Running Head : INTROSPECTION AND SELF - ENHANCEMENT The Why ’ s the Limit : Curtailing Self - Enhancement with Explanatory Introspection

Self-enhancement is linked to psychological gains (e.g., subjective well-being, persistence in adversity), but also to intrapersonal and interpersonal costs (e.g., excessive risktaking, antisocial behavior). Thus, constraints on self-enhancement may sometimes afford intrapersonal and interpersonal advantages. We tested whether explanatory introspection (i.e., generating reasons for why one might or might not possess personality traits) constitutes one such constraint. Experiment 1 demonstrated that explanatory introspection curtails selfenhancement. Experiment 2 clarified that the underlying mechanism must (a) involve explanatory questioning rather than descriptive imagining, (b) invoke the self rather than another person, and (c) feature written expression rather than unaided contemplation. Finally, Experiment 3 obtained evidence that an increase in uncertainty about oneself mediates the effect. Introspection and Self-Enhancement 3 The Why’s the Limit: Curtailing Self-enhancement with Explanatory Introspection Most people, most of the time, see themselves through rose-colored glasses. Whether rating themselves as above-average on personality traits and abilities (Alicke, 1985) or believing themselves less susceptible to bias than the average person (Pronin, Yin, & Ross, 2002)— whether showing selective recall for flattering autobiographical episodes (Sanitioso, Kunda, & Fong, 1990) or engaging in social comparisons that validate a positive self-view (Dunning, 1999)—whether attributing their successes internally and their failures externally (Mezulis, Abramson, Hyde, & Hankin, 2004) or thinking that their own future will surpass that of their peers (Weinstein, 1980)—people by and large evaluate themselves more favorably either than the objective facts warrant (Gosling, John, Craik, & Robins, 1998) or than external observers think justified (Epley & Dunning, 2000). Tellingly, people even believe that they outdo their own doppelgangers: they rate themselves more favorably than they rate their peers on the basis of identical behavioral evidence (Alicke, Vredenburg, Hiatt, & Govorun, 2001). Moreover, egocentric biases like the better-than-average effect are pervasive existing not only in (selfpromoting) individualistic cultures, but also in (self-deprecating) collectivistic cultures (Sedikides, Gaertner, & Toguchi, 2003). All such phenomena can be viewed as forms of self-enhancement. Although perhaps irrational in the normative sense—half of us being forever doomed to be below average—selfenhancement is nonetheless linked to substantial benefits. These include good psychological health (Taylor, Lerner, Sherman, Sage, & McDowell, 2003), better coping with physical illness (Taylor et al., 2003) and traumatic loss (Bonanno, Rennicke, & Dekel, 2005), greater persistence in the face of adversity (Taylor & Brown, 1988), and good social adjustment (Donnellan, Trzesniewski, Robins, Moffitt, & Caspi, 2005). However, self-enhancement is also linked to several substantial costs. Intrapersonal costs include imprudent risk-taking (Baumeister, Heatherton, & Tice, 1993), ineffective action planning (Oettingen & Gollwitzer, 2001), and an increased likelihood of disengaging from Introspection and Self-Enhancement 4 academic studies (Robins & Beer, 2001). Interpersonal costs involve being perceived negatively and treated unpleasantly by others. For example, after a brief period of infatuation, peers come to regard inveterate self-enhancers as conceited, defensive and hostile (Paulhus, 1998), and are generally prone to deride them, if not isolate them interpersonally (Schlenker & Leary, 1982). In addition, concerns about promoting or protecting a favorable public self-image can prompt actions that lead to illness, injury, and death: Notoriously, people from temperate climes often sunbathe for hours to look and feel good among their peers, thereby raising their risk of sunstroke, sunburn, and skin cancer (Leary, Tchividjian, & Kraxberger, 1994). In view of these inauspicious correlates, it is perhaps salutary that self-enhancement, although pervasive, is not inevitable: it varies naturally and can be strategically manipulated. For example, the topic of judgment moderates self-enhancement: people self-enhance less on traits that lack ambiguity (Dunning, Meyerowitz, & Holzberg, 1989) or that they believe they can modify (Dauenheimer, Stahlberg, Spreeman, & Sedikides, 2002). In addition, several interpersonal factors are also known to constrain self-enhancement. These include the similarity of the comparison other to the self (Stapel & Schwinghammer, 2004), the concreteness of the comparison other (Alicke, Klotz, Breitenbecher, Yurak, & Vredenburg, 1995), concerns about preserving close relationships (Tice, Butler, Muraven, & Stillwell, 1995), and social pressures to be accountable (Sedikides, Herbst, Hardin, & Dardis, 2002). However, given the problems that self-enhancement sometimes poses, it is worth exploring what other factors have the potential to curtail it. In this article, we investigate a possible intrapersonal factor: introspection. Varieties of Introspection The human ability to introspect has long fascinated philosophers. Descartes (see Cottingham, Stoothoff, & Murdoch, 1984) regarded reflexive thought as proof of an indubitable self. Introspection has also captivated the attention of psychologists, from the early structuralists (Titchener, 1912; Wundt, 1894) to modern-day experimental social psychologists (Hirt & Markman, 1995; Hixon & Swann, 1993; Wilson, Dunn, Kraft, & Lisle, 1989). Importantly, introspection is considered a uniquely human capacity (Sedikides & Skowronski, 1997, 2000; Introspection and Self-Enhancement 5 Sedikides, Skowronski, & Dunbar, 2006) and its investigation is central to personality and social psychology (Bless & Forgas, 2000; Maio & Olson, 1998; Wilson & Dunn, 2004). Conceptual Distinctions Introspection is the process of looking inward, thinking “about [one’s] thoughts and feelings” (Wilson et al., 1993, p. 33), or about oneself as a whole. However, introspection is not a unitary construct. Indeed, it can be conceptualized in at least two distinct ways. One type of introspection constitutes what we term descriptive introspection. This denotes the act of contemplating what one’s personality is like. When introspecting descriptively, people ask themselves questions like “Do I have (or not have) traits X and Y?” or “To what extent do I have (or not have) traits X and Y?” People then conclude that they possess or lack particular traits to some degree or other. Another type of introspection constitutes what we term explanatory introspection. This denotes the act of contemplating why one does or does not think of oneself in a particular way. When introspecting explanatorily, people ask themselves questions like “Why might I have (or not have) traits X and Y?” or “What are the reasons for my having (or not having) traits X and Y?” People then generate reasons that explain why they either possess or lack particular traits to some degree or other. Descriptive and Explanatory Introspection: A Review of the Literature Descriptive and explanatory introspection, or key elements thereof, have already been operationalized as independent variables in past research. Consider two lines of