Документ взят из кэша поисковой машины. Адрес оригинального документа : http://suny.msu.ru/ru/Lavine's%20article.pdf Дата изменения: Tue Aug 24 11:42:41 2010 Дата индексирования: Mon Oct 1 19:48:22 2012 Кодировка:

Experimentation and Personality, 1

Explicating the Black Box through Experimentation: Studies of Individual Differences and Cognitive Processes

Howard Lavine Department of Political Science, SUNY ­ Stony Brook Stony Brook, NY 11794-4392 Howard.Lavine@sunysb.edu Corresponding Author

Milton Lodge Department of Political Science, SUNY ­ Stony Brook Stony Brook, NY 11794-4392 Milton.Lodge@sunysb.edu

James Polichak School of Law, University of Michigan Ann Arbor, MI 48109 jpolicha@umich.edu

Charles Taber Department of Political Science, SUNY ­ Stony Brook Stony Brook, NY 11794-4392 Charles.Taber @sunysb.edu


Experimentation and Personality, 2

Abstract We advocate for an experimental approach to the study of personality and politics. In particular, we propose an "interactionist" model of political behavior in which the cognitive and behavioral effects of dispositional variables are qualified by experimentally-induced contexts. Our operating assumption is that the political effects of personality do not occur in a contextual vacuum, but instead are magnified by the presence of key precipitating or "activating" features of the political environment. We illustrate the approach with four experimental studies of authoritarianism. Results indicate that the effects of authoritarianism depend critically on the presence of situationally-induced threat. More generally, we argue that interactions between personality variables and experimental treatments can lead to valuable insights about when and why personality will make a meaningful contribution to public opinion and political behavior. Finally, we close with a critique of the traditional skepticism toward experimentation in political science, and suggest that external validity is an overrated virtue when the research goal is the development of theory rather than the description of "real world" phenomena.


Experimentation and Personality, 3 1 Introduction Theory and research on the underpinnings of public opinion is dominated by the study of political predispositions. Individual differences in personality, value orientation, ideology, and cognitive ability are central explanatory constructs in the areas of mass belief systems (Feldman 1988), opinion formation and change (Zaller 1992), and electoral behavior (Campbell, Converse, Miller and Stokes 1960). The centrality of such variables reflects both their key role in recent theoretical accounts of the origins of public opinion as well as longstanding empirical regularities. It is also a function of the methodological dominance of survey research in the field of political behavior. In essence, political predispositions present few measurement obstacles, and therefore are readily incorporated into models of political judgment and choice. Moreover, by using representative samples and naturalistic settings, external validity is maximized. However, the approach has two important weaknesses, one methodological and well-known, the other substantive and less so. First, survey data do not provide for strong causal inference, instead yielding conclusions of the type "X and Y are related," or the pseudocausal "Y is predicted by X," rather than "X is the cause of Y." Second, research in social psychology suggests that the effects of individual differences are not consistent across situational contexts (Mischel, 1968). In harmony with the dictum that "every event depends upon the state of the person and at the same time on the environment" (Lewin, 1936), the effects of predispositions are likely to be variable, depending on features of the political context. In this article, our task is to demonstrate the variable effects of individual differences on political judgment and choice. To do this, we advocate for an experimental approach to the study of personality and politics. However, our goal is not merely to affirm that situations matter (and that the experimental method is handy in this regard); rather, we aim to show that the effects of political predispositions are qualified by experimentally-induced variation in political context. In particular, we argue that the experimental approach can be useful in answering two fundamental questions about the influence of political predispositions: (1) when are they likely to influence political judgment and behavior (a


