The false consensus effect (FCE), the tendency to project our attitudes and opinions on to others, is a pervasive bias in social reasoning with a range of ramifications for individuals and society. Research in social psychology has suggested that numerous factors (anchoring and adjustment, accessibility, motivated projection, etc.) may contribute to the FCE. In this study, we examine the neural correlates of the FCE and provide evidence that motivated projection plays a significant role. Activity in reward regions (ventromedial pFC and bilateral nucleus accumbens) during consensus estimation was positively associated with bias, whereas activity in right ventrolateral pFC (implicated in emotion regulation) was inversely associated with bias. Activity in reward and regulatory regions accounted for half of the total variation in consensus bias across participants (R2 = .503). This research complements models of the FCE in social psychology, providing a glimpse into the neural mechanisms underlying this important phenomenon.
Adaptation to the pressures and pitfalls of a dynamic social environment demands acute sensitivity to the attitudes, perspectives, and opinions of others. Whether official pollsters or ordinary social thinkers, we expend a great deal of effort to understand how others feel about the issues of the day. Nevertheless, empirical research shows that our understanding of others' attitudes is consistently biased by the positions we hold ourselves, a phenomenon known as the false consensus effect (FCE or “consensus bias”; Marks & Miller, 1987; Ross, Greene, & House, 1977). Moreover, this consensus bias has proven remarkably recalcitrant, persisting stubbornly when challenged by social feedback, sometimes even in the face of unanimous disagreement (Krueger & Clement, 1994). Given the importance of understanding others' attitudes, why should our own attitudes exert such a profound impact on our perceptions of social reality, and what mechanisms support this bias?
One prominent theory contends that consensus bias is a consequence of motivated projection—in short, we misperceive others' attitudes because we want to think of ourselves as being in the majority, holding views that are normatively “right” (see Morrison & Matthes, 2011; Sherman, Presson, & Chassin, 1984; Crano, 1983). If the projection of our own attitudes onto others is an instance of motivated projection, we might expect that consensus bias would be associated with neural correlates of social reward. Indeed, social approval has been found to activate neural structures involved in reward learning (Simon, Becker, Mothes-Lasch, Miltner, & Straube, 2014; Izuma, Saito, & Sadato, 2008) such as the nucleus accumbens (NAcc) and ventromedial pFC (VMPFC). Sharing our own attitudes with others has also been associated with reward (participants were willing to forego monetary payment to self-disclose), with corresponding activity in NAcc (Tamir & Mitchell, 2012). If motivated projection contributes significantly to the FCE by enhancing feelings of social approval or as a prelude to social sharing, we might therefore expect between-participant differences in the FCE to covary with activity in these reward regions.
Conversely, to accurately estimate the attitudes of others, regulatory mechanisms may be necessary to overcome the affective lure of our own antecedent opinions. That is, our own attitudes may serve as an evaluative anchor when we consider the attitudes of others, and regulatory mechanisms may help us to detach from this starting point and assess the attitudes of others more objectively (Tamir & Mitchell, 2010; Tversky & Kahneman, 1974). Functional neuroimaging studies have consistently implicated the right and left ventrolateral pFC (RVLPFC and LVLPFC) in emotion regulation (see meta-analysis by Kohn et al., 2014). Both of these regions have also been invoked when individuals must detach from their own perspective (Hartwright, Apperly, & Hansen, 2015; Cohen, Berkman, & Lieberman, 2013). If regulatory mechanisms are required to inhibit the prejudicial pull of one's antecedent attitudes and to accept the possibility that one's own position may not be predominant, then activity in RVLPFC and LVLPFC may be inversely associated with exhibited consensus bias.
Guided by the social psychological literature on the FCE, we sought to test the putative processes of motivated projection and regulatory restraint during consensus estimation in a functional neuroimaging study. We therefore interrogated hemodynamic response using fMRI while participants estimated the attitudes of the ordinary member of a comparison population (other University of California, Los Angeles [UCLA] undergraduates) on contemporary social, personal, and political issues.
