats create more anxiety than one threat, or does one severe threat alone push anxiety to its limit?

As a theory, affective intelligence offers no answers and, in fact, fails to raise the questions. Neuroscience's current inability to offer guidance and the authors' reliance on cross-sectional data explain the void. In our view, these are the very sorts of questions that political psychologists must pose and try to answer.

Accordingly, we posit the following propositions. Although we offer a rationale for each, we emphasize, again, the lack of theoretical and empirical guidance from affective intelligence theory. Our conception of democratic governance emphasizes change and dynamics; affective intelligence predicts only that surveillance will occur under high but not low anxiety.

All of the propositions we offer below stem from the same set of assumptions. Common to all are the following:

  • Threats to common desires are potential sources of anxiety.

  • People differ as to the level of threat that creates anxiety.

  • People differ as to the level of anxiety that causes surveillance.

  • People differ as to the level at which surveillance causes them to change voting behavior.

These assumptions recognize, first, that threats to common desires can change behavior and, second, that the threat → anxiety → surveillance → change-in-behavior chain will reach completion more quickly and with greater effects among some people than others.

Three propositions follow directly from the assumptions.

The Magnitude Proposition

  • Prediction: The greater an external threat to core desires at any moment in time, the greater the change in aggregate voting behavior.

  • Rationale: Bigger threats cause more people to monitor their environments and change their behavior than do smaller threats.

(p.321) The Endurance Proposition

  • Prediction: The longer an external threat to core desires endures, the greater the change in aggregate voting behavior.

  • Rationale: Persistent threats produce anxiety and thus behavioral change among voters who did not become anxious earlier.

The Confluence Proposition

  • Prediction: The greater the number of core desires that are threatened at any moment in time, the greater the change in aggregate voting behavior.

  • Rationale: Simultaneous threats to two or more core values induce anxiety even among those with relatively high anxiety thresholds.

Figure 13.1 illustrates the logic of the magnitude, endurance, and confluence propositions using two domains: crime and unemployment. The horizontal axis represents time. Higher values on the vertical axis indicate greater threat and thus, presumably, a higher level of anxiety.

Note, first, that conditions in either domain can be better or worse at any particular point in time. So, for example, at time 3, the crime rate has gone up, increasing anxiety among some citizens who now see threats to their physical security. The unemployment rate, on the other hand, is too low to be a source of anxiety. Conversely, at times 6 and 7, deteriorating economic conditions induce anxiety by threatening some people's sense of economic security but the now-low crime rate does not. Second, bad conditions can endure for a long or a short time in either domain. Economic conditions, for example, deteriorate at three different times in the hypothetical example. Distinguishing the third time from the prior two is the enduring peak, which continues unabated for three years. In the real world, conditions that threaten people's core political desires vary considerably in their duration.

Finally, bad conditions in different domains can occur simultaneously, creating particularly serious threats to core political desires. In figure 13.1, crime and unemployment rates both peak at time 5. In the United States, civil unrest and crime rates both jumped markedly during the 1960s, much as inflation, unemployment, and international threats all did in the late 1970s. Such periods presumably generate especially high and widely shared levels of anxiety.

Rationality and deliberation also hold a place in affective intelligence theory. In their most widely cited work, Marcus and MacKuen (1993) show (p.322)

Testing Some Implications of Affective Intelligence Theory at the Aggregate Level


(p.323) that voters who report relatively high levels of anxiety about a presidential candidate tend to rely more heavily on issues than on the candidates' partisanship. That is, they make more carefully considered judgments. The authors' most recent statement goes further and states, strongly, that increased anxiety3 induces rational decision making. It tells individual that they are entering a situation of uncertainty, which leads them to rely on learning of alternatives and consideration of available choices as depicted by the rational choice approach (see chapter 6 in this volume).

The assumption that high anxiety leads to rational decision making suggests two additional propositions that predict how rationally acting voters should behave, collectively, in the face of threats to core desires. They are as follows:

The Prospective Evaluation Proposition

  • Prediction: Aggregate voting patterns, as responses to threats to core desires, depend on predictions of future conditions.

  • Rationale: Highly anxious and thus rationally calculating voters successfully predict future conditions at election time (Downs 1957; MacKuen, Erikson, and Stimson 1992).

