ORGANIZATIONAL

BEHAVIOR

AND

HUMAN

DECISION

PROCESSES

39, 162- 183 (1987)

What Makes You So Sure? Effects of Epistemic Motivations on judgmental

Confidence

OFRA MAYSELESS AND ARIE W. KRUGLANSKI Tel-Aviv University Three experiments examined the effects of motivational variables on the epistemic (or knowledge acquisition) process. The motivations considered were need for cognitive structure and fear of invalidity, and the epistemic phenomena studied were subjective confidence and hypothesis generation. In the first experiment a tachistoscopic task was employed to examine the effects of fear of invalidity upon (1) subjects’ initial and final confidence in a hypothesis and (2) shifts in confidence occasioned by successive items of new information. The second experiment replicated and extended the first by investigating the effects of need for structure over and above those of fear of invalidity. The third experiment employed an object recognition task and investigated the process of hypothesis generation assumed to mediate motivational effects on subjective confidence. It was found that both initial contidence and informationally induced confidence shifts were of higher magnitude when subjects’ need for structure was high rather than low, and when subjects’ fear of invalidity was low rather than high. Furthermore, the number of alternate hypotheses generated was higher under high (versus low) fear of invalidity, and low (vs high) need for structure. The above findings were discussed in reference to Kruglanski’s theory of lay epistemology. 0 1987 ~~ademicPress,Inc.

A great deal of research in contemporary social psychology has been devoted to the processes whereby people form their judgments and beliefs. Recent work in this area focused on several dependent variables. For example, a large number of studies concentrated on judgmental errors and biases (for reviews see Kahneman, Slavic, & Tversky, 1982). Several studies have also examined the phenomenon of subjective confidence, which has been investigated mainly within two separate paradigms. One paradigm studied subjective confidence in reference to empirical relative frequencies of correct answers (for a review see Lichtenstein, Fischhoff, & Phillips, 1982). In a typical experiment, subjects respond to questions to which precise answers are known, and are asked to assess Correspondence, including requests for reprints, should be directed to Arie W. Kruglanski, Department of Psychology, University of Wisconsin, 1202 W. Johnson St., Madison, WI 53706. 162 0749-5978187$3.00 Copyright0 1987by AcademicPress.Inc. All rightsof reproductionin any form reserved.

EPISTEMIC

MOTIVATION

163

the likelihood of their answers being correct. For the “well-calibrated” assessor, likelihood judgments of a given magnitude (say, p = .75) are associated with correct answers the corresponding percentage of the time (i.e., 75%). Empirical studies of calibration have shown that probabilities are related to, but are not identical with, proportions of correct answers. The most common deviation from perfect calibration appears to be overconfidence (for example, being .99 confident while only 85% correct). In the second paradigm, subjective probabilities were studied from the Bayesian perspective. Initial findings of this line of work were that generally people are conservative Bayesians who revise their judgments in gross correspondence with the Bayes’ formula albeit in an insufficient amount (for a recent review of this research see Edwards, 1982). However, further studies demonstrated that under some conditions people can render excessive rather than conservative judgments, for example, when making cascaded or hierarchical inference (cf. Schum, DuCharme, & DePitts, 1973; Youssef, 1973; Youssef and Peterson, 1973). Recent theoretical analyses (cf. Cohen, 1982; Fischhoff & Beyth Marom, 1983; Hintikka & Hintikka, 1983) suggest that beyond the issue of conservatism, the Bayesian model might not faithfully depict the process mediating peoples’ judgments. Thus, peoples’ cognitive representations of judgmental tasks may incorporate assumptions at variance with the Bayesian model. For instance, subjects might conceive of “probabilities” as indices of belief intensity rather than as ratios of favorable chances (cf. Cohen, 1982). They might also conceive of events as nonrandom (Cohen, 1982) and fail to take into account all the potentially relevant information or all the potentially relevant alternatives (cf. Hintikka & Hintikka, 1983). The above arguments imply that subjective probability estimates often depend on specific beliefs and assumptions persons bring to bear on judgmental tasks. Beyond such cognitive factors, confidence estimates might be importantly affected by the motivational context (of costs and benefits) in which judgments are rendered (cf. Christensen-Szalanski, 1978, 1980; Fischhoff & Beyth Marom, 1983). The purpose of the present investigation is to explore several such possible motivational effects on subjective confidence using as a point of departure a theory of lay epistemology developed recently by Kruglanski and his associates (e.g., BarTal & Geva, 1985; Bar-Tal, Geva, Bar-Tal, & Yarkin, 1985; Kruglanski, 1980; Kruglanski & Ajzen, 1983; Kruglanski & Freund, 1983; Kruglanski, Friedland, & Farkash, 1984). In particular, we consider the effects on judgmental confidence of two motivations, referred to as “need for structure” and “fear of invalidity,” respectively. Their role in the epistemic process is described below.

