Covariation Bias in Fear-Stimuli & Aversive Outcomes: An Illusory Correlation Study, Lab Reports of Psychology

Download Covariation Bias in Fear-Stimuli & Aversive Outcomes: An Illusory Correlation Study and more Lab Reports Psychology in PDF only on Docsity! used in Experiment 2 was not yet available to us. For this reason, the Sonalert tone used previously in Experiment 1 was used in Experiment 3. Means and Standard Deviations for Salience Ratings of the Shock and Chime + Light Outcomes in Experiment 2 Mean estimates of the conditional probability of outcomes given slide categories in Experiment 1. Mean estimates of the conditional probability of outcomes given slide categories in Experiment 2. Mean estimates of the conditional probability of outcomes given slide categories in the 33% and 50% base-rate conditions of Experiment 3. Page 24 of 24 9/24/2000http://spider.apa.org/ftdocs/abn/1989/november/abn984381.html or spider group. Supplementary analyses failed to indicate that individual differences in fear of one stimulus (e.g., snakes) were correlated with biased assessment of covariations involving the other stimulus (e.g., spiders). 2 Subjects' knowledge of experimental hypotheses was assessed during the debriefing. In each experiment, those subjects who expressed knowledge of the experimental hypotheses were eliminated from the design. Subjects were judged to have guessed hypotheses if they stated that the purpose of the experiment was to assess whether subjects perceived a relation between fear-relevant stimuli and aversive outcomes even though there was no such relation. Examination of the covariation estimates of the few subjects who were designated as "guessers" indicated, overall, little difference from the estimates of comparable (e.g., same degree of fear) subjects who failed to guess hypotheses. If anything, the covariation judgments of the few guessers indicated less bias than those of subjects kept in the design. This is not a surprising outcome given the criteria used to eliminate subjects (see Tomarken, 1988 , for further details). 3 Details of the results from two additional sections of the questionnaire are provided in Tomarken (1988) . 4 The results of omnibus ANOVAS performed on subjects' covariation estimates in Experiments 1—3 are reported in Tomarken (1988) and are available from the authors on request. 5 Because the decision to have subjects rate emotional reactions to slides was not made until the study was already in progress, only 16 of the 25 subjects made emotion ratings. 6 In this article, the major data-analytic strategy is to conduct those planned analyses most capable of addressing the central, unique questions raised by each study. One inevitable by-product of this strategy is some inconsistency across studies in the particular types of planned analyses conducted. The results of analyses that are more consistent, albeit somewhat less tailored to the unique questions of individual studies, are presented in Tomarken (1988) . These latter analyses are conducted in the general manner reported in Experiment 1. 7 Because Experiment 3 was conducted before Experiment 2, the chime-plus-light compound stimulus Page 23 of 24 9/24/2000http://spider.apa.org/ftdocs/abn/1989/november/abn984381.html estimates. As in the present studies, however, high-fear subjects displayed a significant bias to associate slides of snakes with shock. This finding strongly argues against a demand characteristics interpretation of the present findings. In summary, high-fear subjects in the present experiments consistently demonstrated a bias to associate fear-relevant stimuli with shock. Although generally accurate, low-fear subjects demonstrated similar judgmental distortions under particularly potent contextual conditions. These findings suggest that fear is consistently linked to a style of confirmatory processing that may serve to maintain or enhance fear. Finally, our findings also raise the important question of whether preparedness phenomena are restricted to the domain of classical conditioning or are one manifestation of a more general covariation bias. References Alloy, L. B. & Abramson, L. Y. (1979). Judgment of contingency in depressed and nondepressed students: Sadder but wiser? Journal of Experimental Psychology: General, 108, 441-485. Alloy, L. B. & Tabachnik, N. (1984). Assessment of covariation by humans and animals: The joint influence of prior expectations and current situational information. Psychological Review, 91, 112-149. Arkes, H. R. & Harkness, A. R. (1983). Estimates of contingency between two dichotomous variables. Journal of Experimental Psychology: General, 112, 117-135. Beck, A. & Emery, G. (1985). Anxiety disorders and phobias: A cognitive perspective.(New York: Basic Books) Blaney, P. H. (1986). Affect and memory: A review. Psychological Bulletin, 99, 229-246. Boik, R. J. (1981). A priori tests in repeated measures designs: Effects of nonsphericity. Psychometrika, 46, 241-255. Bower, G. H. (1987). Commentary on mood and memory. Behaviour Research and Therapy, 25, 443- 455. Brewer, W. F. & Nakamura, G. V. (1984). The nature and function of schemas.(In R. S. Wyer & T. K. Srull (Eds.), Handbook of social cognition (Vol. 1, pp. 119—160). Hillsdale, NJ: Erlbaum.) Burgess, I. S., Jones, L. M., Robertson, S. A., Radcliffe, W. N. & Emerson, E. (1981). The degree of control exerted by phobic and non-phobic verbal stimuli over the recognition behavior of phobic and non-phobic subjects Behaviour Research and Therapy, 19, 233-243. Butler, G. & Mathews, A. (1983). Cognitive processes in anxiety. Advances in Behaviour Research and Therapy, 5, 51-62. Chapman, L. J. (1967). Illusory correlation in observational report. Journal of Verbal Learning and Verbal Behavior, 6, 151-155. Chapman, L. J. & Chapman, J. P. (1967). Genesis of popular but erroneous psychodiagnostic observations. Journal of Abnormal Psychology, 72, 193-204. Chapman, L. J. & Chapman, J. P. (1969). Illusory correlation as an obstacle to the use of valid diagnostic signs. Journal of Abnormal Psychology, 74, 271-280. Clark, D. M. & Teasdale, J. D. (1982). Diurnal variation in clinical depression and accessibility of memories of positive and negative experiences. Journal of Abnormal Psychology, 91, 87-95. Clark, W. C. (1974). Pain sensitivity and the report of pain: An introduction to signal detection theory. Anesthesiology, 40, 