inquiry. First, Tesser (1978) investigated the consequences of thinking about an attitude object for which a well-developed knowledge base exists. Intensive thinking led to the formation of an evaluatively-consistent belief set, which in turn polarized attitudinal judgments. That is, intensive thinking produced “more univalent, less ambivalent” attitudes (p. 295). Second, Hixon and Swann (1993: Experiment 3) had participants peruse particular dimensions of personality. In particular, undergraduates with low self-esteem pondered the question “What kind of person are you in terms of sociability, likeability, and interestingness?” while weighing up the accuracy of two evaluations—one flattering and one critical—that graduate students ostensibly provided of Introspection and Self-Enhancement 6 them. Consistent with their pre-existing negative self-view, the undergraduates endorsed the critical evaluation over the flattering one. The two lines of inquiry have common elements. First, in terms of procedure, participants either reviewed a stored body of knowledge, or answered a “what” question. Both activities are clearly reminiscent of descriptive introspection. Second, in terms of outcome, participants either consolidated an attitude or confirmed a self-view. Either way, a previously held belief was strengthened. The conjunction of these facts suggests that descriptive introspection is a source of psychological stability (Silvia & Gendolla, 2001). In other lines of research, examples of explanatory introspection are clearly discernible. Wilson and his colleagues have investigated the impact of this type of introspection on attitudes towards various objects (e.g., the self, political candidates, collegiate classes; Wilson, Dunn et al., 1989). Participants wrote down some reasons why they liked or disliked an object and thereafter expressed their attitudes toward that object. Reasons-analysis perturbed attitudes, prompting either a shift in direction or an increase in variability (Wilson et al., 1993). This perturbation was attributed to the temporary accessibility of reasons that, although easily verbalized and subjectively plausible, are nonetheless unrepresentative of the full set of reasons and at odds with dispositional preferences. Similar experimental procedures, findings and explanations apply to a line of research on value change by Maio, Olson, and colleagues (Bernard, Maio, & Olson, 2003a; Maio & Olson, 1998; Maio, Olson, Allen, & Bernard, 2001). Both research programs suggest that explanatory introspection is an agent of psychological change. Explanatory introspection also features in research on explanatory bias. Participants, when instructed to explain why a particular hypothetical outcome might occur, overestimate the likelihood of its occurrence (Ross, Lepper, Strack, & Steinmetz, 1977). The bias is observed regardless of whether the to-be-explained outcome pertains to the self (Kunda & Sanitioso, 1989), to another person (Anderson, 1982), or to an event like a political election (Caroll, 1978) or sporting competition (Markman & Hirt, 2002). As in the attitudes/values literature, Introspection and Self-Enhancement 7 information availability and accessibility have been invoked as underlying mechanisms. In particular, the goal of explaining some outcome prompts an information search that brings outcome-consistent arguments to the forefront of the mind (Tversky & Kahneman, 1974), where they influence, in an assimilative manner, the ensuing judgment (Kunda & Sanitioso, 1989). An alternative account of explanatory bias posits that the goal of outcome explanation prompts a frame of mind in which the explanation (or focal hypothesis; Koehler, 1991) is assumed to be true. Evidence is then reviewed from the perspective of that frame, and thus selectively accumulates in the direction of the focal hypothesis, leading its merits to be overestimated (Hirt & Markman, 1995). Regardless of the underlying mechanism, research on explanatory bias suggests that explanatory introspection has well-defined directional effects. Finally, explanatory introspection features in debiasing research. In a typical task, participants are presented with an event and instructed to explain how it might give rise both to one outcome and to another (alternative or contrary) outcome. This task—variants of which are known as counterexplanation, consider-the-opposite, inoculation, or consider-an-alternative— attenuates the magnitude of the explanatory bias (Anderson, 1982; Hirt & Markman, 1995; Hirt, Kardes, & Markman, 2004; Lord, Lepper, & Preston, 1984). This body of research suggests that explanatory introspection, when it involves a consideration of more than one point of view, exerts a moderating influence on psychological processes. Taking our cue from the above lines of research, we wondered whether explanatory introspection could curtail self-enhancement. We accordingly devised an introspection manipulation that blended elements of a prototypical debiasing manipulation with elements of a typical reasons-analysis manipulation. Specifically, we had participants generate reasons for why they might or might not have a set of important personality traits. Two key features of our adaptation are worth noting. First, our participants focused on the self rather than on a hypothetical person, object, or event. Second, our participants focused on central (or core) facets of the self (Sedikides, 1993). Thus, with the self involved, our particular adaptation likely Introspection and Self-Enhancement 8 facilitated the emergence of motivational processes above and beyond conventional cognitive ones. Any account of underlying mechanisms would need to take this into consideration. Pretesting First off, we ran a pretest in order to identify a set of nomothetic trait dimensions that participants would regard as central to their self-concept. In this pretest—as in all subsequently reported experiments—participants were undergraduates from the University of North Carolina at Chapel Hill, fulfilling an introductory psychology course option. Central trait dimensions can be operationally defined as those that elicit extreme ratings when it comes to three pertinent properties: self-descriptiveness (i.e., either highly selfdescriptive or not at all self-descriptive), valence (i.e., either highly positive or highly negative), and importance (i.e., very important to have or very important not to have). Sixty-five participants duly rated 24 trait adjectives—corresponding to the positive and negative poles of 12 trait dimensions—in terms of all three properties (Table 1). Central trait dimensions were then selected for use, if two conditions were met. First, the positive pole of the dimension had to be rated among the four most self-descriptive, most positive, and most important to have; second, the negative pole of the dimension had to be rated among the four least self-descriptive, least positive (i.e., most negative), and least important to have. These selection criteria yielded three central trait dimensions: honest-dishonest, kind-unkind, and trustworthy-untrustworthy. These trait dimensions were subsequently broken down into two contrasting categories of trait adjective for use in the experiments: central positive (honest, kind, trustworthy), and central negative (dishonest, unkind, untrustworthy). Experiment 1 The objective of Experiment 1 was to test whether explanatory introspection