Laboratory studies of the FCE typically ask for estimates of consensus in the absence of contextual information, but we were also interested in the effects of participants having some information that might be relevant to making the consensus judgment. To this end, we varied the information available about the attitudes of other UCLA students on a trial-by-trial basis, providing participants with false feedback concerning their peers. On Confirmation trials, participants were led to believe that another individual held an attitudinal position comparable with their own, a manipulation that we hoped would reaffirm participants' (biased) intuition that their attitudes were normative or commonplace among their peers. In contrast, on Disconfirmation trials, participants were informed that another individual held an attitude discrepant with their own, a manipulation we believed might encourage participants to restrain (insofar as possible) the tendency toward motivated projection. For the last trial type, No Information trials, participants made consensus estimates without additional feedback.
If motivated projection and regulatory restraint are important contributors to consensus bias, we anticipated that reward regions such as NAcc and VMPFC would drive consensus bias, whereas regulatory regions such as LVLPFC and RVLPFC would attenuate bias. In addition, the role of these regions and their putative psychological processes in consensus bias may interact with informational context. During Confirmation trials, social reward processes may exacerbate consensus bias uninhibited by contradictory feedback, whereas Disconfirmation trials may push participants toward a more critical interrogation of their attitudes and increase the likelihood of successful regulation.
Twenty-nine participants (17 women) were recruited by e-mail and Internet solicitations from the psychology research participant pool at UCLA. All participants had been enrolled as undergraduate students at UCLA for at least two quarters, and none had taken an introductory course in social psychology (to preclude familiarity with the FCE). Participants were judged ineligible if they did not differ from our estimate of the mean UCLA undergraduate attitude on a sufficient number of items. All participants were compensated $40 for their contribution to this research or received course credit. Participants provided written informed consent approved by the UCLA institutional review board. One participant's data are not included in these analyses because of partial acquisition failure (final n = 28).
Attitude Item Selection
Attitude items were selected from a larger set of 155 social, political, and personal issues (e.g., abortion rights, gay marriage, daily flossing, making out on a first date) that had been tested with an online sample of 178 UCLA undergraduates. Participants in this online sample indicated their attitudes toward each issue using a numeric scale ranging from 0 to 100 in integer increments (with anchors 0 = complete opposition, 25 = moderate opposition, 50 = neutrality, 75 = moderate support, and 100 = complete support). These responses provided a reasonable estimate of the mean UCLA undergraduate attitude on each of the 155 issues, and these values were used to determine error of estimation for the scanner task described below.
Before scanning, prospective participants in this study indicated their own attitudes on each of the 155 issues and were eligible to participate only if their responses differed from our estimate of the UCLA undergraduate population mean by at least 15 points on at least 90 items. If participants did not differ in their attitudes from the group mean for the items used, it would not be possible to disambiguate projection from accurate consensus estimation on a trial-by-trial level. As this was a major objective of the study, we felt it was necessary to impose such an inclusion criterion to provide a sufficient number of viable trials for the scanner task. The idiosyncrasies of participants' attitudes on the stimulus issues resulted in the selection of a unique set of attitude items for each individual, on each of which they differed from the UCLA undergraduate mean by at least 15 points. These items were randomly and equivalently divided among the Confirmation, Disconfirmation, and No Information conditions. Across participants, this procedure resulted in an average of 99 trials in total or 33 per consensus estimation condition.
Consensus Estimation Task
While undergoing fMRI, participants estimated the attitude of the ordinary UCLA student on each of the ideographically selected attitude items (see above). During the No Information condition, participants were simply asked to provide their best possible estimate of the attitude that an ordinary UCLA student would have on the given issue. To do this, they used an on-screen scale identical to that used during item selection (as described above) except that the values represented the attitude that the ordinary UCLA student would have, rather than the participant's own attitude.
In the Confirmation and Disconfirmation conditions, participants were provided with on-screen information ostensibly reflecting the attitudes of other UCLA undergraduates. Participants were told that, on each trial, the attitude of a different UCLA student from our larger Internet sample would be presented and that they could use (or disregard) this information in making their consensus estimates. Although this sample actually existed and was used to determine the true norms for each attitude item as described above, participants actually received false information designed to either confirm or disconfirm the presupposition that their own attitudes would be representative of the UCLA undergraduate population as a whole. In the Confirmation condition, participants were provided with an attitude that differed from their own by at most 5 points (in either direction). As all attitude items were preselected so that participants' attitudes were at least 15 points different from the mean, this ensured that the sample attitudes presented in the Confirmation were closer to the participant's own attitude than to the mean UCLA undergraduate attitude. In the Disconfirmation condition, participants were provided with a sample attitude that differed from the actual mean UCLA undergraduate attitude by at most 5 points (in either direction), so that this sample attitude was invariably closer to the actual mean than to the participant's own attitude. In both Confirmation and Disconfirmation conditions, deviations from the participant's own attitude and the mean UCLA undergraduate attitude were selected from a uniform random distribution so as to ensure that the presented attitude fell within the desired range.