164 Motivational

MAYSELESS

AND KRUGLANSKI

Effects in the Knowledge Acquisition Process

According to the theory of lay epistemology, knowledge is acquired in course of a twofold sequence during which a hypothesis may turn into “firm,” or subjectively valid “fact” when it is consistent with all the available evidence, and when no alternative hypotheses, also consistent with the evidence, come to mind. However, in principle it is possible to continue generating further and further hypotheses consistent with any body of evidence (see also Gettys & Fisher, 1979). Thus, the individual’s tendency to terminate the hypothesis-generation process at some point) and grant a given hypothesis the status of firm knowledge requires explanation. The lay epistemic theory proposes that the inhibition (and/or facilitation) of hypothesis generation is affected by variables in two categories: the individuals’ (1) capacity to generate alternative hypotheses on a given topic, and (2) their motivation to do so (for thorough discussions of the hypothesis-generation process see also Eco and Sebeok, 1983; Fisher, Gettys, Maming, Mehle, & Boca, 1983; Gettys & Fisher, 1979). The capacity to generate numerous hypotheses on a given topic has to do with the availability of constructs in a person’s conceptual repertoire and with construct accessibility at a given moment (cf. Higgins & King, 1981). The motivation to produce alternatives to a given hypothesis may be affected at least by three separate types of need: the need for cognitive structure, the fear of invalidity and the need for specific conclusions. The need for cognitive structure refers to the desire to possess some knowledge on a given topic, any definite knowledge as opposed to confusion and ambiguity. A heightened need for cognitive structure is assumed to promote an early closure on a solution to a judgmental problem. In terms of the lay epistemic theory this is referred to as cognitive “freezing” (cf. Freund, Kruglanski,. & Schpitzajzen, 1985; Kruglanski and Freund, 1983). Thus, a person under high need of structure is likely to inhibit the generation of competing alternatives to a given hypothesis as such alternatives might appear to threaten an existing structure. A need for structure might be aroused under a variety of situational conditions, for instance, under the pressure to render a decision or to engage in some action. Alternatively, need for structure could be aroused via instructions stressing the value of order and coherence and/or those disparaging ambiguity or undue complexity. Previous research manipulating the need for structure in such ways found that individuals in which this need was aroused tended to base their judgments more (1) on early (vs late) information, thus exhibiting impressional “primacy effects,” (2) on pre-existing stereotypes, in this sense being theory rather than data

EPISTEMIC

MOTIVATION

165

“driven” (cf. Bechtold & Zanna, 1985; Freund et al., in press; Kruglanski & Freund, 1983). Furthermore, under high (vs low) need for structure individuais tended more to seek comparison with similarly minded others likely to support their views and opinions (cf. Mayseless & Kruglanski, 1985). Functionally opposite to the need for structure is the fear of invalidity. This motivation has to do with the desire to avoid judgmental mistakes, in view of their perceived costliness. If need for structure may often promote a “freezing” of the epistemic process, fear of invalidity may often promote an “unfreezing.” By this is meant an increased tendency to generate alternatives to a currently entertained hypothesis, and/or an increased sensitivity to information inconsistent with the hypothesis. A fear of invalidity might be aroused under various situational conditions, e.g., by instructions stressing the importance of correctness and accuracy, via anticipated evaluation of one’s judgments by significant others, via an expectation that target persons would suffer costs if subjects’ assessments were incorrect, etc. Previous research manipulating the fear of invalidity in such ways found that high (versus low) fear of invalidity suppressed the magnitude of primacy effects, and of the tendency to translate stereotypes into discriminatory judgments (cf. Kruglanski & Freund, 1983; Freund ef al., in press). Similarly, high (versus low) fear of invalidity enhanced social comparison with dissimilarly minded others, presumably because this might forestall the commission of costly mistakes (cf. Mayseless & Kruglanski, 1985). In addition to need of structure and fear of invalidity effects, the judgmental process may often be influenced by a variety of “conclusional needs,” that is, by wishes and desires compatible with some judgmental contents and incompatible with others. Unlike need for structure and fear of invalidity which effects are not assumed to bias the final conclusions in any specific direction, conclusional needs are assumed to sway the judgmental process toward desirable inferences. To the extent, then, that a currently entertained hypothesis appeared desirable to the individual, this might promote a “freezing” on this particular conclusion. To the contrary, if the hypothesis appeared undesirable this might facilitate “unfreezing.” In short, the effects of conclusional needs are assumed to be directional and biasing, and to represent persons’ occasional tendency to engage in “wishful thinking” (for a more extensive discussion see Kruglanski & Ajzen, 1983). The present series of experiments puts some of the above notions to empirical test. Our first experiment focuses on the effects of fear of invalidity on subjective confidence. The second experiment studies the effects of need for structure over and above those of fear of invalidity. The

166

MAYSELESS

AND KRUGLANSKI

third experiment studies the hypothesis generation process (presumed to mediate subjective confidence) as it may be affected by invalidity and structure concerns. EXPERIMENT 1 As noted earlier, a heightened fear of invalidity is expected to promote an “unfreezing” of the hypothesis generation process, stimulating the production of competing alternatives to a given hypothesis or judgment. Assuming that the alternative hypotheses are evaluated simultaneously (or nearly so) confidence should normalize. In other words, probabilities should be additive and the generation of a competing alternative should lower one’s confidence in the initial hypothesis (for a discussion of cases where normalization may not occur see Cohen, 1982; Fischhoff and Beyth Marom, 1983; Shafer, 1978). Second, a subject whose fear of invalidity is high rather than low should generate a greater number of interpretations regarding each new item of received evidence. Assuming, again, that the interpretations are considered simultaneously, this should lower the magnitude of confidence shifts (increments or decrements) occasioned by new evidence. In other words, the tendency to interpret each item of evidence in competing ways may ambiguate its meaning, hence lessen its impact on subjective confidence. Thus, the effect on confidence of new evidence should be lower under high as compared to low fear of invalidity. Our analysis also predicts that more items of relevant information would be sought by subjects whose fear of invalidity is high (versus low) because of the greater number of alternate possibilities they may find necessary to eliminate. Finally, subjects’ confidence in their final judgments should be greater under high versus low fear of invalidity. Two separate lines of reasoning suggest the last prediction: first, to avoid a costly mistake, persons may make special effort to test every conceivable hypothesis concerning the judgment of interest, thus exhausting their cognitive capacity to come up with further relevant ideas. Second, for a person committed to a given judgment, the higher the cost of a possible mistake the stronger should be the conclusional motivation to believe in having avoided it. Such tendency toward “wishful thinking” may also enhance the individual’s confidence in the judgment rendered. To summarize our hypotheses for this experiment we predicted that increasing the fear of invalidity would (a) lower the initial level of confidence in a hypothesis, (b) lower the magnitude of confidence shifts occasioned by new information, (c) heighten the tendency to seek more information before making a final judgment, and (d) heighten the final level of subjective confidence.

