Abstract
“Strong” attitudes often have greater impact than “weak” attitudes. However, emerging research suggests that weak (uncertain) attitudes can substantially influence thinking or behavior. We propose metacognitive reflection as a moderator between traditional strength patterns and these emerging attitude bolstering patterns. Across six studies, research participants encountered a target person who agreed or disagreed with participants’ attitudes. When focused on evaluating the target, attitudes predicted target evaluations better when the attitude was held with certainty (Study 1A), or after certainty had been primed (Studies 2A and 3; strength effects). However, when engaged in attitudinal social comparison (metacognitive reflection), attitudes better predicted target evaluation when the attitudes were held with doubt (Study 1B), or after doubt had been primed (Studies 2B and 3; bolstering effects). Expected change in certainty served as a mediator of attitude effects in metacognitive reflection but not target-focus conditions when doubt had been primed (Study 4).
No analysis of factors that predict thinking, judgment, and behavior would be complete without considering attitudes. Alongside habits and subjective norms, attitudes toward the target itself and toward the behavior have each been shown to predict behavioral intentions and behaviors (e.g., Ajzen & Fishbein, 2005). But not all attitudes are equally impactful on thinking, behavior, and judgment. A wealth of research has identified qualities of attitudes that moderate the attitude’s impact or strength in determining outcomes (see Petty & Krosnick, 1995, for a review). That is, sometimes it is not enough that an attitude be positive toward a particular attitude object. It may also be necessary that the attitude be strong enough to guide the behavior.
What factors influence an attitude’s predictive ability (i.e., strength)? Early research on attitude strength demonstrated that direct experience (rather than indirectly learning) regarding the attitude object led to increased use of those attitudes in determining behavior (Regan & Fazio, 1977). In addition to direct experience, other strength-related features enhance an attitude’s impact, including elaboration, issue importance, and certainty (cf. Petty & Krosnick, 1995). Much research has documented the moderating role of attitude certainty in attitudinal outcomes. When attitudes are held with greater confidence, the attitudes are more likely to persist across time, resist counterattack, and predict thoughts and behavior compared with attitudes held with less certainty (e.g., Bassili, 1996; Fazio & Zanna, 1978; see Glasman & Albarracín, 2006; Tormala & Rucker, 2007, for reviews).
Attitude Uncertainty and the Motivation to Bolster
Despite this general picture linking certainty to strength of attitudes, a growing literature suggests that there are settings in which traditionally weak attitudes nonetheless impact information processing and behavior, possibly in an attempt to bolster or strengthen that weak attitude (e.g., Clark & Wegener, 2013). Of greatest relevance to the current research, consider previous examples of attitude uncertainty motivating people to select information, process information, or identify with groups that support their attitude. That is, when selecting information to read that either agrees or disagrees with one’s opinion, people with opinion uncertainty are most likely to choose (especially novel) information that supports rather than opposes their attitude (Sawicki et al., 2011). Similarly, message recipients with doubt process information that agrees with their attitude more than information that disagrees with their attitude. In contrast, when message recipients are confident, disagreeable information is processed more deeply than agreeable information (Clark & Wegener, 2013). In the group identification domain, Clarkson, Smith, Tormala, and Dugan (2017) have shown that people with uncertain attitudes identify with groups who provide clear attitudinal norms that support their attitudes (but not with groups who do not hold attitudes that would strengthen one’s uncertain attitude). In each of these cases, people can compare their own attitudes with those of others, and specifically seek out information or people that bolster attitudes held with doubt. Therefore, these paradigms may have inadvertently created an opportunity to consider the implications of available information or actions for the properties of one’s attitudes. For instance, the Clarkson et al. (2017) participants were reminded that “groups can have an important influence on people” before indicating identification with a group that did or did not hold relevant attitudes. Such instructions might have prompted participants to consider the possible implications of the group’s attitudinal positions for their own attitude (un)certainty.
These preferences for support over opposition of one’s attitude have also extended to actions taken in expressing support for the attitude. For instance, Gal and Rucker (2010) found that people whose confidence had been shaken were more likely to put effort into persuading others (or intended to advocate more). Consistent with the idea that these individuals were attempting to increase confidence through advocacy, these tendencies were decreased when the need for validation was decreased (i.e., when participants had previously affirmed their identity) or when the advocacy was likely to meet with opposition.
Metacognitive Reflection as a Moderator of Strength Versus Bolstering Patterns
A key missing piece of this literature is identification of moderators when traditional strength patterns are obtained (doubt undermining impact of attitudes) rather than bolstering patterns (doubt leading the attitude to have just as much or more impact as an attitude held with certainty). We believe that one such moderator entails the extent to which people engage in metacognitive reflection. Metacognition refers to “thinking about thinking,” and a key metacognitive construct in social psychology involves the confidence one has in cognitions (Petty, Briñol, Tormala, & Wegener, 2007). The distinction between cognition and metacognition parallels that between primary and secondary cognition. That is, if a person has a primary cognition toward a target (e.g., “taxing junk food is a good idea”), confidence or doubt in that thought would be the secondary cognition (i.e., the metacognition). Thus, the term metacognitive reflection pertains to one’s consideration of how a setting, thought, or action would affect one’s metacognitions such as certainty.
Attention to metacognitions can be juxtaposed with default attention to the target of the cognitions. In many if not most settings, social perceivers attempt to form an impression of a particular target, and attitudes toward that target form input to (primary) thoughts, judgments, or behaviors toward the target. In such impression-formation settings, attitude strength might naturally determine the strength of the attitude’s input to one’s reactions toward the target. Thus, these impression-formation settings would tend to produce a strength pattern, such as attitudes held with certainty having greater impact than attitudes held with uncertainty. When attention is drawn to the metacognitive properties of an attitude, however, then attention is drawn to how the target or reactions toward the target can influence the metacognition. For example, when reflecting on the confidence (certainty) versus doubt (uncertainty) with which one holds an attitude, the situation turns into a validity-testing situation rather than an impression-formation situation. As such, the person’s experiences are more likely to be driven by potential shifts in one’s level of doubt or confidence. That is, targets or reactions that serve to reduce doubt or create confidence may create more favorable experiences than targets or reactions that serve to increase doubt or reduce confidence. When reactions to a target are driven by how that target might strengthen the attitude, bolstering patterns may occur, where attitudes held with doubt have greater influence than attitudes held with confidence.
Such outcomes of metacognitive reflection would seem consistent with Festinger’s (1954) social comparison theory and assumptions that people are motivated to hold “correct” attitudes (Petty & Cacioppo, 1986). That is, when lacking validity (or validity remains uncertain), people should be motivated to engage in actions that enhance the attitude’s perceived correctness and should react favorably to contexts that make them feel better about the validity of their attitude. Thus, a key distinction would contrast such validity-testing situations (where bolstering effects may occur) with many (default) impression-formation situations in which one’s attitude forms input to a thought or action (and strength effects should be more likely).