The Party Reputation Proposition

  • Prediction: Aggregate voting patterns, as responses to threats to core desires, depend on the two parties' reputations in a policy domain.

  • Rationale: If the party with the stronger reputation in a policy domain holds office when a threat in that domain arises, it benefits from that reputation because voters do not see a good alternative.

An Aggregate-Level Test of Affective Intelligence

We analyze the relationship between real-world conditions in three policy domains and voting returns, all measured at the county level.4 In between (p.324) the aggregated conditions and voting returns are assumptions about unobserved individual-level mental processes, assumptions that we take initially from affective intelligence theory. The causal chain is as follows:

Rise of

Increase in

Increase in





of normal vote



The chain's complexity is even greater than this, for there are also aggregation processes: from individual perceptions of threat to collective perceived threat, from individual anxiety to collective anxiety, and from individual votes to collective votes. Luskin (2002a, 2002b; also see MacKuen 2002) has documented the complexities of such aggregation processes. We simplified them by assuming that each of the collective relationships represented by the symbol → maps as a monotonic function. In support of this assumption, we do not expect worsening conditions to reduce collective perceived threat; nor do we expect collective anxiety to decline when collective perceived threat increases or deviations from the normal vote to fall as collective anxiety rises.


To test the propositions, we used data about voting in presidential elections that derived from a comprehensive data archive. The primary unit of analysis in this data archive is the local electorate (counties and major cities).5 The data archive includes election returns for all presidential contests from 1828 to 2000. Unfortunately, the unavailability of some variables restricted this analysis to elections between 1976 and 2000 (about 21,903 cases are available for analysis). We tested only the magnitude, confluence, prospective evaluation, and party reputation propositions; work in progress will address the endurance proposition.


Deviation from the incumbent party's normal vote (incumbentDEV) serves as the dependent variable. It compares the actual presidential vote in any given election to an estimated vote based on the assumption that citizens' habitual voting patterns (voting Democratic, voting Republican, abstaining) (p.325) solely determine their choices. We scaled incumbentDEV so that positive values represent greater-than-expected returns for the incumbent party.

We identified three core desires: physical safety, economic security, and safety from external harm. Five primary variables—current rate and one-year change in crime rate, current rate and two-year change in unemployment rate, and years of international peace—measure threats to the core desires. The national inflation rate and party of the incumbent serve as control variables.

“Years of international peace,” by definition, is a national-level variable.6 Although crime and unemployment rate data exist at all three levels—nation, state, and county—we selected the local-level measures. For one thing, one-year crime and two-year unemployment rates vary considerably across space and time at the local level. When the rates soar, the local media cover the problems extensively. Moreover, bad local conditions more directly threaten citizens' core political desires than do national ones. This is particularly true with respect to crime rates. Crime in local communities and neighborhoods poses a direct and immediate threat to residents' physical well-being; when it increases, people presumably take notice and become anxious. Finally, the local measures offer a good test of the confluence thesis, which posits that overlapping threats will create the highest levels of anxiety among the largest number of voters.

Using the local-level measures entailed making tradeoffs, not the least of which is a reduction in the number of years available for analysis. The U.S. Bureau of Labor Statistics began to publish local monthly unemployment rates in 1974, which is two years earlier than the Federal Bureau of Investigation began to publish local crime rates. National crime and unemployment figures go back much further. Moreover, the county-level data are noisy, a problem that arises because minor fluctuations in the number of jobs or crimes within small counties can cause deceptively large proportional increases.7 We tried to identify all such cases and make proper adjustments, but this alone does not fully overcome the problem. Removing these cases from the analysis does not change the conclusions.

(p.326) Moving from proposition to empirical test requires proper specification of the statistical models. For purposes of testing the magnitude proposition, we included the primary variables introduced above: local crime rate, change in local crime rate, local unemployment rate, change in local unemployment rate, and years of peace. We also included two interactions—local crime rate * change in local crime rate and local unemployment rate * change in local unemployment rate. These interactions allow for the possibility that changes in crime and unemployment rates most strongly affect normal votes when current rates are high. We hypothesize that the main and interactive terms will all linearly and positively affect incumbentDEV (see column 4 in table 13.1).