curtails selfenhancement. We instructed participants to analyze the reasons both for why they might and might not have a particular trait. Additionally, we asked some participants to introspect explanatorily about positive traits, others about negative traits. Participants in the control group engaged in a neutral task irrelevant to self. Our prediction was that, compared to control Introspection and Self-Enhancement 9 participants, explanatory introspection participants would self-enhance less by giving both lower self-ratings on positive traits and higher self-ratings on negative traits. Method Participants and Experimental Design Eighty-eight participants were randomly assigned to a 2 (Cognitive Activity: Explanatory Introspection vs. Control) X 2 (Trait Valence: Positive vs. Negative) balanced factorial design. In this and all subsequent experiments, participants were tested individually and debriefed thoroughly. Procedure Participants assigned to the two Explanatory Introspection cells were instructed to generate reasons for why they might or might not have each of three traits. In the Positive cell, the traits in question were honest, kind and trustworthy, and in the Negative cell, dishonest, unkind and untrustworthy. The instructions read as follows: “We are interested in the reasons why you might or might not have the trait ___Please take a few moments to think about why you might or might not have the trait ___.We want you to analyze very carefully the reasons you might or might not have the trait ___ because this will help you organize your thoughts for subsequent tasks.” Participants were encouraged one final time to analyze very carefully why they both might and might not have each trait, and were then asked to write the relevant reasons down, using a separate page for each trait. Participants assigned to the two Control cells instead listed as many uses as possible—again, positive or negative, depending on the cell— for three everyday objects (spoon, brick, and briefcase; cf. Sedikides, Campbell, Reeder, & Elliot, 1998), and again used a separate page for each item. Participants were told that all pages were theirs to keep if they so desired so as to encourage frank responding. However, all opted to leave the pages behind in the experimental booth. Next, all participants (including controls) rated the self-descriptiveness of three traits (positive or negative, depending on the experimental condition). In particular, they responded to Introspection and Self-Enhancement 10 the question: “To what extent do you think you have the trait ___?” (1 = not at all, 15 = very much). Finally, to explore underlying mechanism, Explanatory Introspection participants (but not Controls) labeled each reason that they generated as either “confirming” or “disconfirming” the trait they had considered. Results Self-Evaluation The three trait self-descriptiveness ratings were internally consistent (α = .95) and so averaged to form a composite index. We then entered this index to a two-way factorial ANOVA (Cognitive Activity X Trait Valence). A significant main effect for Trait Valence emerged: Participants rated positive traits (M = 12.64) as more self-descriptive than negative traits (M = 3.36), F(1, 84) = 1402, p < .001, replicating a well-established finding (Sedikides, 1993). More importantly, this main effect was qualified by a predicted interaction, F(1, 84) = 9.67, p < .005. Explanatory introspection participants regarded positive traits (M = 12.20, SD = 1.38) as significantly less self-descriptive than controls did (M = 13.09, SD = .98), F(1, 42) = 6.10, p < .02, and regarded negative traits (M = 3.68, SD = 1.35) as marginally more selfdescriptive than controls did (M = 3.03, SD = .85), F(1, 42) = 3.66, p < .06. That is, explanatory introspection participants, compared to controls, evaluated themselves less positively and (tendentially) more negatively. In sum, explanatory introspection curtailed self-enhancement. Reasons Generated On the basis of past research, we expected that, during explanatory introspection, participants would engage in autobiographical searches, retrieving episodic or abstract information from long-term memory. This was indeed the case. In this and subsequent experiments, the reasons that participants gave (a) were non-overlapping, and (b) consisted almost uniquely of episodic memories or habitual behaviors, for example, “I [once] lied to parents about where I went at night” (confirming dishonest) and “I [typically] tell people actually what I think about them” (confirming honest). Introspection and Self-Enhancement 11 We also expected that the reasons participants generated would correspond intelligibly to their self-descriptiveness ratings. To begin with, we summed the total number of reasons that each participant labeled as confirming a trait, and then divided this by the total number of reasons they generated for that trait. We derived such a ratio separately for each trait, and then created a composite confirmation index by averaging all three ratios (α = .75). A one-way ANOVA incorporating this index showed that participants generated a significantly higher proportion of confirming reasons when they considered positive traits (M = .73) than when they considering negative ones (M = .38), F(1, 42) = 35.57, p < .001. Interpreted somewhat differently, participants confirmed their positive but disconfirmed their negative traits, replicating past research (Dunning et al., 1989; Sedikides, 1993). More importantly, we investigated whether the confirmation index correlated significantly with participants’ self-descriptiveness scores. It did, r(42) = .72, p < .001. This result suggests that explanatory introspection participants based their self-descriptiveness ratings largely on the reasons they generated. Moreover, this account is in keeping with previous research showing that the generation of supportive thoughts increases the endorsement of personality characteristics (Davies, 2003). However, alternative accounts—for example, that reasons were based on self-descriptions—cannot be definitively ruled out. (We investigate the matter further in Experiment 3.) Note that the correlations between the confirmation index and self-descriptiveness scores for participants considering positive traits (r[20] = .27, p < .23) and negative traits (r[20] = .40, p < .07) did not differ significantly from one another, z = .45, p < .65. Summary Relative to Controls, participants who explanatorily introspected showed an attenuated tendency to self-enhance. In particular, they regarded positive traits as significantly less selfdescriptive, and negative traits as marginally more self-descriptive. Regardless of trait valence, self-descriptiveness scores correlated with confirmatory reasons generated via explanatory introspection, suggesting that self-judgments varied as a function of the accessibility of autobiographical instances. Introspection and Self-Enhancement 12 Discussion What are the psychological mechanisms by which explanatory introspection curtails selfevaluation? Explanatory introspection both reduced the positivity of self-views on positive dimensions and tended to increase the negativity of self-views on negative dimensions. Any comprehensive account must therefore explain why self-enhancement was attenuated in both cases. The first point to note is that, given the ubiquity and common pre-eminence of the selfenhancement