On each trial (see Figure 1), the sample information (ostensibly reflecting the attitude of a single UCLA undergraduate) was presented numerically above the appropriate portion of the scale, with a line denoting the precise location corresponding to the other student's attitude. After the scale (and, if applicable, sample information) had appeared on-screen, participants had 10 sec within which to make their response. Trials were not explicitly separated into feedback and response phases, and sample information remained on-screen until participants had confirmed their response. Trial presentation was self-paced, with a jitter duration commencing immediately after participants' responses were registered. Intertrial jitter was selected from an exponential random distribution with a range of 4–9 sec and a mean value of 5 sec.
Nonsocial color judgment trials were also included as a basic perceptual-motor control condition. On these trials, participants were asked to judge the color of an on-screen square that varied continuously from completely red to completely blue. Participants were instructed to treat the midpoint value of “50” as indicating that the square appeared to them completely purple and neither bluer nor redder in hue. If the square appeared redder than bluer, participants were to select values greater than 50, with 100 indicating that they perceived the square to be completely red. If the square appeared bluer than redder, participants were to select values less than 50, with 0 indicating that the square is completely blue. Participants were instructed explicitly to provide their own judgment regarding the color of the square and to ignore how others might perceive it. Thirty control trials were included in the task for each participant, intermixed with consensus estimation trials.
Trial order was pseudorandomized such that no condition repeated more than twice sequentially and conditions were represented equally over two functional runs.
After completion of the consensus estimation task, participants viewed each attitude item again and indicated (a) their confidence in the accuracy of their consensus estimation and (b) the subjective importance of their attitude on the issue. For both judgments, participants used a 100-point integer scale with anchors at 0 = not at all confident (important), 25 = a little confident (important), 50 = moderately confident (important), 75 = very confident (important), and 100 = extremely confident (important).
fMRI Data Acquisition
All imaging data were acquired using a 3.0-T Siemens Trio scanner (Siemens, Erlangen, Germany) at the Ahmanson-Lovelace Brain Mapping Center at UCLA. Across two functional runs, approximately 650 T2*-weighted EPIs were acquired during the completion of experimental tasks described above (slice thickness = 3 mm, gap = 1 mm, 36 slices, repetition time [TR] = 2000 msec, echo time [TE] = 25 msec, flip angle = 90°, matrix = 64 × 64, field of view = 200 mm). An oblique slice angle was used to minimize signal dropout in ventral medial portions of the brain. In addition, a T2-weighted, matched-bandwidth anatomical scan was acquired for each participant (TR = 5000 msec, TE = 34 msec, flip angle = 90°, matrix = 128 × 128; otherwise identical to EPIs). Finally, we acquired a T1-weighted magnetically prepared rapid acquisition gradient-echo anatomical image (slice thickness = 1 mm, 176 slices, TR = 2530 msec, TE = 3.31 msec, flip angle = 7°, matrix = 256 × 256, field of view = 256 mm).
fMRI Data Preprocessing and Analysis
Preprocessing and ROI Definition
Functional data were analyzed using SPM8 (Wellcome Department of Cognitive Neurology, London, UK). Within each functional run, image volumes were corrected for slice acquisition timing, realigned to correct for head motion, segmented by tissue type, and normalized into standard Montreal Neurological Institute (MNI) stereotactic space (resampled at 3 × 3 × 3 mm). Finally, images were smoothed with an 8-mm FWHM Gaussian kernel.