We added a series of interaction terms to test the confluence proposition. These include three two-way interactions: change in local crime rate * change in local unemployment rate, change in local crime rate * years of peace, and change in local unemployment rate * years of peace. Each interaction accounts for the possible compounding effects of two increasingly bad conditions. To be complete, we also included three- and four-way interactions that include, for crime and unemployment, both the rate and the change in rate (see table 13.1).

Testing the prospective evaluation proposition entailed adding a measure that captures the change in local unemployment rates for the three-month period immediately following the election (change in local unemployment rate3-MONTH FORWARD LAG). A statistically significant relationship between it and incumbentDEV suggests that voters accurately project economic trends affecting their economic well-being and, more generally, that they make sophisticated judgments of political stewardship.

In evaluating the party reputation proposition, we take advantage of the fact that the two major U.S. parties enjoy different performance reputations in two of the domains examined in this analysis: crime and unemployment. Republicans have gained a reputation as a law-and-order party, and Democrats have not. At least since the New Deal, Democrats have held a reputation as supporters of unions and the working class and thus as proponents of full employment. In contrast, most observers see Republicans as generally favoring low interest rates rather than low unemployment rates.

Suppose crime rates jump while Republicans hold office. They should absorb fewer electoral losses than Democrats would under the same conditions. Rationally thinking voters will not expect the Democratic challenger to give higher priority to the crime problem than the Republican incumbent did. Reputation effects with respect to unemployment should work similarly, although, of course, they should favor Democrats. (p.327)

TABLE 13.1: Summary of measures and predicted effects on deviations from the normal vote


Predicted effect

Magnitude effects:

  Local crime rate

  Local unemployment rate

  Δ local crime rate

  Δ local unemployment rate

  Local crime rate × Δ local crime rate

  Local unemployment rate × Δ local unemployment rate

  Years without war

Confluence effects:

  Δ local crime rate × Δ local unemployment rate

  Δ local crime rate × years without peace

  Δ local unemployment rate × years without peace

  Local crime rate × Δ local crime rate × local unemployment rate × Δ local unemployment rate

  Local crime rate × Δ local crime rate × years without peace

  Local unemployment rate × Δ local unemployment rate × years without peace

Rationality effects:

  Δ unemployment rate three months after an election

  Δ local crime rate × Republican incumbent


  Δ local unemployment rate × Republican incumbent

  Local crime rate × Δ local crime rate × Republican incumbent


  Local unemployment rate × Δ local unemployment rate × Republican incumbent

Note: All measures are coded in the same direction, so that each measure is expected to be negatively associated with deviations from the normal vote. The two exceptions are the crime × Republican incumbent interactions.

We tested the party reputation hypothesis with two two-way interaction terms (change in local crime rate * Republican incumbent and change in local unemployment rate * Republican incumbent) and two three-way interaction terms—local crime rate * change in local crime rate * Republican incumbent and local unemployment rate * change in local unemployment rate * Republican incumbent.8 (p.328)

TABLE 13.2: Tests of propositions


Model 1

Model 2

Model 3





National inflation rate




Republican incumbent




Local crime rate




Local unemployment rate




Δ Local crime rate




Δ Local unemployment rate




Local unemployment rate × Δ local crime rate




Local crime rate × Δ local unemployment rate




Years without war




Δ local crime rate × Δ local unemployment rate



Δ local crime rate × years without peace



Δ local unemployment rate × years without peace



Local crime rate × Δ local crime rate × local unemployment rate × Δ local unemployment rate



Local crime rate × Δ local crime rate × years without peace



Local unemployment rate × Δ local unemployment rate × years without peace



Δ Unemployment rate three months after an election


Δ Local crime rate × Republican incumbent


Δ Local unemployment rate × Republican incumbent


Local crime rate × Δ local crime rate × Republican incumbent


Local unemployment rate × Δ local unemployment rate × Republican incumbent






Adjusted R2




(*) p〈.0.5;

(**) p〈.01;

(***) p〈.001.


To facilitate the discussion, we have presented a progression of multiple regression analyses. Table 13.2 reports the results for each model. The first model tests only the magnitude proposition, the second adds terms (p.329) to test the confluence hypothesis, and the third adds yet more terms to test the rationality propositions.