motive (Baumeister, 1998; Sedikides & Gregg, 2003), participants’ levels of selfregard were likely approaching their upper limit. This is because, to the extent that people can self-enhance, they generally will: the balloon of self-regard will rise as far as the ballast of rational and normative constraints permits (Sedikides & Strube, 1997). At the start of the experimental session, our participants, already fairly high-achieving members of a Western culture, would not have been under any special pressure to self-derogate. Their levels of selfregard would likely have been closer to their maximum than their minimum. Thus, their selfregard would have had more room for maneuver in a downward direction than in an upward one, regardless of whether they explanatorily introspected about positive traits or about negative ones. Hence, any factor undermining self-regard would have observably reduced it more than any intrinsically comparable factor promoting self-regard would have observably increased it. The second point to note is that, generally speaking, negative factors exert a greater impact than positive ones (Baumeister, Bratslavsky, Finkenauer, & Vohs, 2001). To take one of numberless examples, the prospect of losing a substantial sum of money strikes most people as more aversive than the prospect of gaining that sum strikes them as attractive (Kahneman & Tversky, 1981). Now, explanatory introspection participants were instructed to consider, not only why they might possess, but also why they might not possess, particular traits. Thus, when those traits were positive, participants considered both why they might possess them (an attractive reflection) and why they might not (an aversive reflection); and when those traits were negative, participants considered both why they might possess them (an aversive reflection) and Introspection and Self-Enhancement 13 why they might not (an attractive reflection). Given the generally greater power of negative factors, it would hardly be surprising if participants’ aversive reflections exerted greater psychological impact than the attractive reflections. If they did—and if, as seemed to have been the case, their self-regard varied as a function of the reasons they generated—then the net result would have been a reduction in self-enhancement. The combination of both dynamics plausibly accounts in general for why explanatory introspection curtails self-enhancement, regardless of whether positive or negative traits are considered. Of course, this is only a distal outline; the proximal details still require filling in. The effects of explanatory introspection are likely proximally mediated by induced variations in the accessibility of self-knowledge (Davies, 2003; Fazio, Effrein, & Falender, 1981; Schwarz et al., 1991). Explanatorily introspecting participants, when attempting to answer self-generated questions about whether they possess or lack personality traits, will engage in retrospective mental simulations (Sanna, 2000) and autobiographical memory searches (Kihlstrom, Beer, & Klein, 2003). Such simulations and searches will prompt consideration of a relatively broad set of plausible alternatives. Participants will bring to mind both instances in which they behaved in a trait-confirming manner and instances in which they behaved in trait-disconfirming manner. The relative accessibility of these instances, accompanied by a state of heightened selfuncertainty (Petty, Brinol, & Tormala, 2002), will then trigger corresponding self-judgments (i.e., trait self-descriptiveness ratings). In terms of the two dynamics discussed above, negatively-toned simulations and searches are liable to be rendered more accessible, or to be weighted more heavily, than positively-toned ones; and, given the normative positivity of selfregard, such negatively-toned simulations will have greater scope for impact. Experiment 2 One purpose of Experiment 2 was simply to replicate Experiment 1. We therefore included experimental and control conditions permitting the effects of explanatory introspection to be tested, both when positive and negative traits were considered. But Experiment 2 had an additional purpose: to pin down the precise preconditions for curtailing self-enhancement Introspection and Self-Enhancement 14 through explanatory introspection. This necessitated some methodological additions and theoretical extensions. First, we wondered whether the active ingredient of our manipulation might be the more general act of asking explanatory questions about personality traits (or anything else) rather than the more specific act of asking explanatory questions about one’s own personality traits. Do inquiries have to be self-directed in order for self-enhancement to be curtailed, or will otherdirected inquiries suffice? Because only self-directed inquiries constitute introspection, this question needed to be addressed. To address it, we directly manipulated the target of scrutiny (Target Type). In particular, we had half the participants consider their own personality traits (Self), and the other half an acquaintance’s personality traits (Other). We predicted that selfenhancement would be curtailed only in the Self condition. Note that this distinction between self-directed and other-directed inquiry parallels one drawn by previous researchers (Klein & Loftus, 1988; Sedikides & Green, 2000), who argued that different cognitive processes are at work when individuals process self-related versus other-related information: elaboration in the first case (i.e., considering a new instance in relation to prior self-knowledge), organization in the second (i.e., considering a new instance in relation to other instances). Second, we further explored the hypothesis that temporary self-knowledge accessibility mediates the impact of explanatory introspection on self-enhancement. As before, we asked all participants in the Explanatory Introspection condition to generate reasons why they might have or not have a set of traits. This time, however, we instructed only half of them to list those reasons in written form, and instructed the other half merely to entertain those reasons in mental form. We labeled this variable Activity Type (Written vs. Mental). We suspected that the requirement to write reasons down would be a critical factor in success of the manipulation. For one thing, the act of writing something down is liable to concretize and stabilize thoughts that would otherwise remain hypothetical and fleeting; this, in turn, is liable to increase durably the accessibility of trait-related thoughts and their derivative associations (cf. Pennebaker, 2003). For another thing, the act of writing something down is liable to engender consistency motivation by Introspection and Self-Enhancement 15 committing participants to the content of statements willingly expressed (Festinger & Carlsmith, 1959) or increasing a sense of accountability (Tetlock, Skitka, & Boettger, 1989); this, in turn, is liable to increase durably the weight ascribed to the underlying thoughts and associations. Hence, we predicted that the effects of explanatory introspection would be present in the Written condition but not in the Mental condition. Third, past research suggests that, whereas explanatory introspection instigates a relatively impartial search of relevant autobiographical details (i.e., one that promotes psychological change), descriptive introspection