Given our specific hypotheses regarding the role of reward and regulatory regions in shaping the expression of consensus bias, all principal analyses were conducted on a priori ROIs. Reward regions were selected from a comprehensive meta-analysis of the literature on subjective valuation conducted by Bartra, McGuire, and Kable (2013). Six-millimeter spherical ROIs were defined based on statistical meta-analytic peaks from their analysis of the decision phase of rewarding trials, during which participants selected between various choice alternatives on the basis of their subjective value (see Table 3 in Bartra et al., 2013, for details). We felt that this particular conceptualization best matched the mechanism of motivated projection hypothesized to underlie the FCE. This procedure yielded a VMPFC ROI centered at MNI of −2, 40, −8 and left and right NAcc ROIs centered at MNI of −6, 8, −4 and 6, 10, −8, respectively. As we did not have separate hypotheses regarding the function of left and right NAcc in this context, the union of these regions was employed as a single ROI for all analyses. Regulatory regions were selected from a recent meta-analysis of the literature on emotion regulation by Kohn et al. (2014). Six-millimeter spherical RVLPFC and LVLPFC ROIs were defined based on the peak activation coordinates associated with cognitive emotion regulation in the left and right inferior frontal gyrus (MNI: −42, 22, −6 and 50, 30, −8; see Table 2 in Kohn et al., 2014, for details). All ROIs were constructed using the automated anatomical labeling toolbox (Tzourio-Mazoyer et al., 2002) of the Wakeforest University Pickatlas (Maldjian, Laurienti, Burdette, & Kraft, 2003).
fMRI Analytic Paradigm
A general linear model was defined for each participant, in which trials were modeled with separate functions corresponding to (1) the initial presentation of the trial and (2) a fixed epoch corresponding to the final 2.5 sec preceding (and including) the participants' final response. The initial portion of the trial differs significantly between conditions, with the Confirmation and Disconfirmation conditions, but not the No Information condition, including on-screen information regarding the attitudes of another UCLA undergraduate. As parameter estimates from this portion of the trial are not directly comparable across conditions, the initial portion of each trial was therefore modeled as a parameter of no interest in the general linear model. Planned comparisons were conducted on parameter estimates corresponding to the final period of each trial (i.e., the last 2.5 sec before participant response), which we believe better corresponds to the period of participants' decision-making and response selection. Both stimulus presentation and response selection were convolved with the canonical (double-gamma) hemodynamic response function. Four regressors of interest were modeled to the response period of the Confirmation, Disconfirmation, No Information, and Control conditions. The model also controlled for 18 motion parameters (three translations and rotations as well as their squares and first-order derivatives) and a junk regressor for acquisitions on which either translation exceeded 2 mm or rotation exceeded 2° in any direction. The time series was high-pass filtered using a cutoff period of 128 sec, and serial autocorrelations were modeled as an AR(1) process.
Consensus bias was computed on a trial-by-trial basis as the error of estimation of a participant's consensus estimate regarding the attitude item (relative to the true mean of our larger, 197-person sample) in the direction of the participant's own attitude on the attitude item (acquired several days before the scan). That is, consensus bias was operationalized as |consensus estimate − true sample mean|(x − 1 if consensus estimate underestimates support of own attitude). Bias values were also capped by the participant's own attitude; that is, participants could not have a bias score greater than the difference between their own attitude and the sample mean. The consensus bias metric used is thus positive when participants overestimate support for their own attitudinal positions in the UCLA undergraduate population, negative when they underestimate support for their own attitudinal positions in the undergraduate population, and 0 if their estimate is accurate. Because this bias metric is sensitive to participants' actual overestimation of support for their own attitudes, we believe it is an effective operationalization of consensus bias for the purposes of imaging research. It is conceptually similar to the “truly FCE” developed by Krueger and Clement (1994).
Parameter estimates were extracted from all ROIs using MarsBaR (Brett, Anton, Valabregue, & Poline, 2002) and entered into multiple regression models (see Results below) with participants' mean consensus bias scores (overall or condition specific, depending on the model under evaluation) as the dependent variable. Additional regions whose activity correlated with between-participant variation in consensus bias were identified by whole-brain analyses, interrogating only gray matter voxels. Monte Carlo simulations implemented in 3dClustSim (from AFNI; Cox, 1996) were used to determine appropriate cluster-size thresholds (70 contiguous voxels) to ensure overall false discovery rate of less than 0.05, when combined with a voxelwise significance threshold of p < .005 within gray matter voxels. All results reported exceed these joint voxelwise and cluster-extent thresholds, except as noted.