Experimentation and Personality, 4 question of prediction) and (2) why, i.e., through which psychological processes (a question of explanation). As we will show in several experiments, the variable influence of predispositions can be profitably studied by crossing aspects of personality with manipulations of context in experimental designs. Our operating assumption is that the political effects of personality do not occur in a contextual vacuum, but instead are magnified by the presence of key precipitating or "activating" features of the political environment. We therefore expect the relationship between personality and political preference to be a contingent one, and that these contingencies can best be elucidated when they are brought under direct experimental control. Beyond the virtue of enhanced prediction, experimental personality designs stand to deepen political scientists' understanding of the linkages between individual differences and political choice. Consider the relationship between political conservatism and opposition to government policies designed to create racial equality. Despite the robustness of this relationship, its meaning is far from clear. For example, does it reflect principled opposition to government intervention in the economy, as Sniderman and Carmines (1997) argue, or racial prejudice, as others (e.g., Sidinius, Pratto and Bobo 1996) have argued? That is, do conservatives oppose such legislation because they believe in individual rather than government responsibility or because they tend to hold negative beliefs about blacks and thus oppose policies designed to help them? Until we develop more powerful methods to pry open the black box, our models of political behavior will remain impoverished at the level of explanation. We do not claim that multivariate survey-based approaches are ill-suited to explicating the relations between individual differences and political judgments and preferences (for excellent examples, see Feldman and Stenner 1997; Sidanius et al. 1996; Sniderman, Brody and Tetlock 1991); what we do contend, and what we hope to demonstrate in this article, is that experiments can provide a powerful methodological tool for addressing this type of inferential problem. In the next section, we introduce more formally the experimental approach to the study of personality and politics. Then, in the bulk of the article, we illustrate the approach with several


Experimentation and Personality, 5 experimental studies of authoritarianism and threat. We then reverse figure and ground by demonstrating how interactions involving predispositions and experimental treatments can shed light on the nature and meaning of the situational or contextual determinants of political reasoning. Finally, we close with a critique of the traditional skepticism toward experimentation in political science, and suggest that external validity is an overrated virtue when the research goal is the development of theory rather than the description of phenomena.

2 An Experimental Approach to the Study of Personality and Politics In our view, political judgment and behavior arise out of the joint (nonadditive) influence of longstanding political predispositions and exigencies of the immediate political context. The core feature of this perspective is that the influence of context on political behavior is not uniform in the electorate. In particular, features of political contexts can be identified in which political predispositions will be predictive of behavior and instances in which they won't. By bringing these environmental influences under direct experimental control, political scientists stand to gain valuable insights into when and why personality will make a meaningful contribution to public opinion and political behavior. In other words, by observing who takes the experimental bait and who doesn't, we can piece together why individual difference effects in politics emerge. But how does this occur? What features of a situation shift the causal locus of behavior to political predispositions, and to which predispositions in particular? Moreover, how can this "interactionist" approach shed light on the origins, nature, and dynamics of personality effects in politics? In our view, political behavior should be predictable (and explainable) in terms of personality when features of the situation match the content associated with a particular dimension of personality. Feature or "template" matching (see Bem and Funder 1978; Funder 1982) occurs when situational forces activate corresponding personality dispositions from memory, thus rendering them temporarily salient or "cognitively accessible." In turn, when the cognitive-behavioral propensities associated with a particular personality style are memorially active ­ when their contents are


Experimentation and Personality, 6 transferred from long-term memory and deposited into working memory ­ they should exert a disproportionate effect on political judgment and decision-making (e.g., Higgins and King 1981; Huckfeldt, Levine, Morgan and Sprague 1999; Iyengar and Kinder 1986; Lavine 2001; Zaller and Feldman 1992). In other words, precipitating situations make personality relevant and thereby strengthen the connections between predispositions and their affective, cognitive, and behavioral manifestations. In the absence of trait-situational feature matching, such predispositions are likely to remain cognitively inactive and therefore relatively unlikely to influence subsequent judgments and behaviors. To illustrate the feature matching approach within an experimental context, consider the following examples, one from the realm of race and politics, and other from the literature on priming and electoral judgment. Within the realm of racial politics, Sniderman and Carmines (1997) challenge longheld wisdom about the determinants of white Americans' racial policy attitudes by embedding a series of randomized experiments within national public opinion surveys. The authors capitalize on the strong inference logic of experimentation to gauge, among other things, the hold of racial prejudice and concerns about social desirability on public opinion. For example, in their "List Experiment," the authors attempt to gain explanatory purchase on ideological differences in attitudes toward affirmative action by varying the format by which respondents report their policy preferences. In the overt response condition, policy attitudes are assessed in the usual manner, and unsurprisingly, liberals are substantially more likely than conservatives to express support for affirmative action. To examine the role of social desirability (among liberals) in producing this ideological effect, Sniderman and Carmines devised a clever covert strategy for assessing respondents' policy attitudes without their knowledge. To do this, one half of the respondents were given a list of three issues (tax increases on gasoline, the high salaries of professional athletes, and corporate environmental pollution), and are asked to list how many of them ­ but importantly, not which ones ­ make them angry. The remaining respondents were given the same instructions and the same list, but with an additional item: "black leaders asking the government for affirmative action." Because there is no way for the interviewer to know which of the items on the list