Looking back at previous research, the likelihood of metacognitive reflection may distinguish between cases in which strength versus bolstering patterns have been observed. When strength patterns have been observed, participant focus seems likely to be on the target of the judgment or behavior. When bolstering patterns have been observed, however, it seems more likely for participant focus to be on their own attitude and its validity. For example, bolstering has occurred when the person is directly seeking or processing information on the topic of one’s attitudes (Clark & Wegener, 2013; Sawicki et al., 2011). Bolstering has also occurred when affiliating with groups with specific attitudinal norms (Clarkson et al., 2017), or when behavior directly seeks support for one’s attitude (Gal & Rucker, 2010). Each of these settings provides information useful for evaluating the validity of one’s attitude that can be used to consider the implications of that support or opposition for the attitude itself and its validity. Of course, this interpretation of the previous literature is just that—an interpretation. Better evidence for this conjecture would provide more direct tests of conditions that we believe should produce strength versus bolstering patterns.
Strength and Bolstering in the Context of Attitude Similarity and Attraction
One of the most consistent influences of attitudes is on attraction to others with similar attitudes (Byrne, 1961). Consistent with balance theory (Heider, 1958), people generally like and prefer to interact with others who agree rather than disagree with them (Montoya & Horton, 2013). The attitude similarity domain should be a particularly interesting and apt setting to examine differences between strength of attitudinal input and motives to bolster weak attitudes. Despite a long history, however, little research on attitude similarity and attraction has examined attitude strength.
One attempt to do so examined attitude importance. Using the multiattitude bogus stranger paradigm, Byrne, London, and Griffitt (1968) concluded that attitude importance only moderated attitude effects on attraction when important and unimportant attitudes conflicted (and participants presumably had to decide which attitudes mattered more to them). In the current research, however, we did not use the bogus stranger paradigm. Instead, we provided information only about a single attitude held by the target. In that context, we believed that attitude certainty could moderate attitude effects on attraction (and we found that they could, even when controlling for attitude importance). This may be, in part, because people forming impressions of targets in the current single-attitude setting do not have to aggregate across many attitudes (similarly, when focused on metacognition, they can do so for certainty in a single attitude rather than overall certainty across many attitudes). Another potential key difference between the old bogus stranger work and the current research is that, unlike topic importance, certainty may be quite directly connected to the evaluation in memory (e.g., see Petty, Briñol, & DeMarree, 2007). Thus, strength effects of attitude certainty would constitute an advance in understanding similarity effects on attraction.
In “baseline” conditions, little information other than the target’s attitude was presented to participants, and they simply formed an impression of the target. This seemed to fit the default contexts described earlier in which perceivers’ attitudes serve as core input to impressions of target likeability. However, if a perceiver instead uses the target’s attitude as information about the validity of the perceiver’s own attitude, this represents exactly the kind of social comparison Festinger (1954) discussed. When engaging in this attitudinal social comparison from a position of high perceived validity, another’s (dis)agreement would seem of little consequence. However, from the vulnerable position of doubting one’s attitude, learning that another person agrees might be considerably more reassuring than disagreement. If so, when thinking of the other person’s opinion as an opportunity to test validity of one’s own opinion, attitude (dis)similarity might matter more for attraction when uncertain rather than certain about one’s attitude.
The Current Research
In the initial studies, we examined the conditions in which certainty might increase attitude impact on judgments of the target (strength effects) and conditions in which uncertainty might increase attitude impact on attraction to the target (bolstering effects). In Studies 1A and 1B, we examined influences of measured attitude certainty. Study 1A examined (default) impression-formation settings in which participants were given information about a target’s position on an issue and were asked to report liking for the target. Study 1B examined settings in which participants were asked to consider the implications of the target’s attitude for their own attitude (metacognitive reflection). In a second pair of studies using the same default (impression-formation) and metacognitive-reflection conditions, we manipulated certainty through an incidental certainty priming procedure (Studies 2A and 2B). In all studies, we expected an Attitude X Certainty interaction, however, the direction of the effect (strength versus bolstering) should depend on evaluative focus. We expected strength patterns to occur in default (impression-formation) conditions (Studies 1A and 2A)—with greater certainty leading attitudes to serve as stronger input to judgments of target likeability. Because the target’s position was the only attitudinal information provided, we expected the primary implication of the target’s view for the participant’s own view—in metacognitive-reflection conditions (Studies 1B and 2B)—would be to provide attitudinal social comparison information (supporting or questioning the validity of the participant’s attitude). Thus, in metacognitive-reflection conditions (Studies 1B and 2B), we expected bolstering patterns to occur—with greater uncertainty leading to stronger influences of attitudes on target liking.
Though Studies 1A to 2B provided evidence consistent with the hypothesized strength versus bolstering patterns, they had not randomly assigned participants to default (impression-formation) versus metacognitive-reflection conditions. Therefore, Study 3 examined the full experimental design by manipulating both certainty and metacognitive reflection. In these studies, the newest and most marked part of the pattern was the change in impact of uncertain attitudes on evaluations of the target. Thus, in Study 4, we sought to provide initial mechanistic evidence that the impact of uncertain attitudes under metacognitive-reflection conditions involved viewing the target’s attitude as helping the participants to feel better about their own position (i.e., to increase attitude certainty).
Studies 1A and 1B
In Studies 1A and 1B, we measured attitude certainty and asked participants to form an impression of and rate the target (Study 1A), or asked participants to consider the implications of the target’s attitude for their own attitude (metacognitive reflection; Study 1B). We expected traditional strength patterns to occur when respondents were focused on forming an impression of the target, and we expected bolstering patterns to occur when respondents engaged in metacognitive reflection. Variations across studies in control variables or other features are described.
Method
Participants and design
Across two data collections, 190 undergraduates (Study 1A N = 139; Study 1B N = 51) at Midwestern universities participated in exchange for partial course credit toward an introductory psychology class. 1 The design consisted of measures of participants’ attitude favorability accompanied by measures of attitude certainty. Because various alternative strength-related properties of attitudes might covary with measured certainty, the studies also included measures of topic importance, knowledge, and ambivalence—enabling analyses that controlled for these variables as well as attitude extremity (calculated as discrepancy from the scale midpoint). Attitude certainty might also be correlated with a more general sense of self-certainty (though previous attitude bolstering effects have occurred based on topic-specific uncertainty controlling for general self-certainty; Clarkson et al., 2017). To examine the topic specificity of the effects, the studies also included measures of certainty in an alternative attitude. If effects were due to more general self-certainty, measures of certainty in an alternative attitude would presumably reproduce the same pattern of effects as certainty in attitudes toward the same topic as the target’s attitude.