EPISTEMIC

MOTIVATION

167

METHOD Subjects and Procedure

Twenty-eight first-year students (12 male and 16 female) in introductory psychology at Tel-Aviv University participated as subjects in the experiment for course credits. On each of 10 trials the subject was shown a digit on a tachistoscope and was asked to identify it. Length of exposure and size of digit were preselected to produce barely visible stimuli resulting in about 50% correct recognition. The experimental exposure times always lasted 4 ms. Prior to the experimental trials subjects were given four warm-up trials at longer exposure: the first at 250 ms, the second at 150 ms, the third at 50 ms, and the last at 4 ms. The true identity of the digits was not disclosed to subjects until the end of the experiment. At each of the 10 experimental trials, a different digit was flashed on the tachistoscope and the subject was asked to identify it. The digits (l-9) were randomly selected and were presented to all subjects in the same order. Subjects were allowed to operate the tachistoscope an unlimited number of times until able to reach a decision regarding the digit’s identity. Then a new digit was displayed. Following each presentation subjects guessed at the digit’s identity and rated their confidence on a scale ranging from 0 (not at all confident) to 100 (confident beyond the shadow of a doubt). On some of the presentations subjects reported seeing nothing on the screen; in such cases (referred to as null presentations) they indicated this on the answer sheet, and proceeded to operate the tachistoscope. Manipulating the fear of invalidity. As the experiment was run fairly late in the semester, all subjects needed experimental credits rather badly. Half of the subjects, randomly assigned to the high fear of inualidity condition, were promised an additional hour of credit if successful in identifying at least 9 of the 10 digits. The remaining subjects, in the low.fear condition, were not promised extra credits nor were they advised of a contingency between correct performance and rewards of any kind. RESULTS

Several different measures were used to test our hypotheses. To obtain a measure of initial confidence, subjects’ first rating of confidence for each digit was averaged across the 10 digits. A similar procedure was adopted for the final confidence ratings: They, too, were averaged across the 10 digits to yield a measure of final confidence. An inverse index of confidence was the number of null presentations regarding which the subject reported having seen nothing on the tachistoscope. We assumed that besides random factors affecting visual perception (like blinking

168

MAYSELESS AND KRUGLANSKI

rate), fear of invalidity should have a systematic effect on the perceptual threshhold. Thus, subjects under high as compared to low fear of invalidity were expected to report more instances of null presentation because their judgmental confidence was so low they may not even venture guessing that a digit was flashed. The mean number of null presentations was calculated for each subject across the 10 digits. Finally, the mean number of null presentations reported prior to subjects’ first guess was also calculated. Our additional aim was to assess the contribution of each new datum (that is, each presentation of the digit) to subjects’ confidence level. By contribution we mean either an upward (incremental) or a downward (decremental) shift in confidence following a given stimulus presentation. As might be recalled, we expected the confidence shifts in either direction to be lower under high as compared to low fear of invalidity. To test this hypothesis the following calculations were performed. For each digit, subjects’ first rating of confidence was subtracted from the second rating (yielding a shift occasioned by the latter presentation), the second rating was subtracted from the third, and so on. The results were added without regard to sign, as both increments and decrements were assumed to be influenced by fear of invalidity level. This sum was then averaged: (1) over all presentations of the digit including the null presentations, and (2) over all presentations of the digit, excluding the null presentations. For each subject these scores were averaged across digits to yield the average confidence shift over all presentations and the average confidence shift over all but the null presentations. Means, standard deviations, and the relevant t values (one tailed) for all the above measures are displayed in Table 1. As may be seen, all our hypotheses were confirmed. Initial confidence in the high-fear condition was lower than in the low-fear condition 0, < .Ol). Furthermore, high-fear subjects sought more information and expressed higher final confidence than did low-fear subjects (JJ < .Ol for both measures). As expected, the number of null presentations in general was greater for high (versus low) fear group (p < .Ol), as was the number of null presentations reported before the first guess (p < .05). The magnitude of confidence shifts per item of information was lower for high (versus low) fear subjects using an index based on all presentations @ < .05), and it approached significance in the same direction using an index excluding the null presentations. DISCUSSION Data reported above lend support to our theoretical predictions concerning fear of invalidity effects on subjective confidence. In addition, the present findings supplement previous results (e.g., Freund et al., in

Experimental condition

2.03*

3.74**

91.46

(11.1)

30.27

68.42 (19.93)

Final confidence

(11.1)

40.75 (15.81)

Initial confidence

* t(26) (one tailed) = 1.7. p < .05. ** r(26) (one tailed) = 2.47, p < .Ol.

r

High fear of invalidity

Low fear of invalidity

TABLE 1

1.09 (1.03) 1.72*

4.89**

0.45 (0.51)

15.12 (7.71)

4.75 (3.7)

Number of presentations

Initial number of null presentations

15.49 (6.19) 1.98*

2.72**

21.67 (9.85)

Confidence change scorebased on all presentations

5.27 (5.91)

0.91 (1.08)

Overall number of null presentations

Dependent measure

1.55

21.17 (7.82)

26.93 (11.46)