Experimentation and Personality, 7 engendered the respondent's anger (unless of course all of them did so), there should be little compunction toward masking any anger that respondents might feel toward affirmative action. However, by subtracting the mean in the three item condition from the mean in the four item condition, the proportion of respondents who are angry about affirmative action ­ presumably untainted by social desirability bias ­ can be readily ascertained. The experimental (overt-covert) response format produced a significant effect on policy attitudes, but only for liberals: On the overt measure, 32.7 percent of liberals and 50.9 percent of conservatives expressed anger about affirmative action. On the covert measure, levels of anger among conservatives increased only slightly, to 59.1 percent; in contrast, levels of anger among liberals shot up to 55.8 percent. Thus, the liberal-conservative policy attitude difference seen on the overt measure is eliminated almost entirely when attitudes are assessed covertly, providing some suggestion that ideological differences in attitudes toward affirmative action are more apparent than real (i.e., produced largely by social desirability pressures among liberals). Several recent experiments in the realm of priming exemplify the feature-matching aspect of the interactional approach (e.g., Iyengar and Kinder 1987; Valentino, Hutchings and White 2002; for a quasiexperimental demonstration, see Krosnick and Kinder 1990). As a class, these studies demonstrate that the influence of a given dimension of personality (or attitude or belief) on political judgment depends on the extent to which that dimension has been made situationally salient. In psychological terms, by cognitively highlighting a particular feature of the environment, experimental manipulations of political context can alter the judgmental influence of individual differences. For example, in their studies of priming and presidential performance, Iyengar and Kinder (1987) experimentally varied whether respondents were exposed to several TV news stories about a particular issue (e.g., defense, inflation, unemployment). Consistent with the priming hypothesis, they found that the impact of ratings of the president's performance on specific issues on overall presidential performance was substantially stronger when the issue had been primed in the news stories. Thus, respondents who had been primed by defense stories were more likely to rate the president's overall performance on the basis of his perceived


Experimentation and Personality, 8 performance on defense than were respondents who were primed by news stories about inflation or unemployment, and vice-versa. The foregoing examples are intended to illustrate the value of the experimental method in elucidating personality effects in political judgment. What we see is that the links between individual differences on one hand and judgment and choice on the other depend critically on environmental context. In the next section, we illustrate more extensively the experimental approach to the study of personality and politics by reporting on a program of research on the personality trait of authoritarianism, and examining how its effects on a variety of attitudinal judgments are conditioned by the presence of a key situational instigator: threat.

3 Authoritarianism and Threat Authoritarianism is among the most widely invoked dispositional concepts in the social sciences. Although originally conceived to explicate the psychological roots of prejudice and intolerance (e.g,. Adorno, Frankel-Brunswik, Levinson and Sanford 1950), it has emerged in more contemporary research as a robust predictor of policy attitudes and voting behavior (Altemeyer 1988; Peterson, Doty and Winter 1993). However, despite a vast empirical literature, authoritarianism has remained a theoretically impoverished concept. Aside from the notion that authoritarians are highly conventional, submissive to legitimate authority figures, and aggressive toward socially sanctioned outgroups, scholars disagree about what authoritarianism is, what its origins are, and what dynamics are involved in its expression. Our aim here is not to stake out new theoretical ground but rather to illustrate the value of the experimental approach in highlighting the relevance of the concept of threat in the activation of authoritarian dispositions. Although threat is routinely implicated as an instigator of authoritarianism, the empirical evidence to date is sparse. At the aggregate level, archival studies have shown that citizens exhibit heightened authoritarian attitudes and behavior (e.g., a concern for power, authoritarian aggression,