In Study 1A, participants were asked to form an impression and make judgments of the target. Study 1A included two focal targets: one expressing support for marriage equality and one expressing support for building additional nuclear power plants. Study 1B was identical to Study 1A, except that Study 1B participants were asked to consider the implications of the target’s attitude for one’s own stance (metacognitive reflection). Also, Study 1B used a single target expressing support for polyamory.
Procedure
Participants completed computerized materials using MediaLab 2008 (Jarvis, 2008). In each case, the study was initially presented as an investigation of general impressions toward issues, events, and people. Participants expressed attitudes toward the focal topic and any control issues, and reported attitude certainty, issue importance, issue knowledge, and attitude ambivalence regarding the various issues. Participants then received an instruction screen prior to receiving one target (in Study 1B) or two targets (Study 1A) who held a specific attitude on an assessed topic. Before encountering the target, all participants in both studies were told, “On the following screens, you will be presented with information about another person and you will be asked to rate them.” In Study 1B, they were also asked to engage in metacognitive reflection. That is, they were told, “When you evaluate them, please think about their opinion and what that opinion might mean for your own stance.”
Predictor variables
Attitude favorability
Participants in Study 1A reported attitude favorability toward each target issue (nuclear power and marriage equality) using two 7-point scales (nuclear power: 1 = worthless, bad; 7 = worthwhile, good; marriage equality: 1 = negative, bad; 7 = positive, good). The attitude scales hung together well (α = .86 and .98, respectively), and were averaged into indices of participant attitude favorability (nuclear power attitude M = 4.08; SD = 1.41; marriage equality attitude, M = 5.93; SD = 1.77). In Study 1B, participants reported attitudes toward polyamory, using two 7-point scales (1 = bad, negative; 7 = good, positive; M = 2.62, SD = 1.58, α = .92).
Attitude certainty
In both studies, attitude certainty was measured using a 7-point scale (“I am confident in my attitude regarding [issue]”; 1 = not at all; 7 = very much; in Study 1A, nuclear power certainty: M = 4.18; SD = 1.90; marriage equality certainty: M = 6.31; SD = 1.13; in Study 1B, polyamory certainty: M = 5.51; SD = 1.78).
Dependent measure
Target evaluation
Evaluation of each target was measured using 5-point scales. In Study 1A, participants learned that “[Target name] supports [issue],” and were asked, “How much do you like [Target name]?” 1 = not at all; 5 = very much; “[Target name] is” 1 = a bad person; 5 = a good person. The two items were reasonably correlated (nuclear power α = .65 and marriage equality α = .89) and averaged to form indices of target evaluation (nuclear power target evaluation: M = 3.08; SD = 0.54; marriage equality target evaluation: M = 3.88; SD = 1.03). Each target was given a different (counterbalanced) name (George White or John Roberts). Participants in Study 1B evaluated “Greg Frank” who “is involved in a polyamorous relationship” (Greg Frank is . . . 1 = a bad person; 5 = a good person; M = 2.82, SD = 0.84).
Results
Regression analyses were performed separately for each target evaluation using mean-centered attitude favorability, mean-centered attitude certainty, and their interaction as predictors (Aiken & West, 1991). For the issue of nuclear power, a main effect of attitude emerged, b = .12, t(135) = 3.71, p < .001. This effect was qualified by the predicted Certainty × Attitude interaction, b = .04, t(135) = 2.59, p = .01. 2 Attitudes had greater impact on target liking when attitudes were held with relatively high certainty (1 SD above the certainty mean), b = .19, t(135) = 5.76, p < .0001, than when attitudes were held with relatively low certainty (1 SD below the certainty mean), b = .04, t(135) = 0.88, p = .38. For the issue of marriage equality, a main effect of attitude emerged, b = .30, t(135) = 6.60, p < .001. This effect was qualified by the predicted Certainty × Attitude interaction, b = .10, t(135) = 2.76, p = .007. Attitudes were a significantly stronger predictor of target liking when attitudes were held with relatively high certainty (1 SD above the certainty mean), b = .37, t(135) = 7.52, p < .0001, rather than low certainty (1 SD below the certainty mean), b = .19, t(135) = 2.87, p = .005. Thus, in Study 1A, uncertainty in one’s attitudes reduced impact of those attitudes on evaluations of the target (see Study 1 of Sawicki, 2013, for another measured attitude confidence study that shows attitude strength effects on attraction). 3
For Study 1B, regression analyses were conducted on the target evaluation with mean-centered attitude favorability, mean-centered attitude certainty, and their interaction as predictors. A main effect of attitude emerged, b = .22, t(47) = 2.70, p = .01. The more a person supported polyamory, the more he or she liked the polyamorous target. Of critical importance to the current research, however, the predicted Certainty × Attitude interaction was also significant, b = −.11, t(47) = −2.09, p = .04. When relatively uncertain (1 SD below the certainty mean), attitudes significantly predicted target evaluation, b = .41, t(47) = 2.78, p = .008. However, when relatively certain (1 SD above the certainty mean), attitudes did not predict evaluations of the target, b = .06, t(47) = 0.80, p = .43. Thus, when people were focused on implications of the target for their own attitudes (metacognitive reflection), attitude uncertainty rather than certainty increased the use of the attitude in evaluating the target person. 4 With the addition of a simple one sentence instruction designed to increase metacognitive reflection, effects of attitude certainty were completely reversed from a traditional strength pattern to a bolstering pattern.
The key test for Studies 1A and 1B was that metacognitive reflection would shift traditional attitude strength effects to become bolstering effects. A fixed effect meta-analysis compared the average of the two Certainty × Attitude effects from Study 1A (r = .224) with the Certainty × Attitude effect from Study 1B (r = −.291). The analysis showed a significant difference, Q(1) = 9.61, p < .002. Though the respective strength and bolstering effects each held when we controlled for a number of variables that might have been correlated with attitude certainty, the evidence for strength or bolstering from Studies 1A and 1B was not causal because certainty was not manipulated. Therefore, Studies 2A and 2B manipulated certainty.
Studies 2A and 2B
Studies 2A and 2B used very similar methods to Studies 1A and 1B. Both studies measured attitudes toward the focal issue and manipulated certainty through a general priming procedure (as in Petty, Briñol, & Tormala, 2002) prior to exposing participants to a target person who took a position on the focal issue. Like Studies 1A and 1B, Study 2A simply asked participants to form an impression of and rate the target, whereas Study 2B asked participants to consider the implications of the target’s attitude for their own stance (metacognitive reflection). As in Studies 1A and 1B, we expected a strength pattern when respondents focused on forming an impression of the target (in Study 2A), and we expected a bolstering pattern under metacognitive reflection (in Study 2B).