272-287. Cohen, J. & Cohen, P. (1983). Applied multiple regression/correlation analysis for the behavioral sciences. (Hillsdale, NJ: Erlbaum) Cook, E. W., Hodes, R. L. & Lang, P. J. (1986). Preparedness and phobia: Effects of stimulus content on human visceral conditioning. Journal of Abnormal Psychology, 95, 195-207. Cornelius, R. R. & Averill, J. R. (1983). Sex differences in fear of spiders. Journal of Personality and Social Psychology, 45, 377-383. Dawson, M. E., Schell, A. M. & Banis, H. T. (1986). Greater resistance to extinction of electrodermal Page 20 of 24 9/24/2000http://spider.apa.org/ftdocs/abn/1989/november/abn984381.html need for future illusory correlation studies that concurrently monitor subjects' contingency estimates and their autonomic responses, to further understand the relation between the two sets of phenomena. This line of inquiry would also have broader implications for the more general issue of the relation between covariation assessment and conditioned responding (e.g., Alloy & Tabachnik, 1984 ; Dickinson & Shanks, 1985 ). Additional Conceptual and Methodological Issues As was noted previously, one important question not addressed by these experiments is the nature of those processes that mediate covariation bias. There may well be a variety of processes that mediate covariation bias. One potential mediator noted previously (see the Discussion of Experiment 3) is inappropriate heuristics for integrating frequency information into an overall judgment of contingency. In addition, Hamilton et al. (1985) showed that illusory correlations can be generated by biased encoding of frequency information even when subjects use appropriate heuristics for integrating that information. The significant correlation in Experiment 3 between pain ratings and shock/phobic covariation estimates among low-fear subjects suggests that more basic perceptual processes may also contribute to covariation bias. One important general question here is the degree to which the effects of individual differences in fear are mediated by emotional or motivational states. High- and low-fear subjects failed to differ on several measures of emotional state (e.g., state anxiety or fear of shock). Furthermore, when differences were observed on specific indexes (e.g., fear in response to slides), these measures failed to predict shock/phobic estimates among high-fear subjects and, as the multiple regression analyses indicated, failed to significantly account for the effects of individual differences in fear on covariation estimates. Perhaps among high-fear subjects in particular, then, those factors that mediate covariation bias are relatively state-independent. Interestingly in this regard, the one significant correlation observed between a state variable (pain) and shock/phobic covariation estimates was observed for low-fear subjects alone. We should caution, however, that the present studies were not explicitly designed to assess the mediating effects of state variables and thus may not have optimally tested this hypothesis. For example, self-reports of affective state were typically made after, not during, the illusory correlation trials. Concerning those factors that mediate covariation bias, one additional methodological issue should be addressed. It could be argued that the illusory biases manifest in the present experiments primarily reflect subjects' susceptibility to "demand characteristics." According to this argument, subjects may have adjusted their actual phobic—shock estimates in a direction consistent with the perceived expectations or hypotheses of the investigators. There are several reasons why a demand characteristics interpretation of the present findings lacks credibility (see also Footnote 2 ). First, it fails to account for important features of the data. For instance, it is not clear how such an interpretation can explain both the consistently significant differences between high- and low-fear subjects when shocks occurred 33% of the time and the striking lack of differences between these two groups when shocks occurred 50% of the time. Second, and perhaps more importantly, the findings of a recent study by Sutton, Mineka, and Tomarken (1989 ; see also Sutton, 1987 ) also argue strongly against a demand characteristics interpretation. In this study, one group of subjects was exposed to snake, mushroom, and flower slides, as in the present studies, and a second group was exposed to slides of damaged or exposed electric outlets, mushrooms, and flowers (cf. Hugdahl & Karker, 1981 ). If demand characteristics were a significant factor in the present paradigm, it would be expected that subjects would perceive a strong correlation between slides of the clearly dangerous exposed outlets and shock. Despite this potential for demand that appears to be at least equal to that of snakes and spiders, the outlet—shock contingency estimates of neither high- nor low-fear subjects were significantly different from relevant comparison Page 19 of 24 9/24/2000http://spider.apa.org/ftdocs/abn/1989/november/abn984381.html acting to promote the maintenance or enhancement of fears and phobias over time. Relevance to Selective Associations and Preparedness As was noted in the introduction, according to Seligman's (1971) and Öhman's (1979 ; Öhman et al., 1985 ) preparedness theories, the associations between fear-relevant conditioned stimuli (CSs) and aversive unconditioned stimuli (USs) are selective associations. Selective associations are demonstrated when subjects more easily learn to associate some combinations of conditioned stimuli (CSs) and unconditioned stimuli (USs) than others. Given the laboratory findings that support Seligman's original proposal (for reviews, see McNally, 1987 , and Öhman et al., 1985 ) and the evidence that similar factors underlie conditioned responding and covariation perception (e.g., Alloy & Tabachnik, 1984 ; Dickinson & Shanks, 1985 ), one can ask whether the covariation bias manifest in the present studies also exemplifies a selective association. In certain broad respects, the present results do seem to demonstrate selective associations between fear- relevant stimuli and aversive outcomes, particularly for high-fear subjects. For example, high-fear subjects consistently overestimated the percentage of trials on which feared slides and shock occurred together, but not the percentage of trials on which either occurred (i.e., their base-rate estimates were accurate). This pattern high-lights the importance of the specific association between feared stimuli and aversive outcomes. Furthermore, as a selective association account would predict, in Experiment 2, high-fear subjects overestimated the covariation between feared stimuli and shock but accurately estimated the covariation between feared stimuli and the chime-plus-light outcome that was matched to shock on salience. Unfortunately, there are several important criteria not addressed by the present studies that have been identified as necessary for definitive demonstration of selective associations (for reviews, see LoLordo, 1979 , and Mackintosh, 1983 ). In particular, it has been argued that definitive demonstration of a selective association requires use of what is known in the conditioning literature as a "double dissociation design" ( LoLordo, 1979 ). In the present context, a double dissociation would be demonstrated by results indicating not only relative overestimation of the covariation between feared stimuli and an aversive outcome but also relative underestimation of the covariation between feared stimuli and another hedonically positive outcome (or relative overestimation of the covariation between fear-irrelevant stimuli and another positive outcome). Unfortunately, the present studies were not designed to test for a double dissociation, because no positive outcomes were used. The issue of whether the present results exemplify selective associations raises the question of the relation between the present findings and previous studies demonstrating enhanced autonomic conditioning to fear-relevant stimuli. A particularly important question is whether similar processes underlie enhanced conditioning to fear-relevant stimuli and biased assessment of the covariation between such stimuli and aversive outcomes. Of prime relevance here are the results of several conditioning studies that indicate enhanced responding to fear-relevant CSs even when subjects no longer expect that a shock US will occur ( Hugdahl, 1978 ; Hugdahl & Öhman, 1977 ; Öhman, Erixon, & Löfberg, 1975 ; but see, e.g., Cook et al., 1986 ). Unfortunately, expectancies for aversive outcomes were not actually assessed in those experiments. Furthermore, the results of a recent study by Dawson, Schell, and Banis (1986) suggest the importance of expectancies in mediating prepared conditioning effects. Dawson et al. found that although fear-relevant CSs induced greater electrodermal resistance to extinction than fear-irrelevant CSs, the former were also associated with greater expectancy of US occurrence. The results of Dawson et al. suggest that differential expectancies for US occurrence may underlie the differential conditioning effects found in previous preparedness studies. In turn, these findings suggest that similar processes may underlie the enhanced conditioning to fear-relevant stimuli and biased assessment of the covariation between such stimuli and aversive outcomes. Clearly, there is a Page 18 of 24 9/24/2000http://spider.apa.org/ftdocs/abn/1989/november/abn984381.html significant main effect for slide category, F (2, 196) = 15.74, p < .0001. Post hoc comparisons revealed that subjects underestimated the frequency of occurrence of flower slides ( M = .34) in relation to both snake ( M = .43) and mushroom, M = .40, slides, ps < .01. An ANOVA performed on subjects' estimates of the percentage of times that each of the three outcomes occurred revealed a significant Base Rate × Outcome interaction, F (2, 196) = 23.69, p < .0001. Follow-up analyses indicated a pattern of differences that correspond to the veridical base rates in the 33% and 50% conditions. That is, there were no significant differences among the outcome estimates of subjects in the 33% base-rate condition, F (2, 102) < 1. In contrast, there were marked differences among the estimates of subjects in the 50% condition, F (2, 94) = 32.76, p < .0001, because of significantly higher estimates of the base rate of shock ( M = .53) in relation to tones ( M = .31) or nothing ( M = .32; both ps < .0001). State Measures A subsidiary purpose of the present experiment was to follow up the anecdotal observation from Experiments 1 and 2 that subjects, particularly high-fear subjects, perceived shocks as more painful when they occurred after phobic slides. The results of analyses were generally consistent with this observation. An ANOVA performed on subjects' pain ratings yielded a significant Fear Level × Slide Category × Time (i.e., beginning vs. end of the procedure) triple interaction, F (2, 196) = 3.95, p < .025. Subsequent simple effects analyses and post hoc comparisons revealed that, at the beginning of the procedure, high-fear subjects perceived the shocks occurring after snake slides ( M = 3.74) as significantly more painful than shocks occurring after either mushroom ( M = 3.14) or flower ( M = 3.20) slides, all ps < .001. Additionally, high-fear subjects rated shocks occurring after snake slides as significantly more painful than did low-fear subjects ( M = 3.29; p < .01). There were no significant differences among the low-fear subjects' estimates of how painful the shocks had been at the beginning of the procedure. Both groups, however, rated the shocks occurring after snake slides at the end of the procedure ( M = 2.17) as more painful than shocks occurring after flower slides ( M = 1.81; both ps < .005). An ANOVA conducted on subjects' ratings of the degree to which they were bothered by the shock revealed a significant main effect for time of estimation, F (1, 98) = 4.41, p < .05. Subjects in the delayed estimation condition retrospectively reported being more bothered by the shock than did subjects in the immediate estimation condition. Analyses of subjects' mean fear of shock ratings and their state anxiety ratings revealed no significant effects (all ps > .05). Supplementary multiple regression and correlational analyses assessed whether subjects' pain and state emotion ratings predicted their shock/phobic covariation estimates. The multiple regression analyses included as predictors dummy variables denoting fear level and base-rate condition and were conducted in a hierarchical manner with higher order interaction terms added in subsequent steps. The only significant effect involving state measures yielded by the multiple regression analyses was a