Confidence change score(excluding null) presentations

MEANS, STANDARD DEVIATIONS (IN PARENTHESES), AND t VALUES FOR THE DEPENDENT MEASURES OF EXPERIMENT 1

170

MAYSELESS

AND KRUGLANSKI

press; Kruglanski & Freund, 1983) in suggesting that the fear of invalidity stimulates the search for alternate hypotheses, and/or information pertinent to deciding among such hypotheses. In previous relevant research fear of invalidity appeared to enhance individuals’ readiness to revise in light of new information otherwise recalcitrant (or “frozen”) hypotheses. In the present experiment we observed a drop in subjective confidence regarding one’s initial hypothesis as well as a drop in the impact of new information on subjective confidence. Particularly striking in this connection is our finding that subjects’ reports of having seen nothing on the screen occurred more often in the high versus the low fear of invalidity condition. A possible interpretation of this result would be that the fear of invalidity affects the tendency to formulate perceptual hypotheses as well as cognitive ones, suggesting a continuity of perceptual and cognitive processes (cf. Cheesman & Merikle, in press). Of interest and potential significance is our finding that confidence in final judgments was higher under high versus low fear of invalidity. This implies that arousing the fear of invalidity may not only decrease the resistance to change of old beliefs but also increase the resistance to change of new ones. Both effects may combine to render the manipulation of fear of invalidity a particularly effective means of producing lasting cognitive change. According to the present theory, higher final confidence level under high (versus low) fear of invalidity could be mediated by either cognitive or motivational factors. On the cognitive side, the more extensive information processing by high (vs low) fear of invalidity subjects could have exhausted their capacity to come up with further hypotheses, and/or induced the perception that they have “worked hard enough” so probably they are correct. On the motivational side, final commitment to a given judgment could have induced a conclusional need to believe in the correctness of one’s answer, or to engage in “wishful thinking” leading to the perception of correctness. Such need should be proportionate to the perceived costs of a possible mistake, hence it should be higher under high (vs low) fear of invalidity. The above cognitive and motivational explanations need not be mutually exclusive; either or both types of factor might produce the observed effects. However, the present data do not afford a decision between those possibilities, and further research (beyond the present scope) is needed to assess their relative merits. In a sense, the finding that greater final confidence obtained under high fear of invalidity is not surprising. After all, high fear of invalidity might well motivate more extensive information processing which, in turn, might inspire higher confidence. However, the present epistemic theory suggests that under some circumstances less (rather than more) extensive processing is associated with higher confidence. A case in point might be

EPISTEMIC

MOTIVATION

171

one in which an individual is under high need for structure, motivation assumed to promote judgments that are at once quick and confident. Furthermore, under high (vs low) need for structure an individual is expected to exhibit higher initial confidence, and higher magnitude of confidence shifts occasioned by new items of information. Those possibilities were explored in our second study described below. EXPERIMENT

2

The design of this study consisted of three experimental conditions. Two of these included respectively manipulations of a high fear of invalidity and a high need for structure. In addition, a neutral control condition was included in which no motivational induction was attempted. According to our theory, the need for cognitive structure should inhibit the generation of competing hypotheses. Thus, we predicted that subjects in the high need for structure (as compared to the control) condition would exhibit (1) higher confidence in the initial hypothesis (as there would be fewer competing alternatives to consider), (2) greater magnitude of confidence shifts occasioned by each new item of information, (3) a less extensive tendency to expose themselves to relevant information. Finally, we predicted that (4) subjects’ confidence in their final judgment would be higher under a high need for structure (as compared to the neutral condition) insofar as such need induces a stronger desire to cling to one’s hypothesis, hence to believe in it more confidently. Beside the predicted need for structure effects, we also expected to replicate our previous findings concerning fear of invalidity effects: on three dependent measures we expected those effects to be opposite to need for structure effects. Specifically, we expected fear of invalidity subjects to exhibit lower intial confidence, lower magnitude of confidence shifts, and higher amounts of information sought than control subjects. However, we also expected that in both motivational conditions the final confidence level would be higher than in the control condition. If confirmed, the latter prediction would indicate that confidence is affected by the presently postulated motivational factors and is not simply a function of the amount of information processed. In particular, we expected need for structure subjects to process substantially less information than the neutral controls yet exhibit higher level of final confidence, comparable to that of fear of invalidity subjects. Subjects and Procedure

Seventy first-year students (28 males and 42 females) in introductory psychology at Tel-Aviv University participated in the experiment for course credits. The procedure resembled in all details that of the previous

172

MAYSELESS

AND KRUGLANSKI

experiment except that 12 experimental digits were now shown instead of the 10 used previously. Manipulating thefear ofinvalidity. Twenty subjects assigned randomly to the fear of invalidity condition were given instructions stressing the importance of correct and exact identification of the digits. An ability to identify the digits correctly was said to correlate with mental concentration and intelligence. Manipulating the need for structure. Twenty subjects assigned randomly to the need for struture condition were given instructions stressing the importance of forming unambiguous, clear-cut opinions. An ability to reach definite decisions on the basis of scarce information was said to correlate with high mental concentration and intelligence. A neutral group of 20 randomly assigned subjects was not given any special instructions beyond the request to identify the digits as seen. The results of a manipulation check (described more fully later) indicated that the fear of invalidity level was not significantly higher (albeit in the expected direction) in the fear of invalidity condition than in the control condition. Therefore, a fourth group of subjects was included in which a “very high” fear of invalidity was manipulated using a stronger induction. Manipulating a “very high” fear of invalidity. Ten subjects were randomly assigned to the “very high” fear of invalidity condition. The procedure used was essentially similar to that employed in the high-fear condition of Experiment 1. The only difference between the two was that now we demanded that the 12 digits will all be correctly identified before awarding an extra credit of 1 h. Recall that in the previous experiment only 9 out of the 10 digits had to be identified to qualify for extra credit. Upon completion of the identification task a questionnaire was handed out to all subjects. Its purpose was to check on the experimental manipulations. The questionnaire consisted of two questions phrased as follows: (1) how personally important for you was it to identify the digits correctly, and (2) how personally important for you was it to reach a decision quickly? The former item was assumed to reflect the degree of subjects’ fear of invalidity, and the latter that of their need for structure. Subjects rated their answers on a 7-point scale ranging from 1 labeled not at all important to 7 labeled very important. After having answered those questions subjects were thoroughly debriefed and enjoined to secrecy. RESULTS

The dependent measures were computed in the same way as in Experiment 1. Table 2 presents a summary of the analysis of variance results performed on all the dependent variables; the table also displays the cell

6.14 .0009

50.51 .oooo5

1.2b (1.5) 0.92b (1.0) 0.71a.b (0.7) 0.31” (0.29)

2.76 .05

1.6O (1.3) 1.32” (0.99) 0.52” (0.48)

(6.6)

6.37b

11.87 Xl001

13.46b (5.12) 23.79b (17.19) 20.446 (14.6) 38.49“ (18.91)

7.08 .0003

* Cells not sharing the same superscript differ significantly at the p < .05 level by t test (one tailed).