Method
Participants and design
Across two data collections, 158 undergraduates (Study 2A N = 103; Study 2B N = 55) at Midwestern universities participated in exchange for partial course credit toward an introductory psychology class. Studies 2A and 2B were identical to Studies 1A and 1B with the following exceptions. Study 2A used a new topic (homeschooling), and participant certainty and target position were manipulated. Therefore, for Study 2A, the predicted effect became a Certainty × Attitude × Target Position interaction. Because Study 2A focused participants on forming an impression of the target, the expectation was that certainty would strengthen attitude effects (with opposing attitude effects when the target takes opposite positions). Like Study 1B, Study 2B used a single target and focused participants on the implications of the target’s attitude for their own stance (metacognitive reflection). Thus, for Study 2B, the expectation was that uncertainty would lead to stronger attitude effects on target evaluations.
Procedure
Participants were told that they would be asked to provide their attitudes about a variety of issues, to recall some memories from their past, and to provide their impression of another person. In Study 2A, participants first expressed their attitude about various issues, including the target issue of homeschooling. In an ostensibly unrelated task, participants were randomly assigned to recall two events in the past when they felt either doubtful or confident (cf. Petty et al., 2002). Participants then evaluated a target person who either supported or opposed the homeschooling of children. Because impression instructions focused participants on the target, attitude strength effects were expected. Study 2B employed the same basic procedure but used the attitude topic of building nuclear power plants (that had previously shown strength effects under impression-formation conditions in Study 1A). Because the instructions in Study 2B focused participants on metacognitive reflection, we expected bolstering effects with uncertainty increasing attitude effects on target evaluation.
Predictor and independent variables
Attitude favorability
Before the certainty manipulation, in Study 2A participants reported attitudes toward homeschooling children, using two 9-point scales (1 = bad, definitely opposed; 9 = good, definitely in favor). Scores were highly correlated (α = .95) and were averaged to form an index of attitude favorability (M = 4.05, SD = 1.82). In Study 2B, participants reported attitudes toward building additional nuclear power plants, using a 9-point scale (1 = not worthwhile; 9 = very worthwhile; M = 5.21, SD = 1.89).
Certainty
In both studies, certainty was manipulated using a priming procedure (cf. Petty et al., 2002). Participants were asked to recall two experiences when they previously had felt doubtful or confident. Although participants wrote about an experience of doubt or confidence unrelated to the focal attitude topic, past research has shown that people can attribute this (un)certainty to a specific attitude or thought in mind just prior to the confidence prime (Study 3 of Petty et al., 2002).
Dependent measures
Target evaluation
In Study 2A, before evaluating the target person, participants were informed that the task was designed to evaluate another person based on their endorsement of a social issue, just as people evaluate politicians before an election. Participants were then randomly presented with the target (Chris Sawyer) who either supported or opposed the homeschooling of children. Participants evaluated the target using three 5-point scales (i.e., How much do you like Chris? How much would you like to meet Chris? How much would you enjoy talking to Chris at a party? 1 = not at all; 5 = very much). Scores were correlated (α = .73) and averaged to form an index of target evaluation (M = 2.77, SD = 0.77). Similarly, in Study 2B, after reading the same instructions with the addition of a metacognitive reflection (see Study 1B), participants evaluated a person (Taylor Johnson), who supported the building of additional nuclear power plants, using the same three 5-point scales as in Study 2A. Scores were correlated (α = .72) and were averaged to form an index of target evaluation (M = 3.00, SD = 0.76).
Results
Expressed certainty manipulation check
To assess whether participants’ responses to the certainty manipulation differed in content, each response was coded by a research assistant unaware of the participant’s condition. The coder rated each essay for the degree of certainty expressed (i.e., “What the participant wrote conveys . . .”) using a 5-point scale (−2 = extreme doubt; 2 = extreme confidence). Participants were instructed to describe two episodes of doubt or confidence, and thus received two ratings that were correlated in both studies (Study 2A, α = .92; Study 2B, α = .98) and averaged to create an index of expressed certainty. In both studies, there were significant certainty condition differences in the described confidence; Study 2A, t(100) = 17.77, p < .0001, 5 and Study 2B, t(50) = 18.70, p < .0001. People in the doubt condition conveyed significantly less certainty (Study 2A M = −0.97, SD = 0.67; Study 2B M = −1.52, SD = 0.70) than people in the confidence condition (Study 2A M = 1.20, SD = 0.56; Study 2B M = 1.75, SD = 0.53). 6
Target evaluation
In Study 2A, regression analyses were conducted on the index of target evaluation with mean-centered attitude favorability, certainty condition, target’s position, and all possible interactions as predictors. The expected Certainty × Attitude × Target Position interaction was significant, b = .13, t(95) = 3.17, p = .002. When the target opposed homeschooling, the Certainty × Attitude interaction was significant, b = −.17, t(48) = −3.05, p = .004. When describing confidence after reporting attitudes, the more favorable the participants’ attitude toward homeschooling, the less they liked a target who opposed homeschooling, b = −.33, t(48) = −3.81, p < .001. When describing doubt after reporting attitudes, the attitudes did not predict reactions to the anti-homeschooling target, b = .01, t(48) = 0.18, p = .86. Results were also supportive of strength effects when the target supported homeschooling. The Certainty × Attitude interaction did not reach conventional levels of significance, b = .09, t(47) = 1.47, p = .15, but the pattern of simple slopes was in the strength direction. When describing confidence after reporting attitudes, the more favorable the participants’ attitude toward homeschooling, the more they liked a target who supported homeschooling, b = .24, t(47) = 2.62, p = .01. When describing doubt after reporting attitudes, the attitudes did not predict reactions to the pro-homeschooling target, b = .07, t(47) = 0.92, p = .36.
In Study 2B, we regressed the target evaluation on mean-centered attitude favorability, certainty condition, and their interaction. The predicted Certainty × Attitude interaction was significant, b = −.11, t(51) = −2.25, p = .03. When describing doubt after reporting attitudes, the more favorable the participants’ attitude toward nuclear power, the more they liked a pro-nuclear-power target, b = .25, t(51) = 3.42, p = .001. However, when describing confidence after reporting attitudes, the attitudes did not predict reactions to the pro-nuclear-power target, b = .02, t(51) = 0.26, p = .79.