The first column in table 13.2 reports the findings for Model 1. Magnitude effects explain about one-third of the variance in incumbentDEV (adjusted R2 = .31). Although some of the estimated coefficients confirmed the proposition that locally based threats shape people's evaluations of incumbents, others did not. Changes in local crime and unemployment rates worked as hypothesized; increases in either one caused the incumbent to receive less electoral support than expected. On the other hand, existing rates had the opposite effects; the higher the crime rate during the year preceding the election or, in the case of unemployment, the two years before the election, the better the incumbent fared at the polls. Moreover, neither of the interactions between rate and change in rate significantly affected aggregate vote outcomes. As expected, the number of years without peace positively and strongly affected incumbentDEV. These first aggregate results, then, offer mixed support for the first proposition derived from affective intelligence theory's central idea that increased anxiety arising from threatening situations leads people to monitor their environments and rethink their normal partisan choices.

The confluence proposition predicts that the confluence of threatening events evokes emotional reactions from larger numbers of people and thus causes greater aggregate vote changes than any one isolated event. To test it (which, to our knowledge, has not been done before), we added interaction terms that measure the effects of simultaneous conditions in any two or all three domains. The estimated parameters in Model 2 tell a compelling story: bad conditions in two or three domains consistently cause greater deviations from the normal vote than bad conditions in one domain.

The third column in table 13.2 reports the statistical estimates of Model 3, which adds variables to test the two rationality propositions. The prospective evaluation proposition predicts that voters will accurately assess future conditions and use those assessments when choosing between the incumbent president and a challenger. Earlier, we used the twenty-four-month change in local unemployment rate to test the magnitude proposition. For Model 3 we used the rate for the three months immediately following the election. Economic prospects appear to cause deviations from habitual voting patterns; improved unemployment conditions three months after the election help the incumbent, and declining ones hurt him. The party reputation proposition predicts that threatening crime conditions will hurt incumbent Democrats more than incumbent Republicans, whereas threatening unemployment conditions will hurt (p.330) incumbent Republicans more. The last four terms in Model 3 all support the proposition.

Consider crime. Increases in local crime rates adversely affect all incumbents' electoral fortunes, but, as predicted, they negatively affect Republican incumbents' electoral margins less than Democrats'. This holds true in general and also when prevailing crime rates are already high. When existing crime conditions look bleak and appear to be getting bleaker, rationally calculating voters conclude that replacing a Republican incumbent with a Democrat will not improve the situation. Behavior in the unemployment domain looks similar, although Democrats benefit. Bad performance on unemployment does not adversely affect Democratic incumbents' electoral fortunes as much as it affects Republican incumbents'.

Overall, then, the estimates support both of the rationality propositions and suggest that the increased anxiety evoked by bad or worsening conditions does not produce irrational, unthinking reactions. To the contrary, and in line with affective intelligence theory, emotions and rational evaluation appear to go hand-in-hand. At the least, the aggregate patterns support the implications that we correctly or incorrectly derived from the theory.

Affective Intelligence Theory and the Study of Politics

We began with a particular conception of democratic governance, assumed anxiety to be the key mental mechanism, and then derived some implications. The operative term, again, is assumed; our data did not allow us to identify levels of anxiety among individual voters. The findings supported most but not all of the propositions we tested.9 Score a point for affective intelligence theory.

The theory's central notion—that heightened anxiety increases attentiveness to one's external environment and thus causes more considered evaluation of that environment—has justifiably gained considerable status and acceptance in the discipline. Marcus and colleagues have documented the empirical veracity of this notion on numerous occasions and in a variety of ways. In formulating our own propositions, however, we uncovered some limits of affective intelligence theory, at least as we understand it. Most crucially and perhaps ironically, the theory focuses on (p.331) change (in environmental conditions, in levels of anxiety, in behavior) yet offers little guidance for dynamic conceptions of politics. It does not predict how anxiety will ebb and flow as a function of changes in real-world conditions, nor does it predict how political behavior will change as conditions change, except in the most general way.

Take, as a concrete illustration, the endurance proposition (which we did not test): the longer a threat exists, the more widespread anxiety will become. It assumes that a persistently bad condition will create anxiety among people who did not become anxious at the condition's outset. Alternatively, such a condition might reduce anxiety over time because more and more people come to accept it as part of their lives. In its current form, affective intelligence theory offers no help in choosing between these two very different predictions.