significant Fear Level × Pain interaction, F (1, 99) = 2.22, p < .03. This effect was due to the fact that pain ratings predicted low-fear subjects' shock/phobic estimates (zero-order r = .46, p < .001) but not high-fear subjects' estimates ( r = .06). Discussion Base-Rate Effects Individual differences in fear had dissimilar effects in the 33% and 50% base-rate conditions. Consistent with the results of Experiments 1 and 2, in the 33% condition, high-fear subjects markedly overestimated the covariation between feared stimuli and aversive outcomes, but low-fear subjects failed to demonstrate any significant biases. In contrast, in the 50% condition, both high- and low-fear subjects demonstrated significant illusory biases. This conclusion is, however, somewhat qualified by the fact that the Fear Level × Base Rate × Slide Category interaction yielded by an omnibus ANOVA did not Page 15 of 24 9/24/2000http://spider.apa.org/ftdocs/abn/1989/november/abn984381.html All planned analyses compared subjects' shock/phobic estimates only to subjects' estimates of the covariation between neutral slides and shock, because the mushroom—shock and flower—shock pairings were the only two with the same veridical contingency as phobic—shock across all conditions. To control experimentwise error, we evaluated all effects tested by the two within-fear group and two between-fears group ANOVAS at α = .05/2 = .025. To insure adequate control of Type I errors, we adjusted the alpha rates for all follow-up analyses accordingly (e.g., simple effects comparisons were evaluated at .025/2 = .0125). Results Covariation Estimates Figure 3 illustrates high- and low-fear subjects' estimates of the proportion of shocks that followed phobic, mushroom, and flower slides in the 33% and 50% base-rate conditions. Figure 3 indicates that the results in the 33% condition replicate those of the previous two experiments. That is, high-fear subjects show a strong bias to associate feared stimuli with aversive outcomes, whereas low-fear subjects show only a faint hint of bias. The covariation estimates in the 50% base-rate condition, however, offer a striking contrast to the pattern evidenced in the 33% condition and in previous experiments. In this condition, the shock/phobic estimates of high- and low-fear subjects are nearly identical and are greater than relevant comparison estimates. The results of the four sets of planned analyses corroborate these observations. Considering first the within—fear level analyses, the Base Rate × Time of Estimation × Slide Category ANOVA conducted within the low-fear group indicated a significant Slide Category × Base Rate interaction, F (1, 52) = 18.33, p < .01. Follow-up simple effects analyses of this interaction indicated that low-fear subjects' shock/phobic and shock/mushroom—flower estimates failed to differ in the 33% condition ( p > .40) but differed significantly in the 50% condition ( p < .001). In contrast, the complementary ANOVA conducted within the high-fear group revealed only a significant main effect for slide category, F (1, 46) = 37.69, p < .0001. This result indicates that across both base-rate conditions, high-fear subjects' shock/phobic estimates were significantly greater than their shock/mushroom—flower estimates. Concerning the two between—fear levels analyses, the Fear Level × Time of Estimation × Slide Category ANOVA conducted within the 33% base-rate condition indicated a significant effect for the Fear Level × Slide Category interaction, F (1, 51) = 9.11, p < .005. Follow-up simple effects analyses indicated that in the 33% condition, high-fear subjects' shock/phobic estimates were significantly greater than low-fear subjects' shock/phobic estimates ( p < .001). The Fear Level × Time of Estimation × Slide Category ANOVA within the 50% condition revealed a different pattern. In this case, the only significant effect was a main effect for slide category, F (1, 47) = 32.37, p < 001 (Fear Level × Slide Category interaction, F (1, 47) < 1, p > .90). When considered in light of the significant interaction yielded by the 33% condition ANOVA , this significant main effect suggests that low-fear subjects' covariation judgments were significantly more similar to those of the high-fear group in the 50% than in the 33% condition. To test this observation, we conducted a supplementary four-way Fear Level × Base Rate × Time of Estimation × Slide Category ANOVA comparing high- and low-fear subjects' covariation bias across the 33% and 50% conditions. Somewhat surprisingly, this analysis failed to reveal a significant Fear Level × Base Rate × Slide Category three-way interaction, although a near- significant trend was evident, F (1, 98) = 2.91, p = .09. Finally, across all analyses, there were no significant effects involving time of estimation. Base-Rate Estimates A Fear Level × Base Rate × Time of Estimation × Slide Category ANOVA on subjects' base-rate estimates of the percentage of trials on which each of the three slide categories occurred revealed a Page 14 of 24 9/24/2000http://spider.apa.org/ftdocs/abn/1989/november/abn984381.html and bother by, shock during the illusory correlation trials; and a six-item questionnaire asking subjects to indicate on 1 to 5 scales (1 = not at all, 5 = extremely ) how painful they found the shocks that they experienced, both at the beginning and the end of the series of trials. Subjects who had been randomly assigned to the delayed estimation condition were scheduled for two experimental sessions, 1 week apart. When initially contacted on the phone, subjects were told that the first session involved different kinds of stimuli presented in different sensory modalities and that the second experiment involved the completion of several questionnaires and surveys. Subjects were not given any explicit information about the relation between the two sessions over the phone. During the first experimental session, the procedure was identical to that of the immediate estimation condition through the completion of the 72 illusory correlation trials. At that time, instead of completing the probability questionnaire and additional measures (as in the immediate estimation conditions), the subject was requested to return the following week to answer questions about what she had experienced that day. Subjects returned to the second session exactly 1 week after the first session, at the same time of day. After written instructions orienting her to the covariation estimation task, the subject completed the probability questionnaire. Then she completed the additional mood and pain questionnaires described earlier. As with the probability questionnaire, these questionnaires required the subject to recall how she felt during the series of trials in the previous session. 