The key test for Studies 2A and 2B was again that metacognitive reflection would shift traditional attitude strength effects toward bolstering effects. A fixed effect meta-analysis compared the size of the two Certainty × Attitude (strength) effects from Study 2A (rs = .40 and .21, respectively) with the Certainty × Attitude effect from Study 2B (in the direction of a bolstering effect; r = −.30). The analysis showed a significant difference between the strength and bolstering patterns, Q(1) = 13.00, p < .0003, replicating the difference between Studies 1A and 1B. As discussed by Fabrigar and Wegener (2016), another way to examine replication is to examine how the addition of the new data changes the meta-analytic support for the effect and the robustness of that support to any existing or future lack of support for that pattern (indexed through changes in a Failsafe number of studies, Rosenberg, 2005, or the number of participants’ worth of data such studies would represent). Adding Studies 2A and 2B to the results of Studies 1A and 1B increased the test statistic for the meta-analytic difference between strength effects (combining Studies 1A and 2A) and bolstering effects (combining Studies 1B and 2B), Q(1) = 22.13, p = .000002. Also, the robustness of the pattern to data averaging no difference between impression-formation and metacognitive-reflection conditions is improved by adding the data from Studies 2A and 2B (Failsafe from Studies 1A and 1B = 2.62 studies × 186 participants/study = 487 participants’ worth of data averaging no difference; Failsafe number when adding data from Studies 2A and 2B = 14.93 studies × 169 participants/study = 2,523 participants). 7
Two studies manipulating certainty provided additional evidence that strength effects under default impression-formation conditions can be shifted to bolstering effects when social perceivers engage in metacognitive reflection. The comparisons between Studies 1A and 1B and between 2A and 2B were each the first of their kind. If each strength or bolstering pattern had been moderately sized (r = .3), given the respective sample sizes, the estimated power for the moderation by metacognitive reflection would have been over .95 for each pair of studies. However, by themselves the tests of bolstering (1B and 2B) had small sample sizes and were therefore underpowered (even if the average observed bolstering effect of approximately r = −.3 were the population value, the power estimates for tests of bolstering would fall between .58 and .61). Thus, follow-up research using a larger sample can help increase confidence in the bolstering effects while also supporting the moderation by metacognitive reflection. Another weakness is that participants were not randomly assigned to impression-formation or metacognitive-reflection conditions. Therefore, in Study 3 we manipulated evaluative focus along with the experience of doubt versus confidence in relation to the attitudinal position.
Study 3
Study 3 differed from the previous studies in a few ways: First, the certainty prime manipulation was modified to center on the specific attitude domain as opposed to general certainty. Second, a new attitude object was used, political conservatism. Political conservatism was chosen to tailor the certainty manipulation more closely to the attitude. It might be difficult for people to generate confident and doubtful experiences with particular attitudes (e.g., nuclear power), but it should be easier for an individual’s more general political approach (similar to recent work manipulating confidence in a value, for example, Blankenship, Wegener, & Murray, 2012). It was hoped that tailoring the certainty manipulation in this way would also make it less vulnerable to interpretation as producing results based on more general self-related variables. Related to this concern, the study also examined the possible role of more global self-views by examining self-esteem (Robins, Hendin, & Trzesniewski, 2001), self-concept clarity (Campbell et al., 1996), and need to belong (Leary, Kelly, Cottrell, & Schreindorfer, 2012) as control variables. 8
Method
Participants and design
Two hundred seven undergraduates 9 participated in exchange for partial course credit in introductory psychology classes. Political conservatism was measured, and confidence in one’s political approach and evaluative focus were manipulated between participants to predict evaluation of a conservative target.
Procedure
The study was presented as an investigation of general impressions toward issues, events, and people. Participants were told that they would provide their attitudes about a variety of issues, to recall some memories from their past, and to provide their impression of another person. After evaluating the target other, participants completed measures of the self-related variables.
Predictor and independent variables
Political conservatism
Before the certainty manipulation and target evaluation measure, participants completed a political conservatism scale modified from Pratto, Sidanius, Stallworth, and Malle (1994). For the first three items, participants rated the extent to which they identified as a conservative (1 = very liberal; 7 = very conservative) regarding foreign, economic, and social issues. The final item assessed political conservatism in general (Overall, I take a . . . 1 = liberal approach; 6 = conservative approach) on a 6-point scale. If a participant endorsed the lowest 3 points of the final item (i.e., 3 or lower), that participant was asked to write about his or her liberal approach in the confidence/doubt manipulation. If a participant endorsed the highest 3 points (i.e., 4 or higher), that participant was asked to write about his or her conservative approach in the confidence/doubt manipulation. All four items were standardized and averaged to form an index of political attitudes (α = .91).
Political attitude certainty
After indicating their level of political conservatism, participants completed a modified certainty manipulation about their political approach. That is, depending on the certainty condition, participants were asked to write about two past experiences of either doubt or confidence in their (liberal/conservative) approach. This modification was employed to focus participants on (un)certainty in their political attitudes, rather than a more general experience of (un)certainty. Although the content of the confidence manipulation was contingent upon a person’s reported political attitude, participants were randomly assigned to write about certainty or doubt.
Evaluative focus
After completing the certainty manipulation, participants read evaluative focus instructions identical to those used in the previous studies (impression-formation vs. metacognitive reflection).
Dependent measure
Target evaluation
After reading evaluative focus instructions, participants evaluated a person (Taylor) who “supports a conservative approach” using a 5-point scale (i.e., How much do you like Taylor? 1 = not at all; 5 = very much; M = 3.15, SD = 0.72). We used this single item because of norms about not mixing politics into polite conservation that could have affected responses to the other items from our previous studies.
Results
Political attitude certainty check
As in previous studies, a t test examined differences in expressed certainty across certainty conditions. A coder unaware of the condition rated each essay for the degree of certainty expressed (i.e., “What the participant wrote conveys . . .”) using a 5-point scale (−2 = extreme doubt in their political attitude; 2 = extreme confidence in their political attitude). The ratings of the two essays were correlated (α = .80) and averaged to create an index of expressed attitude certainty (M = 0.22, SD = 1.39). As expected, a significant mean difference in certainty emerged, t(176) = −6.43, p < .0001. People in the doubt condition described episodes of significantly less certainty (M = −0.33, SD = 1.31) than people in the confidence condition (M = 0.88, SD = 1.18). 10
Target evaluation
Centered regression analyses were conducted on the index of target liking with attitude favorability, certainty condition, evaluative focus, and all possible interactions as predictors. A main effect of attitude favorability emerged, b = .39, t(199) = 7.62, p < .001. The more a participant endorsed a conservative ideology, the more he or she liked the conservative target. This main effect was qualified by the predicted Certainty × Attitude × Evaluative Focus interaction, b = −.13, t(199) = −2.49, p = .013, r = −.17, 95% confidence interval (CI) = [−0.305, −0.036]. When participants were focused on forming an impression of the target, the simple Certainty × Attitude interaction was in the direction of a strength effect but did not reach conventional levels of significance, b = .12, t(99) = 1.78, p = .08. When people recalled past confident experiences regarding political attitudes, political attitudes significantly predicted liking, b = .60, t(99) = 5.75, p < .0001, and marginally more than when participants recalled past doubtful events about political attitudes, b = .35, t(99) = 3.81, p = .0001 (see Figure 1A). When participants engaged in metacognitive reflection, the simple Certainty × Attitude interaction was in the direction of bolstering but again did not quite reach significance, b = −.13, t(100) = −1.75, p = .08. Political attitudes significantly predicted liking when participants recalled past doubtful events, b = .44, t(100) = 4.15, p < .0001, but not when people recalled past confident experiences, b = .18, t(100) = 1.63, p = .11 (see Figure 1B).