In short, we derived our propositions with minimal guidance from affective intelligence theory. We cannot say without qualification, therefore, that the predictions represent proper extensions of the theory. They make sense to us, but that alone does not justify them.

Perhaps we ask too much of affective intelligence theory. Marcus and colleagues build effectively on the study of emotions in psychology and neuroscience, but this research remains in considerable flux. In their companion chapters, Spezio and Adolphs (chapter 4 in this volume) and Huddy, Feldman, and Cassese (chapter 9 in this volume) document and summarize the competing theories and rapidly changing empirical results that characterize both fields. Most fundamentally, this research lacks the requisite theoretic foundations from which to make precise predictions about the dynamics of environment, anxiety, and behavior.

In fact, some of the current literature challenges the exalted status that Marcus and colleagues accord anxiety in their affective intelligence theory, raising two questions for this chapter. First, was anxiety necessary to derive our predictions? Would substituting fear or anger, for example, have generated similar predictions? Second, if anxiety is not necessary, is there reason nevertheless to give it top billing? We begin with substitutability.

In one of the first scientific studies of emotions and political choice, Conover and Feldman (1986) coined the effective phrase, “I'm mad as hell and I'm not going to take it anymore.” They show that bad economic conditions make people angry, and then they vote incumbents out of office. Suppose we had taken this study as our point of departure for deriving implications. Would we have reached the same implications when substituting anger for anxiety? We think so. Conversely, had Conover and Feldman begun with anxiety rather than anger, they might have derived the same set of hypotheses.

(p.332) Huddy, Feldman, and Cassese (chapter 9; also see Brader and Valentino, chapter 8) do not accept the substitutability thesis, however. Drawing on recent neuroscientific and psychological research that challenges the two-dimensional, positive-versus-negative-affect perspective on which Marcus and colleagues (and thus we) draw, they hypothesize that different negative reactions to real-world events produce different evaluations of those events. Their empirical analysis of a three-wave national panel shows that heightened anger and heightened anxiety similarly increased attention to news about the Iraq war. However, the two negative emotions shaped perceptions of the risk of and support for the Iraq war in different ways. Anger reduced estimates of risk and thus promoted support of the war; anxiety worked in the opposite way. In other words, one negative emotion, anger, led to approach, the other, anxiety, to avoidance.

These findings imply that substituting anger for anxiety should change our predictions. What those changed predictions might be, however, we cannot say. For whereas the Conover and Feldman findings readily transfer to our work, the findings of Huddy et al. do not. We can imagine anger and anxiety differentially affecting voter turnout, with anger increasing it and anxiety decreasing it, but the conception we adopted takes change in the direction of the vote, not change in turnout, as its central focus. This suggests that whether different specific emotions differentially affect evaluations and behavior depends heavily on the task citizens are performing.

Fear is another plausible substitute for anxiety. Evolved from the survival demands of our Stone Age ancestors (Barkow, Cosmides, and Tooby 1992; Damasio 1994, 1999; Hauser 1996; Le Doux 1996; Pinker 1997), it is among the most basic of all emotions, and it pervades human life. Would identifying fear as the triggering emotion in our conception of democratic governance have led to different predictions? Again, we see no reason to believe so.10 But we might reach a different conclusion if the task of interest were turning out to vote.

Let us assume that the three negative emotions—fear, anger, and anxiety—produce similar predictions about aggregate changes in partisan voting patterns. Is there a compelling reason, nevertheless and for our specific purposes, to prefer anxiety as the triggering mechanism? If there is, we do not see it, at least from the perspective we adopted in this chapter Adopting fear or anger would work just as effectively as adopting anxiety. In fact, colloquial and far less precise terms such as getting upset or feeling uptight would serve the purpose as well as any of these terms. (p.333) This does not minimize the importance of the distinctions that neuroscience and political psychology seek to make; it underlines the long distance social scientists have yet to travel to connect mental processes and aggregate political patterns.

Until now the discussion reflects the central premise of this volume: without attention to emotions, political scientists cannot fully understand politics. This premise has focused authors' attention on questions such as, What are emotions? Should scholars construe emotions in terms of a single approach-avoidance continuum or as discrete and distinct entities? How and when do emotions affect political decision making? We, too, have adopted the premise, first by incorporating anxiety into our conception of democratic governance and, second, by asking whether only one emotion, anxiety, explains the reported aggregate patterns.