50% base-rate groups. The procedure for subjects in the immediate and delayed estimation 50% groups was identical to that for subjects in the 33% base-rate groups except for differences in the verdical contingencies between slides and outcomes across the 72 illusory correlation trials. For these groups, sequences of slides and outcomes were used so that each of the three slide types was followed by shock 50% of the time and by each of the two neutral outcomes (tone or nothing) 25% of the time. Because shocks occurred with equal probability after each of the three slide categories, it should be emphasized that slide types and outcomes were uncorrelated, as in the 33% base-rate conditions. In addition, as in the 33% base-rate conditions, nine different combinations of slide—outcome sequences were used, and all sequences were randomly assigned to subjects. Data Analysis Consistent with the previous two experiments, planned analyses were used to test the major experimental predictions. In all, four planned ANOVAS were conducted that correspond to the major purposes of Experiment 3 described earlier. First, an ANOVA was conducted within the low-fear group to assess whether the 50% base-rate condition and, secondarily, the delayed estimation condition potentiated illusory biases among low-fear subjects. This analysis compared subjects' shock/phobic estimates to their pooled shock/mushroom and shock/flower estimates. The overall design for the analysis was Base Rate (33% vs. 50%) × Time of Estimation (immediate vs. delayed) × Slide Category (shock/phobic vs. shock/pooled mushroom—flower). This design was also used for a second ANOVA conducted within the high-fear group that assessed the effects of base rate and time of estimation on high-fear subjects' covariation bias. A third purpose of Experiment 3 was to replicate the results of Experiment 2, indicating differences between high- and low-fear subjects in the patterning of covariation bias under 33% base-rate conditions. For this reason, we conducted a Fear Level × Time of Estimation × Slide Category (shock/phobic vs. shock/pooled mushroom—flower) ANOVA within the 33% base-rate condition. The fourth and final planned analysis was a Fear Level × Time of Estimation × Slide Category ANOVA conducted within the 50% condition. Page 13 of 24 9/24/2000http://spider.apa.org/ftdocs/abn/1989/november/abn984381.html Results Salience and Aversiveness of Outcomes As intended, subjects perceived the shock and chime-plus-light outcomes as equally salient. Table 1 illustrates the results for both measures of perceived salience. In support of this conclusion are the results of Fear Level (high/low) × Outcome (chime-plus-light/shock) × Time (beginning of trials/end of trials) ANOVAS conducted on subjects' ratings of the degree to which the two outcomes were noticeable and attention-getting. Only main effects for time were significant, F s (1, 23) = 23.13 (noticeable) and 18.90 (attention-getting), p s < .001. Both outcomes were more salient at the beginning than at the end of the series of trials. As expected, despite matching on salience, subjects perceived the shock as more aversive than the chime-plus-light outcome. A Fear Level × Outcome × Time ANOVA on subjects' ratings of fear of the two outcomes yielded significant main effects for time, F (1, 23) = 33.18, p < .0001, and outcome, F (1, 23) = 43.07, p < .0001, and a significant Time × Outcome interaction, F (1, 23) = 32.80, p < .0001. Subsequent simple effects analyses revealed that subjects reported greater fear of the shock than of the chime-plus-light outcome at both the beginning, F (1, 23) = 56.63, p < .0001, and the end, F (1, 23) = 3.36, p < .025, of the series of trials. The reason for a significant interaction in the presence of two significant, directionally consistent simple effects is the decline in reported fear of the shock over time, F (1, 23) = 41.53, p < .0001 (Time 1 M = 3.47, Time 2 M = 1.80). An additional ANOVA on subjects' ratings of the degree to which they were bothered by the two outcomes yielded significant effects for time, F (1, 23) = 18.51, p < .001, and for the Time × Outcome interaction, F (1, 23) = 9.95, p < .005. Simple effects analyses indicated that the shock bothered subjects significantly more than the chime-plus-light outcome at the beginning of the series of trials, F (1, 23) = 7.13, p < .025, but not at the end, F (1, 23) < 1, primarily because of a decline in the degree to which shock bothered subjects over the course of time, F (1, 23) = 27.70, p < .0001 (Time 1 M = 3.30, Time 2 M = 2.05). Covariation Estimates Despite the fact that the chime-plus-light outcome equaled shock in salience, high-fear subjects still overestimated the covariation between slides of feared stimuli and shock. Indeed, as is shown in Figure 2 , high-fear subjects' shock/phobic estimates were notably greater than all relevant comparison estimates of the probability of outcomes given slide categories. In contrast, low-fear subjects failed to demonstrate a notable tendency to overestimate the covariation between phobic slides and shock in relation to comparison covariation pairings. The results of the two planned interaction contrasts corroborated these observations. Both contrasts revealed significant Fear Level × Slide Category interactions. These effects indicate that, in comparison with low-fear subjects, high-fear subjects estimated that the covariation between phobic slides and shock was greater than the covariation between phobic slides and the chime-plus-light outcome, t (23) = 2.12, p < .05, and was greater than the covariation between the fear-irrelevant mushroom and flower slides and shock, t (23) = 2.32, p < .05. Subsequent simple effects comparisons indicated that high-fear subjects' shock/phobic estimates were significantly greater than low-fear subjects' shock/phobic estimates, t (23) = 3.82, p < .005, and were greater than their own chime-plus-light/phobic and shock/mushroom—flower estimates, t s(24) = 3.84 and 3.29, respectively, p s < .01. In contrast, low-fear