Predicted values for target evaluation based on relative attitude favorability (−1 SD or +1 SD), certainty condition (doubt or confidence), and evaluative focus instructions (impression-formation or metacognitive reflection): Study 3.
Study 3 provided experimental evidence that a person’s evaluative focus when encountering another person determines how (un)certainty impacts the similarity–attraction link. When participants focused on forming an impression of the target, they used their attitude as a way to determine the likeability of the target and stronger attitudes served as stronger input to those judgments. The opposite pattern emerged when people reflected on the implications of (dis)agreement for their own attitudes. When engaging in metacognitive reflection, a bolstering pattern emerged, and certainty tended to attenuate similarity effects on attraction as in Studies 1B and 2B. That is, experiencing certainty made validation by the other less needed and less impactful than when experiencing uncertainty. This pattern is also quite interesting for the attraction literature, in that validation via agreement has recently been resurrected as a potential reason for attitude similarity effects on attraction (without considering differences in certainty of social perceivers’ attitudes; Singh et al., 2017).
Study 3 presented an experimental manipulation of evaluative focus that replicated the previous meta-analytic comparisons between impression-formation (target-focus) and validity-testing (metacognitive-reflection) settings. Meta-analytically combining the results of Study 3 with those from the previous studies again increased the meta-analytic support for the Certainty × Attitude × Evaluative Focus interaction—across the three studies, Q(1) = 27.92, p < .000001, Failsafe number = 28.12 studies × 180 participants/study = 5,050 participants. Because the Certainty × Attitude patterns varied in strength across studies and were only marginal in Study 3, we also separately examined the strength versus bolstering patterns across studies. As summarized in Table 1, in impression-formation conditions (in Studies 1A, 2A, and impression-formation conditions of Study 3), the meta-analytic combination of effect sizes associated with the Certainty × Attitude interaction was highly significant. The test of heterogeneity of strength effects across studies was nonsignificant (p > .5), and the CIs were highly overlapping (see Table 1). The combined index of the strength pattern was statistically stronger (lower p value) than the index based only on Studies 1A and 2A, and adding the impression-formation conditions of Study 3 also increased robustness of the strength-pattern Certainty × Attitude interaction to future null data (comparing the Failsafe number from Studies 1A and 2A with the Failsafe number when adding impression-formation condition data from Study 3).
Effect Sizes and 95% Confidence Intervals for Attitude × Certainty Effects Across Studies.
Note. ImpForm = Impression formation. MetaRefl = Metacognitive reflection. Positive effect sizes reflect stronger attitude effects with certainty compared with uncertainty (strength patterns), and negative effect sizes reflect stronger attitude effects with uncertainty compared with certainty (bolstering patterns).
When research participants were asked to focus on implications of the other’s view for their own stance (metacognitive reflection), the overall pattern fell in the opposite direction. The meta-analytic summary across samples was significant. The test of heterogeneity of bolstering effects across studies was again nonsignificant (p > .3), and the CIs were highly overlapping (see Table 1). This combined index of the bolstering pattern was also statistically stronger than that based only on Studies 1A and 2A, and adding the metacognitive-reflection conditions of Study 3 also increased robustness of the bolstering-pattern Certainty × Attitude interaction to future null data (comparing Failsafe number from Studies 1B and 2B to the Failsafe number when adding the metacognitive-reflection condition from Study 3).
Though Studies 1B, 2B, and 3 demonstrated that attitudes associated with doubt can have relatively strong influences on related judgments, they did not directly address the proposed reasons for the shift in relative impact of doubt-associated attitudes across target-focus versus metacognitive-reflection conditions. In part, the assumption is that participants view the (dis)agreeing other as capable of influencing their own level of attitude uncertainty. Thus, in Study 4, we aimed to more directly examine the perceived effects on attitude uncertainty that may be responsible.
Study 4
The shift in impact of uncertain attitudes in the previous studies is presumed to be driven by expectations about the influence of (dis)agreement on one’s current level of attitude uncertainty. That is, disagreement is unattractive because the person would not help alleviate one’s doubt, whereas agreement is attractive because the person could resolve uncertainty and strengthen the attitude. If these assumptions are correct, expectations about potential change in attitude uncertainty should determine target evaluations but primarily when people engage in attitudinal social comparison (metacognitive reflection). When the connection between the target and attitude uncertainty is less apparent (when merely forming an impression of the target), perceived effects of agreement on participant uncertainty should be weaker. To hone in on the mechanism behind increased impact of attitudes associated with uncertainty, all participants in Study 4 were primed with doubt after reporting their attitudes.
Method
Participants and design
One hundred sixty-seven undergraduates volunteered to participate at a university library. Favorability of attitudes toward banning alcohol-related advertisements was measured, and evaluative focus was manipulated between participants to predict evaluation of a pro-ban target.
Procedure
Participants first expressed their attitude about a variety of issues, including the target issue of banning alcohol-related advertisements. In an ostensibly unrelated task, all participants were asked to recall previous events of doubt. Participants then encountered evaluative focus instructions (impression-formation or metacognitive reflection) before indicating liking for a person who supported the banning of alcohol-related advertisements. Finally, participants indicated the extent to which meeting with the target would impact their certainty about banning alcohol-related advertisements.
Predictor and independent variables
Attitude favorability
Participants first reported attitudes toward banning alcohol-related advertisements using two 9-point scales (1 = bad; not useful; 9 = good; very useful; M = 3.49, SD = 1.58).
Evaluative focus
After completing the doubt prime, participants read instructions identical to those used in the previous studies that were designed to manipulate their evaluative focus (to merely form an impression of the target or to think about what the other person’s opinion might mean for their own).
Dependent measures
Target evaluation
After reading evaluative focus instructions, participants indicated liking for a pro-ban person using three 5-point scales (i.e., How much do you like Taylor? How much would you like to meet Taylor? How much would you enjoy talking to Taylor at a party? 1 = not at all; 5 = very much). Liking scores were only modestly correlated (α = .55) but were averaged to form an index of attraction (M = 2.77, SD = 0.66).
Expected change in attitude certainty
After reporting attraction, participants responded to the question, “How much would meeting Taylor impact your opinion regarding banning alcohol-related ads? It would . . .” on an 11-point scale (−5 = decrease my certainty about this issue; +5 = increase my certainty about this issue; M = 0.79, SD = 1.97).