instigates a relatively biased search (i.e., one that preserves psychological consistency; Tesser, 1978; Hixon & Swann, 1983). Hence, only explanatory introspection should curtail self-enhancement: descriptive introspection should merely maintain it. We tested this hypothesis by manipulating Inquiry Type (Explanatory vs. Descriptive). In particular, half of the participants considered the reasons why they (or someone else) did or did not possess particular traits (Explanatory), whereas the other half merely considered the extent to which they (or someone else) did or did not possess particular traits (Descriptive). (Note: In the Mental condition, Descriptive participants thought about the extent of trait possession, whereas in the Written condition, they committed those thoughts to paper.) We predicted that only explanatory participants (inquiring about self) would show moderation of self-regard on positive traits and extremification of self-regard on negative traits. Finally, we modified our key manipulation slightly to reinforce its construct validity. In Experiment 1, both explanatory and control participants were free to take as much time as they needed to complete the task at hand. This methodological imperfection left the door open for possible confounds. For example, explanatory participants may have taken longer than control participants. If so, then the findings of Experiment 1 may simply have been due to more protracted cognitive activity. Hence, we standardized the task completion time to eliminate such temporal confounds. Specifically, all participants were allotted three minutes per trait. In summary, Experiment 2 tested the boundary conditions of the self-enhancement curtailment effect observed in Experiment 1. We predicted that this effect would be observed Introspection and Self-Enhancement 16 only (or primarily) when participants (a) engaged in self-directed inquiries (as opposed to otherdirected ones), (b) listed relevant considerations in writing (as opposed to merely mentally entertaining them), and (c) engaged in explanatory (as opposed to descriptive) introspection. Method Participants and Experimental Design One-hundred and sixty participants were randomly assigned to one of 16 experimental conditions yielded by a 2 (Target Type: Self vs. Other) X 2 (Activity Type: Written vs. Mental) X 2 (Cognitive Activity: Explanatory vs. Descriptive) X 2 (Trait Valence: Positive vs. Negative) balanced factorial design. A further 20 participants were randomly assigned to one of two control conditions (Trait Valence: Positive vs. Negative) identical to those in Experiment 1. The purpose of these control conditions was to test the replicability of Experiment 1, and to permit an additional test of the hypotheses of Experiment 2. Procedure Participants in the Explanatory condition were instructed to generate reasons (in written or mental form) for why someone (either they or another person) might or might not have three traits (either positive or negative). Instructions and traits dovetailed those of Experiment 1. Participants in the Descriptive condition were instructed to describe the extent to which someone might or might not have each trait. Participants in the Self condition directed their trait-related inquires towards themselves, whereas those in the Other condition directed their trait-related inquiries towards an acquaintance. Before beginning, the latter wrote down the name of an acquaintance, and then stated (a) how many times they had interacted with him or her, (b) how well they knew him or her, and (c) how positive or negative their impression of him or her was. On average, participants reported that they had interacted with the acquaintance several times (M = 5.36 times) but they did not (yet) know him or her very well (M = 3.69, on a 9-point scale ranging from 1 = not well Introspection and Self-Enhancement 17 at all to 9 = very well), although they had nonetheless formed a mildly positive impression of him or her (M = 6.24, on a 9-point scale ranging from 1 = very negative to 9 = very positive). Participants in the Written condition were instructed to list, on a separate sheet for each trait, the reasons (or thoughts) they had generated. Participants in the Mental condition were instructed that they need not to write anything down: it would suffice to generate the relevant reasons (or thoughts) in their head. After being asked to generate reasons why (or thoughts about the extent to which) they might and might not possess each trait, all experimental participants were informed that they could generate as many or as few reasons (or thoughts) as they wished, but that they must do so within three minutes. Participants in the Control condition, working to the same deadline, were instructed to list as many uses as possible for a spoon, brick, and briefcase. All but 11 participants opted to leave the reasons pages behind in the experimental booth. The final manipulated factor, Trait Valence, applied to both experimental and control participants. In different conditions, the former considered either three positive or three negative traits, and the latter either positive or negative uses for three objects. Finally, all participants completed self-descriptiveness trait ratings, as they had in Experiment 1. Results and Discussion Self-Evaluation Being internally consistent (α = .94), the three trait self-descriptiveness ratings were again averaged to form a composite index. We then entered this index into a four-way factorial ANOVA (Target Type X Cognitive Activity X Trait Valence X Activity Type). Replicating Experiment 1, a significant main effect for Trait Valence emerged, with participants endorsing positive traits (M = 12.21) more strongly than negative traits (M = 3.92), F(1, 144) = 1164, p < .001. Importantly, this main effect was qualified by a three-way interaction between Target Type, Cognitive Activity, and Trait Valence, F(1, 144) = 3.96, p < .05. To clarify its meaning, we then examined the two-way Cognitive Activity X Trait Valence interaction separately for Introspection and Self-Enhancement 18 each level of Target Type (Other vs. Self). For Other, the interaction was not significant, F(1,72) < 1; for Self, it was, F(1, 72) = 4.11, p < .05. Specifically, Explanatory participants in the self condition endorsed positive traits marginally less strongly than Descriptive participants (Ms = 11.68 vs. 12.63), F(1, 36) = 3.44, p < .07; they also endorsed negative traits nonsignificantly more strongly (Ms = 4.62 vs. 3.92), F(1, 36) = 1.22, p < .28. This suggests that, averaging across Activity Type, explanatory introspection curtailed self-enhancement overall (relative to descriptive introspection). However, the above three-way interaction was in turn qualified by Activity Type, to yield the predicted four-way interaction, F(1, 144) = 4.67, p < .04 (Table 2). We decomposed it by examining the three-way Cognitive Activity X Trait Valence X Activity Type interaction separately for each level of Target Type (Other vs. Self). For Other participants, the three-way interaction was not significant, F(1, 72) < 1, p < .99; for Self participants, it was, F(1, 72) = 6.13, p < .02. To further clarify our findings, we then decomposed this significant three-way interaction for Self participants in two ways. First, we examined the two-way Cognitive Activity X Trait Valence interaction for each level of Activity Type (Mental