subjects' shock/phobic estimates failed to differ from comparison estimates. State Emotion There were no differences between high- and low-fear subjects in state anxiety, t (23) < 1. Fear Level × Page 10 of 24 9/24/2000http://spider.apa.org/ftdocs/abn/1989/november/abn984381.html alert people to important events. The exact chime used in New York City subway cars to alert passengers that the doors are about to close (Vapor Industries of Chicago, IL) was acquired. When activated, this chime produces two sequential tones of different pitches. Pilot testing was used to set the loudness at a level that was salient but nonaversive (approximately 70 dB). The second component of the salient stimulus was four small GE Type 44 lightbulbs, with translucent colored lenses, that flashed on and off at the rate of six times per second. The four lights were mounted next to the midpoint of the top, bottom, left, and right edges of the projection screen on which slides were shown. The duration of the chime-plus-light outcome was 2 s, the same as for the shock and tone used in Experiment 1. Procedure The procedure was identical to that of Experiment 1, with the following exceptions: Because of concern that the shock workup might induce subjects to attend more to the shock than to the chime-plus-light outcome during the trials, subjects also received a chime-plus-light workup before the experimental trials. This workup consisted of five occurrences of the chime-plus-light outcome, spaced approximately 7 s apart. The order of introduction to the chime-plus-light outcome and to the shock workup was counterbalanced. After completing the probability questionnaire assessing conditional probabilities and base-rate probabilities, subjects completed the following three questionnaires, in the order indicated: (a) a four-item questionnaire that asked subjects to indicate on 1 to 5 scales how noticeable (1 = barely, 5 = highly ) and attention-getting (1 = not at all, 5 = extremely ) they found the shock and chime- plus-light outcome to be at both the beginning and the end of the experimental procedure; (b) an eight-item questionnaire that asked subjects to indicate on 1 to 5 scales the degree to which they feared and were bothered by the shock and chime-plus-light outcomes at both the beginning and end of the experimental procedure; and (c) the state version of the Trait—State Anxiety Inventory ( Spielberger, Gorsuch, & Lushene, 1970 ). In addition, using 0 ( not at all ) to 8 ( extremely ) rating scales, subjects rated how much interest, happiness, fear, and disgust they experienced when viewing each of the four phobic slides that they were exposed to during the illusory correlation trials. 5 Data Analysis In Experiment 1, the only direct comparison between high- and low-fear subjects was a planned contrast conducted on their shock/phobic estimates. In the present study, we used two planned interaction contrasts to compare high- and low-fear subjects in the overall patterning of their covariation estimates. 6 The first contrast, Fear Level (high vs. low) × Slide Category (shock/phobic vs. chime-plus- light/phobic) assessed whether high- and low-fear subjects differed in the tendency to associate phobic slides with shock in relation to the chime-plus-light outcome. We used as a comparison to subjects' shock/phobic estimates their chime-plus-light/phobic estimates, rather than pooling across the chime- plus-light and nothing outcomes because the major purpose of this study was to assess whether subjects demonstrate covariation bias even when exposed to an outcome that matches shock in salience. The second contrast assessed whether high- and low-fear subjects' shock/phobic covariation estimates were greater than their estimates of the covariation between non-fear-relevant slides and shock. Subjects' mushroom and flower estimates were pooled for this comparison, the design of which was Fear Level (high vs. low) × Slide Category (shock/phobic vs. shock/pooled mushrooms and flowers). Both contrasts were evaluated at the .05 level. To minimize experimental error, we adjusted the criterion alpha level for subsequent simple effects comparisons to .025. Page 9 of 24 9/24/2000http://spider.apa.org/ftdocs/abn/1989/november/abn984381.html latter group demonstrated what can best be described as a mild bias, with shock/phobic estimates significantly greater than some relevant comparison estimates, but not others. The nature of subjects' covariation estimates other than shock/phobic is an additional important feature of the data. None of these other estimates differed significantly from the veridical probability of .33. This high level of accuracy indicates that high-fear subjects' overestimation of the shock/phobic contingency is not due to a generalized inability to perform the experimental task. The lack of differences among subjects' estimates of the base-rate proportions of slide categories and of outcomes suggests the same point. Experiment 2 Experiment 2 was designed to assess precisely what features of the shock were linked to feared stimuli. As noted previously, we originally reasoned that high-fear subjects might be particularly likely to associate fear-relevant stimuli specifically with an aversive outcome, because such a bias would exemplify cognitive processing that confirms fear. In the first experiment, however, it was likely that shock was not only the most aversive of the three outcomes, but also the most salient, or attention- getting. It could be argued, then, that the illusory correlations found in Experiment 1 were due to a general bias to link feared stimuli with any salient outcome. The high likelihood that fear-relevant slides were also the most salient slide category for high-fear subjects lends support to this line of reasoning, because previous studies have shown that illusory correlations can be induced merely by co-occurrences of highly distinctive stimuli ( Chapman, 1967 ; Hamilton, Dugan, & Trolier, 1985 ; Hamilton & Gifford, 1976 ; McArthur & Friedman, 1980 ). To address this issue, we replaced the nonaversive tone used in Experiment 1 with a nonaversive but highly attention-getting stimulus that consisted of a combination of an alerting chime and flashing colored lights. Pilot work with this new stimulus suggested that subjects would perceive the chime-plus- light outcome as equal to shock in salience but as markedly less aversive. Despite