Behavioral Effects of Social Information on Consensus Bias
Consistent with the extensive behavioral literature on the FCE, consensus bias scores were significantly greater than zero both overall and for each information condition individually (Mall = 12.174, t(27) = 15.265, p < .001; MCon = 19.071, t(27) = 18.604, p < .001; MNoI = 10.320, t(27) = 9.950, p < .001; MDis = 8.272, t(27) = 10.445, p < .001). Mean consensus bias scores were not related either to mean estimate confidence ratings (r = .17, ns) or to mean attitude importance scores (r = −.185, ns). Overall, there was a marginally significant inverse correlation between mean consensus bias and mean RT, averaging across all conditions (r = −.344, p = .073). Mean bias in the Confirmation condition was inversely correlated with mean RT to Confirmation trials (r = −.399, p = .035), but this relationship did not hold for the Disconfirmation or No Information conditions.
Repeated-measures ANOVA revealed a substantial effect of Information condition (Confirmation, Disconfirmation, or No Information) on participants' exhibited bias (F(2, 54) = 80.580, p < .001). Participants showed greater bias in the Confirmation condition than in the No Information condition (MCon = 19.071 vs. MNoI = 10.320, t(27) = 9.095, p < .001). Participants also showed significantly less bias in the Disconfirmation condition than in either the No Information condition (MDis = 8.272 vs. MNoI = 10.315, t(27) = −2.279, p = .031) or the Confirmation condition (MDis = 8.272 vs. MCon = 19.071, t(27) = −11.509, p < .001).
The presentation of sample information also affected participants' RTs (F(2, 54) = 5.137, p = .007). Predictably, both the Confirmation and Disconfirmation conditions resulted in longer RTs than the No Information condition (MCon = 4.541 vs. MNoI = 4.323, t(27) = 3.077, p = .005; MDis = 4.522 vs. MNoI = 4.323, t(27) = 2.367, p = .025). However, the Confirmation and Disconfirmation conditions did not differ in RT (MCon = 4.541 vs. MDis = 4.522, t(27) = 0.274, p = .786).
Participants' confidence in their consensus estimates was also affected by the presentation of sample information on-screen (F(2, 54) = 4.673, p = .011). The Confirmation condition increased participants' confidence in their consensus estimates relative to the No Information (MCon = 68.761 vs. MNoI = 66.064, t = 2.662, p = .013) and Disconfirmation (MCon = 68.671 vs. MDis = 65.676, t = 2.568, p = .016) conditions, but the Disconfirmation condition did not decrease participants' confidence in their estimates relative to No Information (MDis = 65.676 vs. MNoI = 66.064, t(27) = −0.361, p = .720). Perceived attitude importance was not influenced by information condition (F(2, 54) = 1.068, p = .351).
On average, participant response in the color judgment control trials was not biased in favor of either color (red or blue) along the continuum presented (mean signed error: MErr = −0.048, t = −0.057, p = .955). Participants were not terribly inaccurate in their color judgments (mean absolute error: MAbsErr = 9.818 of a 100-point scale), but this error was significantly different from zero (t = 24.594, p < .001). RTs were shorter for the Control trials than for the Consensus Estimation trials (MConsensusRT = 4.460 vs. MColorRT = 3.206, paired-sample t = −8.388, p < .001), suggesting that the color judgment task was slightly easier to perform.
Taken together, these results suggest that participants integrated the affirming and challenging information into their consensus estimates in the manner intended. The Confirmation and Disconfirmation trials took slightly longer to complete, on average, and the information provided had the expected impact on participants' demonstrated bias—enhancing and diminishing the consensus bias on Confirmation and Disconfirmation trials, respectively. Participants were slightly more confident when the sample information confirmed the normativity of their beliefs than when no information was provided. It is worth noting that, consistent with the observed consensus bias, participants were relatively confident in their estimates in all conditions. Finally, because attitude items were assigned to experimental conditions randomly, it is reasonable that there should not be significant differences in perceived attitude importance.
Neural Correlates of Between-participant Variation in Consensus Bias
Given our assumptions about the reward and regulatory processes underlying the FCE, we sought to assess whether between-participant variation in critical ROIs would predict variation in participants' observed levels of consensus bias. Specifically, as outlined above, we anticipated that reward activity in the VMPFC and NAcc would be associated with greater consensus bias, whereas regulatory activity in the RVLPFC and LVLPFC would be associated with diminished bias.