vs. Written). For Mental participants, the interaction was not significant, F(1, 36) < 1; for Written participants, it was, F(1, 36) = 8.12, p < .02. In terms of simple effects, Explanatory participants (who wrote down their inquiries) endorsed positive traits significantly less strongly than Descriptive participants (who wrote down their thoughts), F(1, 18) = 6.92, p < .02; they also endorsed negative traits marginally more strongly, F(1, 18) = 2.14, p < .14. As predicted, self-enhancement curtailment occurred only when self-directed explanatory inquiries took written form. Second, we examined the two-way Activity Type X Trait Valence interaction for each level of Cognitive Activity (Descriptive vs. Explanatory). For Descriptive participants, the interaction was not significant, F(1, 36) < 1; for Explanatory participants, it was, F(1, 36) = 10.92, p < .002. In terms of simple effects, Written participants (who wrote down why they did or did not possess traits) endorsed positive traits less strongly than Mental participants (who Introspection and Self-Enhancement 19 merely contemplated why they did or did not possess traits), F(1, 18) = 4.78 , p < .05; they also endorsed negative traits more strongly, F(1, 18) = 6.14, p < .05. As predicted, self-enhancement curtailment occurred only when self-directed writings documented reasons for possessing or lacking traits. In summary, we confirmed all hypotheses regarding the boundary conditions of the effects observed in Experiment 1. Self-enhancement was curtailed when participants (a) considered their own traits rather than those of another person, (b) wrote down what they considered rather than merely keeping it in mind, and (c) inquired into why those traits were held as opposed to the extent to which they were held. Supplementary analyses. With a view to replicating the results of Experiment 1 and more robustly testing our hypotheses, we conducted additional planned comparisons between experimental and control participants. In particular, we examined three types of participants: (a) those who explanatorily introspected about their own personality traits in written form (Self/Written/Explanatory, or SWC); (b) those who reflected upon the extent of their own personality traits in written form (Self/Written/Descriptive, or SWD); and (c) those who considered possible uses for three everyday objects in written form (Control, or CON). We principally sought to investigate whether SWC participants self-enhanced less than CON participants, replicating the results of Experiment 1. However, we additionally sought to investigate whether (a) the SWD and CON participants self-enhanced similarly with one another, but (b) together self-enhanced more than SWC participants. This would establish the essential comparability of the Descriptive Introspection manipulation (newly featured in Experiment 2) and the Control manipulation (also featured in Experiment 1). Any effects of explanatory introspection would therefore be tested relative to a consistent baseline in Experiments 1 and 2. We duly regressed the composite self-descriptiveness index onto three predictors: a main effect contrast for Trait Valence (Positive = 1, Negative = -1); two main effects contrasts to test predictions (a) and (b) above respectively [(a) SWC = 0, SWD = +1, CON = -1; (b) SWC = 1, SWD = -.5, CON = -.5]; and two interaction contrasts created by multiplying the contrast values Introspection and Self-Enhancement 20 for the Trait Valence main effect by the contrast values for each of the Cognitive Activity main effects. All relevant means and standard deviations are presented in Table 3. First, we compared SWC participants to SWD and CON participants combined in terms of their Trait Valence differentials. The critical interaction contrast was significant, B = -.21, t(54) = -3.67, p < .001. Next, we conducted both main effect contrasts for Positive and Negative traits separately. As predicted, the difference between SWD and CON participants was not significant for Positive traits, B = .06, t(27) = .37, p < .75, or for Negative traits, B = .08, t(27) = .43, p < .65. These results attest to the comparability of the Descriptive and Control introspection conditions. Also as predicted, the difference between SWC participants, and the SWD and CON participants combined, was significant for both Positive traits, B = -.51, t(27) = -3.11, p < .01, and Negative Traits, B = .39, t(27) = 2.21, p < .05. Self-enhancement was significantly curtailed among SWC participants relative to SWD and CON participants. Reasons We will start by providing examples of reasons that participants listed in the Cognitive Activity (Explanatory vs. Descriptive) X 2 (Trait Valence: Positive vs. Negative) conditions, when the target type was the self and the activity type was written. These examples are: “I am always straightforward and tell a person how it is” (confirming honest, Explanatory Positive condition); “Sometimes I tell people things that others don’t want me to tell them” (confirming untrustworthy, Explanatory Negative condition); “People always tell me how nice I am” (confirming kind, Descriptive Positive condition); and “It is too tiring to be nice all the time” (confirming unkind, Descriptive Negative condition). In Experiment 1, Explanatory participants rated (following the manipulation) the degree to which each trait was self-descriptive, and then labeled the reasons they had listed as either confirming or disconfirming each trait. However, this practice was vulnerable to confounds involving self-perception (Bem, 1972) or dissonance (Festinger & Carlsmith, 1959). That is, participants’ reasons-labeling decisions may have been driven, at least in part, by a need to maintain consistency with the prior self-descriptiveness ratings. For example, participants who Introspection and Self-Enhancement 21 rated themselves as honest may subsequently have come to perceive the reasons they listed as confirming their honesty, especially if they valued being honest, or their reasons admitted of interpretation. To partly address this possibility, we asked two independent coders, unaware of the hypotheses under study, to label each reason that Explanatory participants listed as either confirming or disconfirming each relevant trait (for either Self or Other). The coders agreed 96% of the time and resolved disagreements though discussion. We proceeded by computing a confirmation index for each participant (α = .81) as in Experiment 1. Next, we entered this index into a Target Type X Trait Valence ANOVA. Replicating Experiment 1, participants were more likely to generate reasons confirming positive traits than reasons confirming negative traits (Ms = .86 vs. .27), F(1, 25) = 36.90, p < .001. However, this effect was qualified by an interaction, F(1, 25) = 5.75, p < .02. Participants showed a weak explanatory tendency to confirm positive traits less for Self (M = .80) than for Other (M = .92), F(1, 12) = 1.64, p < .22, combined with a marginal tendency confirm negative traits more for Self (M = .46) than for Other (M = .15), F(1, 13) = 4.24, p < .06. In our view, this makes it less likely that consistency motivation led participants to revise their reason-labels in light of their self-descriptiveness ratings. If they had, then the tendency to confirm positive and disconfirm negative traits should have been more pronounced in the more personally consequential Self condition