this matching on salience, it was hypothesized that high-fear subjects would continue to overestimate the contingency between feared stimuli and shock in relation to comparison covariation judgments. The results of Experiment 1 indicated that high-fear subjects had a significantly greater bias to associate phobic stimuli with shock than did low-fear subjects. We predicted the same results in Experiment 2 and, in addition, directly assessed differences in the patterning of covariation estimates between the high- and low-fear groups using planned interaction contrasts. Method Subjects Subjects were 25 women selected from the introductory psychology pool at the University of Wisconsin–Madison. On the basis of their responses to the SNAQ or SPIQ, high- and low-fear subjects were selected according to the same criteria used in Experiment 1. There were 6 high-snake-fear and 7 high-spider-fear subjects and 7 low-snake-fear and 5 low-spider fear subjects. In addition to these subjects, 3 additional subjects who participated in the procedure were dropped from the design because of equipment failure, knowledge of experimental hypotheses, and failure to understand covariation questions, respectively. Apparatus The experimental apparatus was identical to that used in Experiment 1 with the exception that the tone used previously was replaced by a more salient combination of a chime and flashing colored lights. We reasoned that an excellent prototype for a salient stimulus would be the chimes used in public settings to Page 8 of 24 9/24/2000http://spider.apa.org/ftdocs/abn/1989/november/abn984381.html There were 72 trials, on each of which a slide was followed by an outcome. Each slide was either fear- relevant (snake or spider, depending on experimental condition) or neutral (mushroom and flower). Slide duration was always 8 s. One of three outcomes occurred immediately at slide offset: a 2-s shock at the intensity level set during the shock workup, a 2-s tone, or nothing at all. The intertrial interval was randomly determined by means of sampling from a uniform distribution with a mean of 20 s and a range of 10 to 30 s. Overall, there were three different sequences of slides and three different sequences of outcomes, resulting in nine possible patterns of pairings of slide and outcome sequences. Each subject received one of these nine patterns, as determined by random assignment. For each possible pattern of pairings, there was no correlation between any category of slide and any type of outcome. That is, across the 72 trials, the conditional probability of any outcome given the prior occurrence of any slide category was always equal to .33. This value was also the base-rate probability of occurrence of any slide category and outcome. In the development of slide and outcome sequences, a given slide category was never followed by the same outcome on more than two consecutive occasions. At the completion of the 72 trials, the subject completed the probability questionnaire (described later), on which she indicated her perception of the relation between slides and outcomes. After completion of this questionnaire, the experimenter ascertained that the subject understood it, both by directly asking her and by reviewing her responses. The subject was then debriefed and paid $5. Probability Questionnaire This questionnaire was given to subjects at the completion of the series of trials to assess their perceptions of contingency. For purposes of the present article, this questionnaire was divided into the following two sections. 3 Covariation questions. The items in this section were the primary variables used to test experimental hypotheses. Written instructions explained to subjects that they were being asked to answer the following sort of question: Given that a specific category of slide occurred, on what percentage of trials was that category of slide followed by a specific outcome? These estimates are formally identical to estimates of the conditional probability of an outcome, given the prior occurrence of a slide category, or p (outcome/slide type). Subjects then completed nine scales corresponding to each combination of slide category and outcome. A sample item is as follows: Given that you saw a flower slide, on what percentage of those trials was the flower followed by a shock? Subjects responded to questions by marking an X at the point on the line best representing their estimated percentage. Base-rate questions. In this section, subjects made overall estimates of the percentage of trials on which each of the three slide types occurred and on which each of the three outcomes occurred. These base-rate estimates were made on the same 0—100 scale. Quantification of probability estimates. Page 5 of 24 9/24/2000http://spider.apa.org/ftdocs/abn/1989/november/abn984381.html these 45 subjects, 5 subjects participated in the procedure but were dropped from the design for the following reasons: failure to feel shocks throughout the series of trials because of habituation ( n = 2), knowledge of experimental hypotheses ( n = 2), 2 and clear failure to understand questions asking subjects to indicate covariation perceptions ( n = 1). Apparatus The experiment was conducted in a two-room suite. Subjects sat in a comfortable recliner directly facing a 116 × 120 cm projection screen hung on a wall and facing away from the experimenter's room. The experimenter monitored subjects and stimulus presentations through a small 30 × 45 cm two-way mirror on the wall separating the two rooms. A Radio Shack TRS-80 microcomputer controlled the timing and presentation of stimuli during all phases of the experiment. Experimental stimuli were 35-mm color slides from one fear-relevant category (either snakes or spiders, depending on experimental condition) and from two neutral categories (mushrooms and flowers). Mushrooms and flowers were used as neutral categories to maximize comparability with previous studies assessing conditioning to fear-relevant and fear-irrelevant stimuli (e.g., Öhman et al., 1976 ). Four unique slides from each category were used, the majority of which had been used in previous conditioning studies (e.g., Cook et al., 1986 ). Six copies of each slide were made to generate the 72 slides (24 from each category) used in all. Slides were presented through the two-way mirror by a Kodak Model 550 Carousel projector housed in the observation room