Experimentation and Personality, 9 submission, cynicism, superstition) during periods marked by social, economic, and political threat (e.g., high unemployment, high crime, civil disorder, war; Doty, Peterson and Winter 1991; McCann 1997, 1999; Sales 1972, 1973). At the individual level, Altemeyer (1988) has reported substantial correlations between his measure of right-wing authoritarianism and perceptions of the world as a dangerous place. More in line with the interactionist perspective, Feldman and Stenner (1997) argued that the punitive and intolerant dispositions of authoritarians should be activated by the presence of threat, and that the connection between authoritarianism and political preference should therefore be strongest when threat is high. This is just what they found: when threat was high ­ in the form of perceived ideological diversity in particular ­ relations between authoritarianism and a variety of political attitudes, including policy issues, attitudes toward ingroups and outgroups, stereotypes about minorities, and political values, were substantially strengthened (see also Greenberg et al 1990).

4 The Present Studies The studies that we report in this article are conceptually similar to the trait-situational matching work of Feldman and Stenner, with two key changes. First, rather than measuring pre-existing levels of perceived threat, we manipulate threat experimentally. This allows us to make more confident inferences about the causal role of threat in activating authoritarian dispositions. Second, rather than examining political attitudes and beliefs congruent with the ideological content of authoritarianism (e.g., intolerance toward nonconformists and minorities), our work encompasses on a more diverse set of psychological reactions plausibly linked to authoritarianism, including heightened automatic cognitive recognition of threatening stimuli, biased or "motivated" information seeking, and biased evaluation of persuasive message arguments. By exerting direct experimental control over the presence and type of threat, and by examining a diverse set of dependent variables, we hope to achieve a deeper understanding of what authoritarianism is and how it works. Our first two experiments examine whether the hypothesized sensitivity to threat among authoritarians can be detected on an unobtrusive, nonreactive information


Experimentation and Personality, 10 processing task (i.e., response latency) that bypasses "controlled" or intentional cognitive processing. Experiment 3 examines whether authoritarianism is associated with biased information seeking in the presence but not in the absence of threat. Finally, in Experiment 4, we examine whether authoritarianism is associated with a preference for threat- rather than reward-based persuasive messages. Given the variety of threats to which authoritarians appear to be sensitized, we do not advocate for any particular type of threat ­ whether political, economic, personal, or something else ­ as the primary activator of authoritarian predispositions. Instead, to increase the generalizability of our findings, we manipulate threat in different ways across the experiments.

4.1 Study 1: Threat, Authoritarianism, and Automatic Cognitive Responses In this section, we explore the possibility that the origins of authoritarianism lie in a general ­ and uncontrollable ­ sensitivity and reactivity to threats to the self, whether those threats emanate from political (e.g., threat to one's cultural values) or personal (e.g., natural disasters, car accidents, AIDS) events. In particular, we examine whether a generalized perception of threat can be detected on an unobtrusive, nonreactive task that bypasses "controlled" or intentional cognitive processing. To do this we employed a cognitive methodology to directly assess whether representations of threatening concepts (e.g., reading the word "mugger" on a computer screen) are automatically activated (i.e., without intention, effort, or control, see Wegner and Bargh 1998) in the cognitive systems of high authoritarians. Participants responded to a set of noun words normed for level of threat on the basis of a pretest (e.g., "mugger," "cancer," "crime," "poison," "telescope," "poetry," "potato"). Specifically, participants determined as quickly as possible (by pressing one of two keys on the keyboard) whether each letter string presented on a computer screen represented a legal English word or a pronounceable nonword (e.g., "shrac"). Theoretically, automatic word recognition ­ the speed with which respondents are able to identify words ­ is an index of the baseline accessibility of the corresponding concepts (Meyer and Schvaneveldt 1971).1 Thus, in our studies, individual differences in the speed of lexical access for high


Experimentation and Personality, 11 threat words reflect differences in the accessibility of or sensitivity to threatening concepts. We expected that reaction times would depend jointly on threat and authoritarianism such that high authoritarians would respond more quickly to threatening but not nonthreatening words.