than in the less personally consequential Other condition. As in Experiment 1, we examined the relation between participants’ self-descriptiveness ratings and the confirmation index derived from participants’ own reason-labelings. The correlation was again significant, r(27) = .85, p < .001, suggesting that participants partially based their self-descriptiveness ratings on the reasons that they generated, although the reverse causal path cannot be ruled out. As before, no significant difference emerged in participants’ propensity to form online self-evaluations (z = .10, p < .92) after explanatorily introspecting about positive traits, r(12) = .63, p < .02, and after explanatorily introspecting about negative traits, r(13) = .61, p < .02. Introspection and Self-Enhancement 22 Summary Experiment 2 achieved several substantive objectives. First, it replicated the selfenhancement curtailment effect observed in Experiment 1. Second, it ruled out a potential rival explanation for the effect, namely, that it was merely due to more protracted thinking. Third, Experiment 2 extended Experiment 1 by identifying several key boundary conditions of the selfenhancement curtailment effect. It showed that explanatory cognition is essential (descriptive cognition does not suffice); it showed that self-directed cognition is essential (other-directed cognition does not suffice); and it showed that that written expression is essential (abstract contemplation does not suffice). Finally, Experiment 2 provided further correlations between listed-reasons and self-ratings suggesting that the changes in the acute accessibility of selfknowledge lie at the heart of the self-enhancement curtailment effect. Experiment 3 In Experiment 3, we sought to test whether explanatory introspection curtails selfenhancement by reducing self-certainty (Petty et al., 2002). The experiment followed a five-step procedure. First, participants rated themselves on three positive traits. (For simplicity, we omitted negative traits). We labeled these ratings pre-introspection self-descriptiveness, or SDPRE. Second, we introduced the manipulation: participants were randomly assigned to introspect explanatorily, to introspect descriptively, or to perform a control task. Third, participants rated how certain they were that they possessed the three positive traits; that is, they indicated how sure they were about SDPRE. We labeled these ratings pre-introspection selfdescription certainty, or CERTPRE. Fourth, participants re-rated themselves on the same three traits. We labeled these ratings post-introspection self-descriptiveness, or SDPOST. (This dependent measure corresponds to the main dependent measure of Experiments 1 and 2.) Fifth, participants re-rated how certain they were that they possessed the three positive traits; that is, they indicated afresh how sure they were about SDPOST. We labeled these ratings postintrospection self-description certainty, or CERTPOST. Introspection and Self-Enhancement 23 What pattern of results would suggest that a reduction in self-certainty was responsible for the impact of explanatory introspection on self-enhancement? Just this: After explanatorily introspecting, participants should be relatively less certain about their original self-views. This decrease in certainty should in turn shape their post-manipulation self-views, now revised downwards. However, after re-expressing their revised self-views, participants’ self-certainty should rebound. In more technical terms, we expected that Explanatory participants (relative to both Descriptive and Control participants) would, following the manipulation, have lower CERTPRE ratings, because they would now be less certain of their original self-views. Such participants would also have lower SDPOST ratings, controlling for SDPRE ratings, because explanatory introspection would have curtailed their proclivity to self-enhance. Most importantly, variations in self-certainty would also mediate the effects of the manipulation on self-views; that is, CERTPRE ratings would mediate the effects of the manipulation on SDPOST. However, following the expression of SDPOST, self-certainty would be restored: no differences between conditions in CERTPOST would be observed. Method Participants, Experimental Design, and Procedure Fifty-one participants were assigned randomly to one of three conditions: explanatory introspection (Explanatory), descriptive introspection (Descriptive), and object-use generation (Control). Thus, the experiment featured a one-way balanced between-subjects design. Procedures were largely identical to those of Experiment 2 (in the Self and Written conditions). As in Experiment 1, all participants left the entire booklet behind. Participants completed SDPRE ratings for three traits: honest, kind, and trustworthy. The manipulation followed. Finally, all participants completed CERTPRE ratings, SDPOST ratings, and CERTPOST ratings. Measures Introspection and Self-Enhancement 24 SDPRE ratings. Participants responded to two items for each trait. The first read “Please rate yourself, relative to other college students your own age, on the trait ___” (1 = lower 5%, 10 = upper 5%). The second read, “Please rate yourself, relative to other people in general, on the trait ___” (1 = lower 5%, 10 = upper 5%). We averaged both items for each trait to create three indices, (α = .91, .85, and .95, for honest, kind, and trustworthy, respectively). Next, we averaged these indices to create a final composite index, SDPRE (α = .85). Higher scores indicate higher pre-manipulation levels of trait self-descriptiveness. CERTPRE. Participants responded to three items for each trait. The first read, “How certain are you of the accuracy of the ratings you made a few moments ago in reference to the trait ___?” (1 = not at all certain, 15 = very certain). The second read, “How confident are you in the accuracy of the ratings you made a few moments ago in reference to the trait ___?” (1 = not at all confident, 15 = very confident). The third read, “How sure are you that the ratings you made a few moments ago about the trait ___ reflect your true level of the trait ___?” (1 = not at all sure, 15 = very sure). We averaged the three items for each trait to create three indices (α = .94, .96, and .95, for honest, kind, and trustworthy, respectively). Next, we averaged these indices to create a final composite index, CERTPRE (α = .80). Higher scores indicate greater certainty about pre-manipulation levels of trait self-descriptiveness. SDPOST. Participants responded to three items for each trait. The wording was varied slightly in order to discourage reflexive repetition of previous responses. The first item read, “How descriptive of you is the trait ___?” (1 = not at all descriptive, 15 = very descriptive). The second read, “To what extent do you think you have the trait ___?” (1 = not at all, 15 = very much). The third read, “How well does the trait ___ describe you?” (1 = not well at all, 15 = very well). We averaged the three items for each trait to create three indices (α = .93, .89, and .97, for honest, kind, and trustworthy, respectively), and then averaged these indices to create a final Introspection and Self-Enhancement 25 composite index, SDPOST (α = .77). Higher scores indicate higher levels of post-manipulation trait self-descriptiveness. CERTPOST. These items were identical to those used for CERTPRE, with one minor