of the suite. The projected slide image was 52 × 72 cm. Shock was generated by a Farrall I&nacute;struments Corporation Mark 300C stimulator equipped with a remote control switch allowing computer control of shock delivery. Shock was delivered by a concentric Tursky electrode. The maximum current that could be delivered was 5 mA. A tone (60 dB), delivered by a Sonalert, was used as the neutral-tone outcome. The Sonalert was placed on the wall 29 cm to the right of the midpoint of the vertical edge of the projection screen. Procedure One male and two female undergraduate research assistants served as experimenters. After completion of the consent form, the experimenter introduced the experimental procedure by stating that the experiment involved different kinds of stimuli presented in different sensory modalities. Then, with the Tursky electrode attached to the proximal lateral surface of the subject's upper left arm, the experimenter and subject began the shock workup procedure designed to set the intensity level of the shock. This procedure was completed when the intensity was set at a level judged by the subject to be on the borderline between uncomfortable and painful. At the completion of the shock workup, the experimenter informed the subject about the experimental task by delivering the following instructions: Over the next one-half hour or so, you'll be seeing three categories of slides. Any given slide will be followed by one of three possible outcomes. You will either hear a tone [pointing to Sonalert], or you'll feel a shock [pointing to shock electrode], or you'll experience nothing. Pay close attention to what is happening because your task is to determine whether or not there is a relationship between any category of slide and any of the outcomes following the slide. Once again, focus in on whether there is a relationship between different categories of slides and the three outcomes. After delivering these instructions, the experimenter returned to the observation room and started the computer-directed routine controlling stimulus presentations. Page 4 of 24 9/24/2000http://spider.apa.org/ftdocs/abn/1989/november/abn984381.html some inconsistencies in results, particularly for experimentally induced negative affective states in normal subjects (e.g., Bower, 1987 ; Isen, 1985 ), this evidence has shown a tendency to process information in a manner that is not only congruent with emotions but that may also serve to confirm emotions and, by this means, promote their maintenance (for a review, see Isen, 1984 ; Leventhal & Tomarken, 1986 ). Particularly relevant to the focus on individual differences in the present study are the results of studies comparing clinical depressives and nondepressives on memory and judgmental tasks. These studies have shown that depressives often selectively remember information in such a way as to confirm the pessimistic view of self and world that appears to be an important component of depression in many cases (e.g., D. M. Clark & Teasdale, 1982 ; Derry & Kuiper, 1981 ; for a review, see Blaney, 1986 ). Conversely, nondepressed individuals demonstrate optimistic biases that may well support the maintenance of positive affect and help prevent the development of depression ( Alloy & Abramson, 1979 ; Lewinsohn, Mischel, Chaplin, & Barton, 1980 ). The present studies were designed, in part, to test the hypothesis that human fears and phobias are associated with a similar bias to process information in a manner that would logically serve to confirm fear and promote its maintenance. Although fear-induced confirmatory processing could conceivably take a variety of forms, we reasoned that a particularly likely and important manifestation would be biased judgment of the covariation between feared stimuli and aversive events. Simply put, the major reason for this prediction is that overestimation of the contingency between feared objects and aversive events would appear to be a particularly direct and powerful way to confirm or maintain fear. Distorted judgment of this sort would almost certainly validate the appraisal of the feared stimulus as threatening, and heightened perceptions of threat are an important component of fear and anxiety (e.g., Beck & Emery, 1985 ; Butler & Mathews, 1983 ; Mathews & MacLeod, 1985 ). From this perspective, then, heightened perceptions of threat are both a product of fear-induced covariation bias and a factor that serves to maintain or enhance fear. In summary, we used an illusory correlation paradigm to assess whether subjects overestimate the covariation between fear-relevant slides and shock, despite extensive exposure to a series of trials that indicate the complete absence of any correlation between slide categories and outcomes. Preparedness theory, and the results of its previous empirical tests in the conditioning literature, suggested that both high- and low-fear subjects would demonstrate a significant bias. Previous evidence concerning the effects of individual differences in fear and anxiety suggested that high-fear subjects in particular would demonstrate covariation bias. These two different predictions were tested in Experiment 1. The subsequent two experiments were designed to replicate the results of Experiment 1 and to extend them by assessing some of the underlying mechanisms and boundary conditions for fear-induced illusory correlations. Experiment 1 Method Subjects Subjects were 45 paid female volunteers selected from a larger sample recruited at student unions and dormitories. Only female subjects were used because of the large sex differences in snake and spider fear (e.g., Cornelius & Averill, 1983 ). At the time of recruitment subjects completed either the SNAQ or the SPIQ, standardized self-report measures of snake and spider fear, respectively ( Klorman, Weerts, Hastings, Melamed, & Lang, 1974 ). Respondents were selected for participation in the study if they scored within the upper 20th percentile (high-fear groups) or lower 35th percentile (low-fear groups) on the snake or spider fear questionnaires. 1 Four groups were formed: high fear–snake ( n = 10; SNAQ M = 18.44, SD = 3.71), low fear–snake ( n = 11; SNAQ M = 2.40, SD = 1.43), high fear–spider ( n = 13; SPIQ M = 15.33, SD = 2.61), and low fear–spider ( n = 11; SPIQ M = 2.22, SD = .67). In addition to Page 3 of 24 9/24/2000http://spider.apa.org/ftdocs/abn/1989/november/abn984381.html