Means and standard deviations (in parentheses) Experimental conditions Very high 27.43b 90.8” 18.28’ fear of invalidity (10.04) 6.82 (10.65) 6.03b Fear of 35.39 71.06b (2.94) invalidity (21.85) (22.05) 34.18b 57.98’ 5.14036 Neutral (22.85) (2.16) (27.3) 50.84” 3.2C Need for 78.03p,b (17.38) (1.47) structure (16.0)

ANOVA summaries F 4.06 P .Ol

18.07’ (4.9) 26.8b,c (18.1) 32XWb (21.5) 42.6” (20.0)

4.54 ,005

6.6b (0.69) 5.Y.b (0.64) 5.7” (0.95) 5.5” (1.21)

2.76 .05

2.9’ (0.W 3.4b,’ (1.29) 4.P (1.22) 5.27” (1.34)

7.51 .0003

Confidence Confidence Initial General change change scoreRated Rated Initial Number of number of null number of null score-all excluding null importance of importance of Final confidence confidence presentations presentations presentations presentations presentations validity decisiveness

Dependent measure

TABLE 2 ANOVA SUMMARIES, MEANS, AND STANDARD DEVIATIONS (IN PARENTHESES) FOR THE DEPENDENT* MEASURES OF EXPERIMENT 2

H

s

g tl 8 z i;

174

MAYSELESS

AND KRUGLANSKI

means and the standard deviations for the various experimental conditions. Manipulation Checks Results of analyses of variance performed on answers to the two manipulation checks are also included in Table 2. It turns out that subjects in both the “very high” fear of invalidity and the “high fear of invalidity” condition rated the perceived importance of correct answers as higher than subjects in the control and need for structure conditions. However, only the “very high fear-invalidity” group differed significantly from the latter two groups (p < .05), which did not significantly differ among themselves. On the “importance-of-decisiveness” question the need for structure group differed significantly from each of the remaining groups (p < .05), exhibiting the highest mean score. The “very high” fear of invalidity group differed significantly from the rest, albeit in the opposite direction, exhibiting the lowest mean score (p < .05). It thus appears that only manipulations of the very high fear ofinvalidity, and the need for structure were sufficiently powerful to produce significant effects. In the following description of our major findings we will, therefore, concentrate particularly on results from the latter two conditions. Confidence data. As shown in Table 2, the confidence data generally support our hypotheses: initial confidence of subjects in the very high fear of invalidity condition was lower than of subjects in the need for structure condition (p < .OS).The remaining two groups occupied intermediate positions on the confidence dimension, but they differed significantly only from the needfor structure group. In line with expectation, the very high fear of invalidity subjects requested the highest number of stimulus presentations whereas the need for structure subjects requested the lowest number of presentations. These two extreme groups differed significantly from each other (p < .05), but only the “very high” group differed significantly from the control. In addition, the overall number of null presentations was significantly higher for the very high fear of invalidity group vs the need for structure group, as was the number of null presentations reported prior to the first guess. In both cases the remaining two groups fell in between the former, extreme conditions. Also, the average magnitude of confidence shifts was significantly lower (p < .05) for the very highfear ofinvalidity subjects vs the needfor structure subjects using (1) an index based on all the presentations and (2) an index excluding the null presentations. As expected, the final confidence of the very high fear of invalidity subjects did not differ significantly from that of the need for structure

EPISTEMIC

MOTIVATION

175

subjects. Also as expected, the final confidence in both the above groups was significantly higher than in the neutral condition (p < .05). DISCUSSION

The results of our second study corroborate and extend those of the previous experiment. Specifically, present results replicate the effects on subjective confidence of a high fear of invalidity. Furthermore, need for structure appears to affect subjective confidence in the manner suggested by the lay epistemic model. As expected, high (vs low) need for structure appears to result in higher initial confidence, and a stronger impact on confidence of new information. It also results in lower frequency of null presentations, or reports of having seen nothing on the screen. Thus, subjects seem able to formulate perceptual hypotheses less hesitantly under high vs low need for structure. Of theoretical interest is the finding that our need for structure subjects processed considerably lower amounts of information than the fear of invalidity subjects yet exhibited approximately equal levels of final confidence. This results lends credence to our suggestion that subjective confidence cannot be fully explained by reference to the amount of information processed. Instead, confidence seems also influenced by various motivational forces as suggested above. Our predictions for Experiments 1 and 2 rested on the assumption that effects of motivational variables on subjective confidence are mediated by the process of hypothesis generation. For instance, we assumed that high fear of invalidity facilitates the generation of alternative hypotheses which, in turn, may lower subjective confidence. By contrast, high need for structure was assumed to inhibit the generation of alternative hypotheses, which may bolster subjective confidence. So far, however, our assumptions concerning the motivational determinants of hypothesis generation were not directly tested, The next experiment in the series was. therefore, designed to provide some information on that issue. EXPERIMENT

3

Subjects and Procedure

Thirty first-year students (16 males and 14 females) in introductory psychology at Tel-Aviv University participated in the experiment for course credits. Subjects were randomly assigned to three conditions: (1) a high need for structure condition, (2) a high fear of invalidity condition, and (3) a neutral condition where no special motivation was manipulated. Testing was carried out individually. Each subject was shown 10 enlarged photographs of parts of common objects. The photos were taken from unusual angles so that the objects were not readily recognizable. Subjects