modification. Each item referred to certainty about the accuracy of “...the ratings you JUST made in reference to the trait ___.” We averaged the three items for each trait to create three indices (α = .97, .98, and .97, for honest, kind, and trustworthy, respectively). Next, we averaged these indices to create a final composite index, CERTPOST (α = .83). Higher scores indicate greater certainty about post-manipulation levels of trait self-descriptiveness. Results and Discussion Self-Evaluation All means and standard deviations for the self-evaluation results are presented in Table 4. Did explanatory introspection reduce self-description certainty? We subjected CERTPRE ratings to a one-way ANOVA. The main effect was significant, F(2, 48) = 3.14, p < .05: the pattern suggested that Explanatory participants (M = 12.03) were less certain about their traits than both Descriptive participants (M = 13.03) and Control (M = 13.49) participants. We used planned comparisons to pin down the locus of the effect. Specifically, after standardizing certainty ratings, we devised linear contrasts that (a) compared Explanatory participants to Descriptive and Control participants combined, and (b) compared Descriptive participants to Control participants. We simultaneously entered these orthogonal contrasts as predictors of the standardized certainty ratings. As predicted, Explanatory participants were less self-certain than Descriptive and Control participants combined, B = -.32, t(48) = -2.39, p < .03, but Descriptive and Control participants did not differ in their self-certainty, B = -.10, t(48) = -.77, p < .45. Thus, explanatory introspection reduced certainty about pre-introspection self-descriptiveness ratings. Introspection and Self-Enhancement 26 Did explanatory introspection curtail self-enhancement (after controlling for preintrospection self-descriptiveness)? We subjected SDPOST ratings to a one-way ANCOVA, with SDPRE ratings serving as a covariate. The main effect for the manipulation was again significant, F(2, 47) = 3.83, p < .05: the pattern suggested that Explanatory participants (M = 12.31) regarded the positive traits as less self-descriptive than both Descriptive participants (M = 13.14) and Control participants (M = 13.62). Unsurprisingly, the effect of SDPRE ratings on SDPOST ratings was also significant, F(1, 47) = 7.70, p < .01. Next, we devised linear contrasts analogous to (a) and (b) described above. We simultaneously entered both contrasts, together with SDPRE ratings, as predictors of SDPOST, after again standardizing both sets of ratings. Descriptive and Control participants did not differ in terms of their SDPOST ratings, B = -.16, t(47) = -1.29, p < .25. However, Explanatory participants regarded the positive traits as less self-descriptive than did Descriptive and Control participants combined, B = -.31, t(47) = -2.44, p < .02. Thus, even after controlling for SDPRE ratings, explanatory introspection curtailed self-enhancement, replicating both previous experiments. Did self-description certainty statistically mediate the impact of explanatory introspection on self-descriptiveness? To determine whether CERTPRE mediated the impact of explanatory introspection (characterized in terms of the two linear contrasts—[a] and [b] above) on SDPOST, we adopted Baron and Kenny’s (1986) analytic strategy. We had already satisfied one requirement—that the independent variable should significantly predict the dependent variable. Specifically, we had found that explanatory introspection led to relatively lower SDPOST ratings (adjusted for SDPRE ratings). We had also already satisfied another requirement—that the independent variable should significantly predict the proposed mediator. Specifically, we had found that explanatory introspection led to relatively lower CERTPRE ratings. We now sought to satisfy the final requirements—(a) that the proposed mediator, CERTPRE ratings, significantly Introspection and Self-Enhancement 27 predict the dependent variable, adjusted SDPOST ratings, controlling for the independent variable, explanatory introspection and (b) that, in the same analysis, the predictiveness of the independent variable is reduced significantly. We succeeded. Specifically, when adjusted SDPOST ratings were regressed on CERTPRE ratings, and on the two linear contrasts (a) and (b), the effect of CERTPRE ratings persisted, B = .68, t(46) = 7.39, p < .001, but the key linear contrast (a), previously significant, became nonsignificant, B = -.11, t(46) = -1.23, p < .25. Importantly, a significant indirect effect of that contrast on SDPOST ratings via CERTPRE emerged, z = 2.05, p < .05. In summary, the impact of explanatory introspection on post-introspection self-descriptiveness ratings was mediated by certainty about pre-introspection self-descriptiveness ratings. Was certainty restored following post-introspection self-descriptiveness ratings? We subjected CERTPOST ratings to a one-way ANOVA. Contrary to what was found for CERTPRE ratings, this main effect was not significant, F(2, 48) = 1.05, p < .40. Explanatory participants (M = 12.91) were nearly as certain about their post-introspection self-descriptiveness ratings as were Descriptive participants (M = 13.37) and Control participants (M = 13.70). For completeness, we ran the same planned contrasts as before, (a) and (b). Unsurprisingly, neither attained significance: (a) B = -.19, t(48) = -1.31, p < .20; (b) B = -.09, t(48) = -.61, p < .60. Reasons Dovetailing the results of Experiments 1 and 2 (for positive traits), explanatory participants generated reasons that they labeled as confirming their self-descriptiveness ratings (77%). However, the design of Experiment 3, unlike that of previous experiments, permitted the disambiguation of two competing causal alternatives: Did Explanatory participants use reasons as a basis for (generating) their self-descriptions? Or did they use their self-descriptions as a basis for (labeling) their reasons? Support for the first alternative would be signaled by (a) a significant positive correlation between the confirmation index and SDPOST ratings, and (b) no Introspection and Self-Enhancement 28 significant positive correlation between the confirmation index and SDPRE ratings. Support for the second alternative would be signaled by the reverse pattern. Like before, we computed a confirmation index (α = .71) and correlated it with SDPOST ratings. The correlation was significant, r(15) = .77, p < .001. However, the corresponding correlation with SDPRE ratings was not, r(15) = .37, p < .14. Moreover, the difference between the two correlations was marginal, z = 1.66, p < .10. Thus, a pattern emerged consistent with the first alternative (and with our favored interpretation of relevant findings of Experiments 1 and 2). Explanatory participants based their self-descriptiveness ratings on the products of their introspections, and did not label their reasons in light of their newly revised self-views. Summary Experiment 3 established that explanatory introspection curtails self-enhancement by decreasing self-certainty. Three lines of evidence supported this assertion. First, explanatory introspection decreased participants’ certainty about their self-views. Second, this decrease in self-certainty fully mediated self-enhancement curtailment. Third, after re-expressing self-views, participants recovered their former levels of self-certainty.