We conclude by questioning the premise itself: is reference to emotions essential to predicting the aggregate patterns we identified and, more generally, to the study of politics? To answer it requires that scholars not be blinded by momentary disciplinary emphases. In both psychology and political science, the emphasis on emotions has waxed and waned. Prior to the 1960s, the two disciplines, psychology especially, placed emotions (and feelings and motivations) at the center of their theories. As other contributors to this volume know, in the ensuing 25 years researchers lost sight of emotions and emphasized cognition and information processing instead. Psychologists, and thus political scientists, have now rediscovered them. This book reflects that rediscovery.

Suppose we had presented our conception of democratic governance 20 years ago. There would have been no references to emotions, and other scholars would not have questioned their absence. Yet our predictions would have mirrored those we offered above. One plausible conclusion: macro-conceptions of politics, or at the very least, ours, work well without reference to emotions.11 We believe this to be a valid conclusion if taken literally.

It fails to recognize, however, that incorporation of emotions enriched the theoretical foundation of our work. Thinking in terms of emotions led to a fuller explication of the causal chain than we otherwise would have made. Moreover, it fails to acknowledge the value of and the need for continual interplay between individual- and aggregate-level studies. Assessing the relative values of appraisal theory and the somatic marker (p.334) hypothesis (Spezio and Adolphs, chapter 4 in this volume) or the differential effects of anger and anxiety (Huddy, Feldman, and Cassese, chapter 9 in this volume) requires analysis at the level of the individual brain or lower. The crucial debates will and should occur at these levels. Political scientists would be remiss, however, if they did not continually try to determine whether the micro-level findings are consistent with higher-level political patterns. This is one of the most formidable challenges students of public opinion and political psychology face. It is tempting, therefore, for researchers to work solely at one or the other level of analysis. Unfortunately, this singular focus limits the discipline's capacity to understand politics and, in particular, the roles of citizens in it.


(1.) We skirt one key matter in this chapter: determining when an increasingly bad condition becomes a threat. Different people will see the same condition differently; some will interpret it as a threat, others will not. In fact, the complexity is greater than this. People might not see a bad condition as a threat until they feel anxiety. We adopt a frankly loose posture in this chapter and assume that a condition or a change in condition objectively becomes a threat once it reaches a certain threshold, which we do not specify. At that point, some people feel threatened, others do not. As the condition worsens further, additional people feel threatened.

(2.) Affective intelligence theory also identifies a key role for heightened enthusiasm, which occurs under positive conditions. Given the focus on threats to core values, this chapter does not consider the impact of enthusiasm.

(3.) Proponents of affective intelligence still have not precisely defined high anxiety or provided a good empirical measure. We agree with Spezio and Adolphs (chapter 4 in this volume) that, intuitively, very high levels of anxiety should impair performance.

(4.) In terms of level analysis, at least, our study is the polar opposite of that by Redlawsk, Civettini, and Lau (chapter 7 in this volume).

(5.) For more information about the data collected for this project see (Nardulli 2005, appendix I, available at www.pol.uiuc.edu/faculty/nardulliresearch.html).

(6.) We are indebted to Scott Gartner for providing the data necessary to construct the variable for years of peace. He provided us with a data set for the correlates of war that lists the beginning and ending of all major military conflicts in U.S. history. The variable for years of peace ranged from 1 to 15 in our analysis pool.

(7.) For theoretical reasons noted above, we needed to test the interactive effects of these variables. This required multiplying “noisy” first-order terms, which only compounds the problem.

(8.) To expedite interpretation of the results, we changed the coding of the party incumbency in the unemployment interaction terms so that 1 equals Democrat and 0 equals Republican.

(9.) Unfortunately, the large number of cases almost ensures statistical significance of the estimated coefficients. Fortunately, nearly all of the parameters represent substantively meaningful relationships.

(10.) Marcus (2002) himself has argued that fear and anxiety are largely one and the same.

(11.) Note that this conclusion is itself time-bound; if neuroscience and psychology advance such that precise predictions about the effects of different emotions can be made, it might no longer hold.