Results
Target evaluation
Centered regression analyses were conducted on the index of target evaluation with attitude favorability, evaluative focus condition, and their interaction as predictors. Unsurprisingly, a main effect of attitude emerged, b = .16, t(163) = 5.21, p < .0001. The more a person supported the banning of alcohol advertisements, the more he or she liked the pro-ban target. The predicted Attitude × Evaluative Focus interaction did not reach significance, b = .05, t(163) = 1.72, p = .09, r = .13, 95% CI = [−0.019, 0.28], but the pattern of simple slopes was in the predicted direction. When doubtful and engaging in metacognitive reflection, attitudes more strongly predicted attraction, b = .21, t(163) = 4.93, p < .0001, than when participants were doubtful and merely forming an impression of the target, b = .11, t(163) = 2.45, p = .02.
Expected change in attitude certainty
Centered regression analyses were conducted on the index of expected change in attitude certainty with attitude favorability, evaluative focus condition, and their interaction as predictors. Only a significant main effect of attitude emerged, b = .16, t(163) = 3.33, p = .001. The more a person supported the ban, the more he or she expected that meeting the pro-ban target would increase their own attitude certainty.
Mediation
Might expectations about changes in attitude certainty account for effects of evaluative focus on target evaluation for people who are doubtful? We first ran a centered regression analysis on the index of target evaluation with expected change in attitude certainty, evaluative focus, and their interactions as predictors. There was a main effect of expected change in attitude certainty on liking, b = .05, t(163) = 2.21, p = .03. The more a person expected the target to increase attitude certainty, the more that person liked the target. There was also a main effect of Evaluative Focus, b = −.44, t(163) = −2.47, p = .02. Liking for the target was greater when forming an impression of the target than when engaged in metacognitive reflection. Most importantly, these main effects were qualified by a significant Expected Change in Attitude Certainty × Evaluative Focus interaction, b = .07, t(163) = 2.64, p = .009. When people engaged in metacognitive reflection, the more a person expected the target to increase certainty about banning alcohol ads, the more favorably that person evaluated the target, b = .12, t(163) = 3.33, p = .001. When people were focused on judging the target other, however, expectations about the target’s impact on attitude certainty did not predict target evaluations, b = −.01, t(163) = −.31, p = .75.
We then assessed whether this Expected Change in Attitude Certainty × Evaluative Focus interaction might account for the Attitude × Evaluative Focus pattern. We ran a simultaneous regression with both interaction terms and all main effect predictors in the same statistical model. If the Expected Change in Attitude Certainty × Evaluative Focus interaction remains significant in this regression but the Attitude × Evaluative Focus interaction is reduced in its magnitude, this suggests that the effect of attitude (similarity) on attraction may be at least partly due to expectations about the target’s ability to increase certainty. This analysis revealed that the Attitude × Evaluative Focus interaction dropped in magnitude; from b = .05, t(163) = 1.72, p = .09, to b = .03, t(163) = 1.09, p = .28. In contrast, the Expected Change in Attitude Certainty × Evaluative Focus interaction remained significant, b = .05, t(163) = 2.21, p = .03; previously b = .07, t(163) = 2.64, p = .009 (see Figure 2).
Graphical depiction of the mediational analyses reported in Study 4.
To test the moderation of the mediation, we used nonparametric bootstrapping analyses (Preacher & Hayes, 2008). The model examined the mediation of attitude effects on attraction by expected change in attitude certainty across different conditions of evaluative focus (see Figure 2 for a graphical depiction of the mediational model). Results based on 1,000 bootstrapped samples indicated that the indirect effect of attitudes on attraction through expected change in attitude certainty was significant under metacognitive-reflection conditions, b = .02; 95% CI = [0.001, 0.061], but was nonsignificant under impression-formation conditions, b = −.01; 95% CI = [−0.04, 0.01]. These two indirect effects also differed significantly from one another, b = .03; 95% CI = [0.002, 0.091]. Thus, mediational analyses converge in their support for expectations about the potential for attitude uncertainty reduction as motivating similarity effects on attraction when people are focused on the implications of the target’s attitude for their own uncertain attitude.
In Study 4, evaluative focus tended to moderate the impact of doubt-associated attitudes, with metacognitive reflection tending to increase attitude use in making attraction judgments. Study 4 also offered initial mediational evidence that uncertainty motivates attraction to like-minded others because agreement appears capable of increasing the participant’s attitude certainty, at least when engaging in attitudinal social comparison (metacognitive reflection). When people consider the implications of the other for their own uncertain attitude, they evaluate the other in ways that parallel that other’s potential to impact that uncertainty. Coupled with the previous statistical controls for more general self-related variables, the mediational pattern for potential changes in attitude certainty might also suggest that some form of topic-specific attitude validation is at work (under metacognitive-reflection conditions).
General Discussion
Across several studies, the data support the hypothesized role of metacognitive reflection in determining whether attitude strength effects or attitude bolstering effects occur. When participants merely form an impression of a target, traditional strength patterns were observed (high levels of certainty increasing attitude-consistent evaluations of the target). That is, when one’s attitude serves primarily as input to judgment of a target, certainty increases the weight of that attitude in making the judgment. However, the pattern changed when research participants focused on the implications of the target’s attitude for their own attitude (i.e., the metacognitive implications of attitudinal social comparison). In such settings, a bolstering pattern was observed (uncertainty increased attitude-consistent attraction).
These different patterns help explain why a growing number of studies display bolstering patterns (i.e., strong influences of attitudes even when held with uncertainty) despite nearly 40 years’ worth of research showing the opposite pattern—certainty increasing the impact of prior attitudes on thinking, judgment, and behavior. The majority of traditional strength effects have occurred in settings where the focus of the actor is on the judgment or behavior target, and one’s attitude serves as an input to the relevant judgment or behavior. However, the settings in which bolstering patterns occur seem to involve a different evaluative focus for social perceivers. In these cases, it seems that social perceivers think instead about the implications of the target or behavior for the validity of their own views (consistent with motives to hold correct attitudes). When this more metacognitive focus takes the fore, attitude uncertainty can motivate people to actively seek support for the attitude, or can generate favorable reactions to discovered support for the attitude. That is, concerns about potential invalidity can lead people to seek attitude-consistent information (Sawicki et al., 2011), process attitude-consistent information deeply (Clark & Wegener, 2013), revel in group memberships that support one’s views (Clarkson et al., 2017), advocate for one’s position (especially when others seem potentially open to that position; Gal & Rucker, 2010), or like others who hold similar attitudes (the current research).