Participants and Procedure Ninety four undergraduate native speakers of English (n=49 men; n=45 women) at the State University of New York at Stony Brook participated in the study for extra course credit.2 Upon arrival at the lab, participants completed a lexical decision task on the computer. One hundred eighty four test trials were preceded by 20 practice trials. Half of the test trials consisted of legal English words and half consisted of nonword letter strings. Participants were instructed to respond to each target letter string "as quickly as possible without making too many errors." After completing the lexical decision component of the study, participants completed a survey that included Altemeyer's (1988) 30-item Right-Wing Authoritarianism Scale (=.89). Identification of Threatening and Nonthreatening Words. Prior to the study, one hundred five separate SUNY ­ Stony Brook undergraduates participated in a pretest in which they rated the degree to which 92 noun words possessed "dangerous" and "useful" attributes for human survival. Ratings were made on an eight point scale where 1="not at all dangerous to (useful for) human survival" to 8="extremely dangerous to (useful for) human survival." From this word list, ten words (cancer, snake, mugger, plague, crime, collision, lava, tobacco, quicksand, and poison) with mean dangerous ratings above 5.0 (M=6.85) and mean usefulness ratings below 3.0 (M=2.00) were chosen for the threatening word condition. Ten words (potato, telescope, tree, shoe, leaf, clothes, wood, wool, canteen, and ointment) with mean usefulness ratings above 5.0 (M=6.29) and mean dangerousness ratings below 3.0 (M=1.86) were chosen for the nonthreatening word condition. To create reaction time (RT) scores for each participant for each of the 20 words, we first averaged each participant's 10 RTs within each of the two within-subject cells (i.e., threatening and


Experimentation and Personality, 12 nonthreatening word conditions). Then, we subtracted from these means the mean RT for the 82 remaining words not appearing in the given condition. Negative numbers thus indicate quicker (i.e., more accessible) responses. This procedure controls for the extraneous effects of individual differences in overall speed of RT by creating scores that represent the extent to which the condition RTs for each participant represent a "fast" or a "slow" mean response time relative to each participant's overall RT from the 92 word trials (see Fazio, 1990; Lavine, 1997).

Results To examine whether high authoritarians responded more quickly than low authoritarians to the threatening (but not the nonthreatening) words, we performed a tertile split on RWA scores (i.e., we used the top and bottom thirds of the authoritarianism distribution) and performed a 2 (authoritarianism: low vs. high) x 2 (word type: threatening vs. nonthreatening) mixed effects analysis of variance with repeated measures on the second factor. Our hypothesis translates into an interaction between authoritarianism and word type such that high authoritarians are predicted to respond more quickly to the threatening but not the nonthreatening words. This interaction was the only effect to achieve significance, F(1, 59) = 3.32, p < .05 (main effect Fs<1).3 Follow-up contrasts revealed that high authoritarians responded marginally more quickly to the threatening words than did their low authoritarian counterparts, t(59) = 1.49, p < .10 (one-tailed). Interestingly, this pattern was reversed for the nonthreatening words (i.e., low authoritarians responded faster), t(59) = 1.59, p < .10 (see Table 1). Insert Table 1 here These results indicate that high authoritarians' sensitivity to threat can be observed on a subtle task that measures automatic responses that are generally thought to preclude conscious control. In our next experiment, using an automatic priming task, we attempt to provide further support for the idea that threat selectively activates the responses of high authoritarians.


Experimentation and Personality, 13 4.2 Study 2: Priming Threat If high authoritarians are chronically more sensitive to threat than their low authoritarian brethren, priming threatening concepts should facilitate lexical responses to semantically related target concepts to a greater extent among high than low authoritarians. For example, priming the word "arms" should provide stronger activation of the concepts "weapons," "guns" and "war" among high than low authoritarians. In Study 2, we used a priming paradigm to further explore whether threat-related concepts are especially cognitively accessible in the belief systems of high authoritarians.