176

MAYSELESS

AND KRUGLANSKI

were first challenged to list down all the conceivable hypotheses that came to mind regarding an object’s identity. They also rated on a IOOpoint scale their confidence in each hypothesis. Second, subjects noted their final answer, and their confidence in it. The final answer was recorded in a separate booklet so as to set it apart from the initial judgment, and to see whether confidence ratings have changed between the two judgments. Manipulating the need for structure. Instructions to subjects in the need for structure condition stated that the experiment investigates people’s ability to reach firm decisions on the basis of scarce information. Subjects were further told that this particular ability is of considerable functional importance, and constitutes a significant component of general intelligence. Manipulating the fear of invalidity. Instructions to subjects in the fear of invalidity condition stated that the experiment studies people’s capacity for correct visual recognition under limited informational circumstances. This particular ability was said to have considerable functional significance, and to constitute a component of general intelligence. After receiving the foregoing instructions, subjects were handed the first booklet in which they registered their hypotheses and, subsequently, the second booklet in which they recorded their final answers. This was followed by a debriefing session in which subjects were fully advised of the experiment’s true purpose and of the deceptions involved. RESULTS

Several dependent measures were used. First, we computed for each subject the mean number of hypotheses generated per each stimulus object. We also computed the mean initial confidence level over all the hypotheses recorded in the first booklet, and the mean final confidence level over the 10 final answers entered in the second booklet. Finally, we computed the mean initial confidence level in hypotheses later chosen as final. Results of analyses of variance performed on the above measures are displayed in Table 3 along with the appropriate means and standard deviations. Number of Hypotheses

Generated

The number of hypotheses measure yielded sign&ant differences between conditions (P’ = 26.46, p -=z.OOOl). As expected, subjects in the high fear of invalidity condition generated more hypotheses than did their counterparts in the neutral condition (p < .05). Also as expected, subjects in the high need for structure condition generated less hypotheses than subjects in the neutral condition @ < .05).

EPISTEMIC

TABLE ANOVA

177

MOTIVATION 3

SUMMARIES, MEANS* AND STANDARDDEVIATIONS (IN PARENTHESES)FORALL DEPENDENTMEASURESOF EXPERIMENT3 Dependent measures Initial confidence in hypothesis finally chosen

ANOVA F P

Final confidence in hypothesis chosen

Initial confidence over all hypotheses

Number of hypotheses generated

summaries 9.85 .0006

Means and standard deviations Experimental conditions 68.606 High fear of invalidity (13.93) Neutral 39.70” High need for structure

10.18

7.66 .002

26.16 .oool

72.2gb

57.226

(17.19)

(17.72)

.ooo5

43.35”

34.24”

2.46’ (0.55) 1.63b

(14.51) 58.09

(14.89)

(13.39)

(0.39)

64.33b

57.6gb

1.18”

(15.69)

(11.95)

(14.49)

(0.15)

* Cells which do not share the same superscript differ significantly at the (p < .05) level by f test (one-tailed).

Confidence Ratings As shown in Table 3, our effort to differentiate between the final and initial phases of the judgmental process was not very successful. Although there was a slight increase in confidence between initial and final ratings it did not reach statistical significance. Thus, it seems possible that subjects arrived at their final decision while still working on the first booklet. Such conclusion is further supported by various behavioral and verbal indications manifested by subjects during the experimental session. Consequently, we can regard both initial and final ratings of confidence as referring essentially to final confidence. It is of further interest to note that both types of rating exhibit the same striking trend: ratings of need for structure and fear of invalidity groups are significantly higher than those of the neutral group. Granting our interpretation that all contidence ratings refer to final confidence, the present result replicates the finding of our previous two experiments, where need for structure and fear of invalidity subjects attained higher final confidence than did neutral controls. DISCUSSION

Our primary aim in the present experiment was to demonstrate the link

178

MAYSELESS

AND KRUGLANSKI

between epistemic motivations (notably, the need for structure and the fear of invalidity) and the number of alternate hypotheses a person generates on route to a judgment. The results yield a clear picture. Our fear of invalidity subjects generated more hypotheses than did neutral controls, whereas our need for structure subjects generated less hypotheses. Above data accord well with predictions of the lay epistemic model. It is also noteworthy that while fear of invalidity subjects generated more hypotheses than did need for structure subjects, their final confidence level was about the same and significantly higher than that of neutral controls. As combined with findings of our second experiment, this result augments the conclusion that confidence level is not exclusively determined by purely cognitive (or informational) considerations and is at least in part affected by the kind of motivational factors outlined in the lay epistemic theory. GENERAL

DISCUSSION

The present series of experiments explored the effects of need for structure and fear of invalidity on subjective confidence. We predicted that high (vs low) fear of invalidity would result in lower initial confidence, weaker impact on confidence of new information, requests for greater amounts of information, and higher final confidence. All the above predictions were borne out by the results of our first study. We further predicted that high (vs low) need for structure would result in higher initial confidence, stronger impact on confidence of new information, requests for lesser amounts of information, and higher final confidence. Results of our second study support the above predictions while also replicating our previous findings concerning fear of invalidity effects. The third experiment replicated the final confidence findings of the former two studies. Despite considerable differences in tasks employed, final confidence was higher under high fear of invalidity or high need for structure as compared to a neutral control condition. Even more importantly, the results of Experiment 3 supported our notions about the mediating variable which might account for motivational effects on subjective confidence: the number of alternate hypotheses generated during the judgmental process. As expected, greater number of such hypotheses was generated under high fear of invalidity and lesser under high need for structure, both as compared with a neutral control condition. It is noteworthy that the experimental task employed in our first two studies differed substantially from that employed in the last study. Unlike in Study 3, in both Studies 1 and 2 the number of possible alternatives and their identities (digits from 1 to 9) were known in advance. Thus, availability considerations as well as creativity probably played a minor