Limitations
The current research addressed the proposed moderator of metacognitive reflection fairly directly. Although judgment instructions did not focus attention specifically on validity per se, attitudinal social comparison has long been thought to draw attention to validity concerns (Festinger, 1954). Future research could profit by addressing more subtle means of instigating metacognitive reflection. Though Study 4 included assessment of perceptions that meeting the target would help enhance perceivers’ attitude certainty, future research could also benefit from additional assessment of validity consideration under metacognitive reflection. Current results were consistent with the general idea of metacognitive reflection. That is, perceptions of likely certainty changes did not relate to liking when merely forming an impression of the target but did predict liking when focused on implications of the target’s attitude for one’s own uncertain attitude. Because of the need for brevity in the context where we collected data, however, we used only a single measure of likelihood of a confidence boost. Also, though meeting the target might provide an opportunity for mutual support of one’s attitudes, it could be that learning of the person’s attitude alone led to perceived increases in attitude confidence. There is still much to learn about efforts to restore certainty when engaging in metacognitive reflection.
The attitude strength and bolstering patterns were produced using self-reported attitude certainty alongside previously validated manipulations of thought- or attitude certainty (i.e., the general confidence priming procedure of Petty et al., 2002; the topic-specific confidence prime of Blankenship et al., 2012). Checks on the material generated as part of the primes verified the desired certainty content differences, and the manipulated conditions produced results that paralleled those using self-reported attitude certainty. However, as previously noted, the manipulations did not produce differences in reported attitude certainty (or alternatives, such as self-certainty). These null effects are difficult to interpret. For example, the null effects on manipulation checks could be driven by the undesirability of appearing uncertain, a reactance against an apparent manipulation of confidence (e.g., compensatory reports of confidence after describing instances of doubt), or insensitive measurement. A number of factors in the results, however, point toward the influences of topic-specific perceptions of validity of one’s views. These include the parallel patterns based on topic-specific attitude confidence measures (that do not occur when using certainty of attitudes on alternative topics, Studies 1A and 2A), the use of a certainty manipulation specific to the attitude topic (Study 3), and the impact of the evaluative focus manipulations (that focus people on the implications of the target’s view for their own specific attitude). Also, when more general self-confidence or self-clarity variables have been measured, they have not produced results that parallel the patterns obtained using topic-specific attitude certainty measures (see Sawicki, 2013, for additional discussion). Thus, whereas data support a role of topic-specific attitude certainty in the current effects, there are no parallel indications that a more general form of (self) certainty played a crucial role in the current effects.
Future Directions
The current approach represents a previously unexamined perspective on attitude similarity effects on attraction. That is, similarity’s predictive hold over attraction appears more complex than long assumed. Similarity-attraction effects are well documented; however, the current research identifies at least two moderating variables that work in concert with one another: the level of certainty with which the attitude is held and the focus of the perceiver on judging the target or on the implications of the target for the person’s own attitude. There might be a variety of other conditions that either strengthen or weaken attitude similarity effects on attraction.
Future research might also consider other strength-related properties of attitudes. Similar to motives to reduce uncertainty, previous research on message-based persuasion and selective exposure to attitude-relevant information suggests that people are often motivated to reduce the level of ambivalence they have toward an attitude object (Clark, Wegener, & Fabrigar, 2008; Sawicki et al., 2013). It seems plausible that social perceivers would like others who are perceived as likely to help them resolve their ambivalence. If so, it could be that people would prefer to meet or spend time with a person who agrees with them rather than disagrees with them, much like ambivalent people process proattitudinal messages more deeply than counterattitudinal messages (Clark et al., 2008).
Whereas doubt and ambivalence have both been associated with discomfort (see van Harreveld, van der Pligt, & De Liver, 2009), it is less clear whether the types of bolstering effects observed in the current research would take place for other types of weak attitudes. It could be that many traditional properties of attitudes would moderate the impact of attitudes on liking of an agreeing or disagreeing other when focused on forming an impression of the target. However, when engaging in metacognitive reflection, “weak” attitudes that are not uncomfortable (such as attitudes toward unimportant targets or attitudes with low knowledge) might not show the same kinds of bolstering effects as in the current research. It could even be that, in some settings, metacognitive reflection might enhance the strength pattern by drawing additional attention to the metacognitive property itself (rather than reversing to become bolstering effects as in the current research). Thus, a number of questions for further investigation concern why the kind of metacognitive reflection we examined produces bolstering. A likely necessary precondition is being able and willing to engage in a moderate to high amount of thought. Before bolstering can occur, people would likely need to assess the current amount of certainty, and determine whether that level is lacking compared with a desired level of confidence (Chaiken, Liberman, & Eagly, 1989). Individuals should also possess the lay theory that attitude strength is generally malleable or at least changeable in this specific instance (Akhtar & Wheeler, 2016). Bolstering would be unlikely if individuals believe that their current doubt is unable to be resolved. A final potential “active ingredient” of the metacognitive-reflection instructions is social comparison. Bolstering occurred after a consideration of the connection between the target person’s attitude and that person’s own position on the issue. This comparison likely highlights the potential for agreement to benefit (or disagreement to exacerbate) the current desire for increased confidence. Future research should parse out which of these factors is necessary versus sufficient in producing the bolstering that occurs under metacognitive reflection.
With all of these potential qualifying conditions, the bolstering effects are likely to be more “fragile” than the strength effects. Although our meta-analytic combinations suggested similarly sized strength and bolstering patterns, the small samples in Studies 1B and 2B likely inflated the size of the observed bolstering effects in those studies. We are not making claims about the size of those effects, however, but are focused on the direction of effects (that were consistent across samples) and on the likely role of metacognitive reflection in moving people away from the more typical strength pattern. The current results, alongside a growing set of studies in which uncertainty enhances attitude effects on judgments, information processing, or behavior, suggest that metacognitive focus may serve as a key moderator in strength versus bolstering patterns (even if strength effects eventually prove to be generally larger and more stable than bolstering effects). We look forward to examining implications of the current work on metacognitive reflection for the many settings in which bolstering-type patterns have been observed.
Footnotes
Acknowledgements
The comments of committee members Christopher Agnew, Kipling Williams, and Meghan Norris are acknowledged with gratitude. Comments by Richard Petty, Pablo Briñol, and the members of the 2010-2013 Group for Attitudes and Persuasion at Ohio State University are also appreciated. We thank Eric Bookmyer, Christin Bryant, Claudia Cash, Anna Habib, Emily Manke, Jenny Leidermeier, Jonathan Pentz, Ellie Hess, Rebecca Weisshaar, and Neesha Hathi for their assistance with data collection and coding.
Authors’ Note
This article is based in part on the doctoral dissertation completed by Vanessa Sawicki under the direction of Duane T. Wegener.
Declaration of Conflicting Interests
The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.
Funding
The author(s) received no financial support for the research, authorship, and/or publication of this article.
Notes
Supplemental Material
Supplementary material is available online with this article.
References
Supplementary Material
Please find the following supplemental material available below.
For Open Access articles published under a Creative Commons License, all supplemental material carries the same license as the article it is associated with.
For non-Open Access articles published, all supplemental material carries a non-exclusive license, and permission requests for re-use of supplemental material or any part of supplemental material shall be sent directly to the copyright owner as specified in the copyright notice associated with the article.