Participants and Procedure Ninety one undergraduate native speakers of English (n=31 men; n=59 women; 1 subject failed to respond to the gender item) at the State University of New York at Stony Brook completed 144 trials of a prime-target lexical decision experiment.4 Half of the target words consisted of nonword letter strings. Forty eight of the remaining 72 trials consisted of legal English word prime-target pairings in which the primes consisted of homographs (i.e., words with multiple meanings) with threatening and neural meanings (e.g., "arms," "beat"). Participants were instructed to attend to both the prime and the target words, but to respond ("as quickly as possible without making too many errors") only to the target word in each pair (in terms of whether or not the letter string is a legal English word). Two characteristics of the pairing were manipulated: (1) The target words were either semantically related or unrelated to the primes ("arms-weapons" vs. "arms-book"), and (2) the target words were related either to the threatening or the neutral connotations of the primes (e.g., "arms-weapons" vs. "arms-legs"). If high authoritarians are more sensitized to threat than are low authoritarians, priming threat-related concepts should facilitate responses to semantically related target words to a greater extent among high than low authoritarians when the targets are linked to the threatening but not the neutral connotation of the prime. Thus, priming effects should be stronger among high than low authoritarians for the pairing "armsweapons" but not "arms-legs." This translates into a two-way interaction between authoritarianism and


Experimentation and Personality, 14 type of prime-target relatedness (threat-related vs. neutral-related prime-target conditions; threatunrelated and neutral-unrelated conditions excluded). Moreover, authoritarians should respond more quickly to target words when they are semantically related to the threatening connotations of the primes but not when they are unrelated to the threatening connotations of the primes. Thus, priming effects should be stronger among high than low authoritarians when the target word "weapons" is preceded by the prime word "arms" (threat-related condition) but not when "weapons" is preceded by the prime word "beat" (threat-unrelated condition). This translates into a two-way interaction between authoritarianism and prime-target relatedness (threat-related vs. threat-unrelated prime-target conditions; neutral-related and neutral-unrelated conditions excluded). To create RT scores for each participant for each of the four within-subject conditions (threatrelated, threat-unrelated, neutral-related, neural-unrelated), we first averaged each participant's RTs from the 12 prime-target pairs comprising each condition. Then, we subtracted from each of these means the mean RT for the 36 prime-target pairs comprising the three other conditions. For example, final RT scores for the threat-related condition (e.g., "arms-weapons") were created by subtracting each participant's mean RT for the 36 pairs comprising the three additional within-subject conditions from the participant's mean RT score in the threat-related condition. Negative numbers thus indicate quicker (i.e., more accessible) responses. As in Study 1, this procedure eliminates the extraneous effects of individual differences in overall speed of responding.

Results To evaluate whether the effects of priming were moderated by authoritarianism, we performed two focused analyses of variance. First, to determine whether greater response facilitation occurred among high than low authoritarians for threat-related (e.g., "arms-weapons") but not neural-related ("arms-legs") prime-target pairs, we performed a 2 (authoritarianism: low vs. high [based on a tertile split]) x 2 (type of prime-target pair: threat-related vs. neutral-related) mixed effects ANOVA with


Experimentation and Personality, 15 repeated measures on the second factor. Only the expected interaction of authoritarianism and type of prime-target pair approached significance, F(1, 54) = 2.26, p = .07 (main effect Fs<1). As can be seen in Table 2, high authoritarians responded more quickly than low authoritarians to threat-related (t[54] = 2.31, p < .05) but not neutral-related (t[54] = -1.52, ns.) prime-target pairs. Second, to determine whether greater response facilitation occurred among high than low authoritarians for threat-related (e.g., "arms-weapons") but not threat-unrelated ("beat-weapons") primetarget pairs, we performed a 2 (authoritarianism: low vs. high) x 2 (type of prime-target pair: threatrelated vs. threat-unrelated) mixed effects ANOVA with repeated measures on the second factor. The analysis produced a main effect for authoritarianism, F(1,54) = 3.98, p = .05; across priming conditions, high authoritarians responded more quickly (M = 6.44 msecs) than low authoritarians (M = 23.84 msecs). Although the authoritarianism x type of prime-target pair interaction did not reach significance (F[1, 54)] = 1.46, ns.), the effect of authoritarianism was significant for threat-related primes (t[54] = 2.63, p < .01) but not for the threat-unrelated primes (t[54] = ­0.19, ns.; see columns 1 and 3 of Table 2). Insert Table 2 here Study 2 provides corroborating evidence that authoritarian individuals are especially sensitive to threat, and that this sensitivity can be primed on an automatic cognitive processing task. Specifically, when primed with threatening concepts, high authoritarians responded more quickly to semantically related concepts, but only if the related concepts (the target words) were semantically linked to the t