EPISTEMIC

MOTIVATION

179

role in the first two experiments and possibly a greater role in the third experiment. From the present theoretical perspective, the same motivational variable may affect hypothesis generation regardless of whether the hypotheses are chosen from a fixed set of known alternatives or are creatively constructed from an open array of possibilities. Some of our findings are directly consistent with the above assumption, in particular the final confidence data which exhibited the same pattern across all three studies. Somewhat less directly, the number of hypotheses generated, investigated in our third study, seemed affected by the present motivational variables in ways consistent with the assumption that it mediates subjective confidence, investigated in our first two studies. However, to argue more cogently for similarity of motivational effects, further research is needed where both hypothesis generation and subjective confidence would be measured under different judgmental tasks. A Social Desirability Interpretation A possible alternative explanation of some of our need for structure effects could be phrased in terms of the social desirability artifact. Specifically, in our two last experiments need for structure was manipulated via instructions portraying as (socially) desirable the ability to form judgments quickly and decisively. Therefore, subjects might have tried to convey the desired image of judgmental decisiveness by reporting high confidence levels throughout the experiment without actually experiencing high confidence, and by requesting minima1 amounts of information prior to announcing their final judgments. It is noteworthy, however, that the social desirability explanation does not readily account for some of the remaining effects observed in our need for structure conditions. For instance, the social desirability hypothesis can not cogently explain why need for structure subjects exhibited more drastic confidence shifts than did their counterparts in the neutral and fear of invalidity conditions: after all, radical confidence shifts may be taken to signify capriciousness and whimsy seemingly out of step with ideals of cool decisiveness implicit in our need for structure instructions. Furthermore, the social desirability hypothesis does not readily explain the lower number of hypotheses generated by subjects in the need for structure condition as compared to those in the neutral, or fear of invalidity, conditions. Recall that our instructions challenged subjects to record all possible hypotheses that came to mind. Thus, if our need for structure subjects were primed to impress the experimenter they could have shown off their intelligence and creativity by producing a maximal number of such hypotheses. Yet quite the opposite occurred. Finally, the social desirability hypothesis cannot readily explain the present fear of

180

MAYSELESS

AND KRUGLANSKI

invalidity effects. Specifically, it seems difficult to discern a simple logical link between the presumptive desire to appear correct and the lower display of initial confidence, lower magnitude of confidence shifts, or the more extensive informational search exhibited by our high fear of invalidity subjects as compared to neutral controls. In sum, the social desirability hypothesis, while able to account for some of our findings in the need for structure condition, does not seem to account for all such findings, nor does it tit the obtained fear of invalidity effects. Thus, it seems reasonable to conclude that the social desirability hypothesis represents a less than convincing alternative to the lay epistemic explanation of the entire data pattern observed in the present studies. Implications

for Other Social Cognitive

Research

Beside supporting the lay epistemic conception of the judgmental process, present findings have implications for several issues of current interest in the field of social cognition. Essentially, our results imply the need to impose motivational limiting conditions regarding several currently prevalent generalizations about judgmental behavior such as the following. Judgmental conservatism. The present findings are pertinent to the issue, noted earlier, of whether persons are conservative or excessive by comparison to the Bayesian formula (cf. Cohen, 1982; Edwards, 1982; Fischhoff & Beyth Marom, 1983). Our results imply that the rate at which subjective probabilities are revised may depend on persons’ motivational states. Thus, the revision might be slower under a high fear of invalidity but faster under a high need for structure. In short, fear of invalidity might induce a tendency to be conservative, whereas need for structure might induce a tendency to be excessive. Those possibilities could be profitably pursued in further research. Hypothesis testing. Our present findings contain several implications concerning the recently active research topic of hypothesis testing (cf. Snyder & Swann, 1978; Trope & Bassok, 1983). Consider, for instance, Snyder and Swann’s (1978) proposal that in the testing of hypotheses (about themselves or other people or events) people predominantly seek to confirm their preexisting notions. This implies a pervasive insensitivity to disconfirming evidence, and (presumably) a reluctance to generate alternatives to a currently tested hypothesis. According to our present findings, however, the tendency to confirm one’s hypothesis could depend on one’s motivational state: persons experiencing high need for structure, for example, might seek to confirm their hypotheses in a manner implied by Snyder and Swann. Yet persons under high fear of invalidity might act very differently. Rather than tending to confirm their

EPISTEMIC MOTIVATION

181

hypotheses come what may, such individuals might adopt a “skeptical” stance for prolonged intervals during which they might contemplate numerous competing alternatives to currently considered notions. Implications for everyday situations. Generally, the present series of studies underscores the considerable influence of epistemic motivations on the experience of subjective confidence. Extrapolating to everyday situations, the present findings imply that in environments characterized by high demand for structure (e.g., in a regimented military ambience, or in a culture stressing the value of order), persons are likely to exhibit considerable self-assurance and a tendency to “leap to conclusions” from limited evidential bases. Similarly, persons exposed to novel, confusing circumstances (e.g., a teenager about to enter adulthood, an immigrant transplanted to a foreign land) may often experience an acute need for guiding structure to which they might respond by adopting inappropriately high levels of self-assurance. This may be often paralleled by ego (or ethno) centric “I (we) know best” attitudes and by close-mindedness to alternative points of view. Finally, it is of interest to speculate about stable individual differences in needs for structure and validity, and about how these may affect persons’ judgmental styles. Consider the perfectionist who, prior to making a commitment, continues for inordinate amounts of time to check out all conceivable alternatives to a judgment or a decision. Such a person might harbor a persistently high fear of invalidity. At the other extreme, consider the overconfident “know-all” individual never lacking an assured point of view on any topic. Such individual might harbor a persistently high need for structure. Above notions are at present quite speculative and in need of extensive empirical probing for their validation. They do, however, illustrate some research directions in which the present work could be extended. REFERENCES Bar-Tal, D., & Geva, D. (1985). A cognitive basis of international conflicts. In W. G. Austin & S. Worchel (Eds.), The social psychology of intergroup relations. (2nd ed.). New York: Nelson-Hall. Bar-Tal, D., Geva, N., Bar-Tal, Y., & Yarkin, K. (198.5).Planning and performing interpersonal inferaction: A cognitive-motivational approach. Unpublished manuscript, TelAviv University. Bechtold, B., & Zanna, M. P. (198.5).Unpublished data, Waterloo University. Cheesman, J., & Merikle, P. M. (in press) Word recognition and consciousness. In D. Besner, T. G. Weller, & G. E. MacKinnon (Eds.), Reading research: Advances in theory and practice (Vol. 5). New York: Academic Press. Christensen-Szalanski, J. J. J. (1978). Problem solving strategies: A selection mechanism, some implications and some data. Organizational Behavior and Human Performance, 22, 307-323. Christensen-Szalanski, J. .I. J. (1980). A further examination of the selection of problem-

182

MAYSELESS AND KRUGLANSKL solving strategies. The effects of deadlines and analytic aptitudes. Organizational havior and Human Performance,

Be-

25, 107-122.

Cohen, L. J. (1982). Are people programmed to commit fallacies? Further thoughts about the interpretation of experimental data on probability judgments. Journal for the Theory of Social Behavior, 12, 251-274. Eco, V., & Sebeok, T. A. (Eds.). (1983). The sign of three. Bloomington: Indiana Univ. Press. Eco, V. (1983). Horns, hooves, insteps: Some hypotheses on the types of abduction. In V. Eco & T. A. Sebeok (Eds.) The sign of three. Bloomington: Indiana Univ. Press. Edwards, W. (1982). Conservatism in human information processing. In D. Kahneman, P. Slavic, & A. Tversky (Eds.), Judgment under uncertainty: Heuristics and biases. Cambridge, MA. Fischhoff, B., & Beyth Marom, R. (1983). Hypothesis evaluation from a Bayesian perspective. Psychological Review, 90(3), 239-260. Fisher, S. D., Gettys, C. F., Marning, C., Mehle, T., & Boca, S. (1983). Consistency checking in hypothesis generation. Organizational Behavior and Human Performance, 31, 233-254. Freund, T., Kruglanski, A. W., & Schpitzajzen, A. (in press). The freezing and unfreezing of impressional primacy: Effects of the need for structure and the fear of invalidity. Personality

and Social Psychology

Bulletin.

Gettys, C. E, & Fisher, S. D. (1979). Hypothesis plausibility and hypothesis generation. Organizational Behavior and Human Performance, 24, 93- 110. Higgins, E. T., & King, G. (1981). Accessibility of social constructs: Information processing consequences of individual and contextual variability. In N. Cantor & J. KihlStrom (Eds.), Personality, cognition and social interaction. Hillsdale, NJ: Erlbaum. Hintikka, J., & Hintikka, M. B. (1983). Sherlock Holmes confronts modern logic: Toward a theory of information seeking through questioning. In V. Eco & T. A. Sebeok (Eds.) The sign of three. Bloomington: Indiana Univ. Press. Kahneman, D., Slavic, P., & Tversky, A. (Eds.). (1982). Judgment under uncertainty: Heuristics and biases. Cambridge, MA: Cambridge Univ. Press. Koriat, A., Lichtenstein, S., & Fischhoff, B. (1980). Reasons for confidence. Journal of Experimental Psychology: Human Learning and Memory, 6, 104- 118. Kruglanski, A. W. (1980). Lay epistemo-logic: Process and contents: Another look at attribution theory. Psychological Review, 87, 70-87. Kruglanski, A. W., & Ajzen, I. (1983). Bias and error in human judgment. European Journal of Social Psychology, 14, 1-44. Kruglanski, A. W., & Freund, T. (1983). The freezing and unfreezing of lay inferences: Effects on impressional primacy, ethnic stereotyping, and numerical anchoring. Journal

of Experimental

Social Psychology,

19, 448-468.

Kruglanski, A. W., Friedland, N., & Farkash, E. (1984). Laypersons’ sensitivity to statistical information: The case of high perceived applicability. Journal of Personality and Social Psychology, 46, 503-518. Lichtenstein, S., Fischhoff, B., & Phillips, L. D. (1982). Calibration of probabilities: The state of the art to 1980. In D. Kahneman, P. Slavic, & A. Tversky (Ed%), Judgment under uncertainty: Heuristics and biases. Cambridge MA: Cambridge Univ. press. Mayseless, O., & Kruglanski, A. W. (1985). Reconsidering the role of perceived similarity in social comparison: A lay epistemic interpretation. Unpublished manuscript, TelAviv University. Schum, D. A., DuChame, W. M., & DePitts, K. E. (1973). Research on human multistage probabilistic inference processes. Organizational Behavior and Human Performance, 10(3), 318-348.

EPISTEMIC MOTIVATION

183

Shafer, G. (1978). Non-additive probabilities in the work of Bernoulli and Lambert. Archive for History

of Exact Sciences, 19, 309-370.

Snyder, M., & Swann, W. B., Jr. (1978). Hypothesis-testing processes in social interaction. Journal of Personality and Social Psychology, 36, 1202-1212. Trope, Y., & Bassok, M. (1983). Information gathering strategies in hypothesis testing. Journal of Experimental Social Psychology, 19, 860-876. Youseff, Z. I. (1973). The effects of cascaded inferences on the subjective value of information. Organizational Behavior and Human Performance, N(3), 359-363. Youseff, Z. I., & Peterson, C. R. (1973). Intuitive cascaded inferences. Orgnnizational Behavior and Human Performance, 10(3), 349-358. RECEIVED: May 31, 1985