Parameter estimates were extracted from these regions during the response period, for each information condition (Confirmation, Disconfirmation, and No Information) versus control, and entered into a multiple regression model as predictors of between-participant variation in consensus bias. As noted above, RT differed as a function of condition and was marginally inversely associated with consensus bias. To rule out any possible effects due simply to variation in RT, this variable was also included as a regressor of no interest. We first assessed whether mean task-related activity (averaging over conditions relative to control) in the ROIs would significantly predict mean consensus bias (again averaging bias scores across conditions). In this model, activity in VMPFC, bilateral NAcc, RVLPFC, and LVLPFC together significantly predicted about half of the variance in participants' mean consensus bias (model: F(5, 22) = 4.455, p = .006, R2 = .503). In addition, the neural predictors independently accounted for a significant proportion of variance in consensus bias scores: Bias was positively associated with activity in NAcc (t = 2.303, p = .031, partial correlation r = .441) and VMPFC (t = 2.164, p = .042, partial correlation r = .419) but negatively associated with activity in RVLPFC (t = −2.192, p = .039, partial correlation r = −.423). Activity in the LVLPFC was not significantly associated with consensus bias (t = 0.287, p = .777, partial correlation r = .061). These results are consistent with our predictions regarding the role of reward and regulatory processes in consensus estimation, insofar as reward-related regions (NAcc and VMPFC) were more active in participants who exhibited greater mean levels of bias, whereas the RVLPFC was recruited more by participants whose estimates were less biased (see Figure 2).
Similar results were uncovered when trials were analyzed in a condition-specific manner, that is, when parameter estimates extracted from the a priori ROIs during a given condition were used as predictors of observed bias during that condition. For No Information trials, the overall model (including NAcc, VMPFC, RVLPFC, LVLPFC, and RT as predictors) remained significant (model: F(5, 22) = 5.636, p = .002, R2 = .562). Consensus bias scores during the No Information condition were positively associated with activity in NAcc (t = 2.734, p = .012, partial correlation r = .504) and VMPFC (t = 2.122, p = .045, partial correlation r = .412) during this condition and negatively associated with activity in RVLPFC during this condition (t = −3.204, p = .004, partial correlation r = −.564). Again, the association between activity in the LVLPFC and consensus bias was not significant for the No Information condition (t = 0.905, p = .375, partial correlation r = .189).
A comparable model also significantly predicted between-participant variation in bias observed during the Disconfirmation condition (model: F(5, 22) = 4.060, p = .012, R2 = .414). Consensus bias scores during the Disconfirmation condition were positively associated with activity in the VMPFC (t = 2.684, p = .013, partial correlation r = .488) and marginally associated with activity in the NAcc (t = 2.019, p = .055, partial correlation r = .388) but negatively associated with activity in the RVLPFC (t = −2.572, p = .017, partial correlation r = −.473). Parameter estimates from LVLPFC did not significantly predict bias in the Disconfirmation condition (t = 0.951, p = .351, partial correlation r = .195). In the Confirmation condition, a multiple regression model including RT and ROI parameter estimates from the ROIs did not significantly predict mean bias (model: F(5, 22) = 1.399, p = .263).
Taken together, these results demonstrate that consensus bias is positively associated with activity in regions (bilateral NAcc and VMPFC) implicated in the subjective experience of reward and negatively associated with activity in a key regulatory region (RVLPFC) during both the Disconfirmation and No Information conditions. These are precisely the trials in which participants ought to be uncertain about the status of their own attitudes vis-à-vis those of their peers and in which the interplay of motivated reasoning and regulation is expected to shape observed bias. In the Confirmation condition, the same regions do not seem to predict consensus bias, perhaps because participants may take the confirmatory feedback at face value most of the time.
Whole-brain analyses were conducted to determine whether brain regions other than the a priori ROIs would show significant associations with between-participant variation in consensus bias. Interestingly, this analysis revealed a cluster in the left precuneus that was positively associated with observed bias during the No Information condition (peak MNI: −3, −58, 16; t = 4.155, k = 117). Given the role of the precuneus in retrieval processes (Kim, 2013), this result provides tentative evidence that biased retrieval may support errors of consensus estimation, even when social feedback is unavailable for direct assessment.
In this investigation, we conducted a functional neuroimaging test of a prominent social psychological account of the FCE, which views consensus bias as a consequence of motivated reasoning/projection (see Marks & Miller, 1987). The results provide support for the theoretical importance of motivated projection in shaping the expression of the FCE but also highlight participants' (limited) capacity for regulatory restraint—a factor not fully considered in previous accounts of the FCE. In the No Information and Disconfirmation conditions, established reward regions (NAcc and VMPFC) were associated with a tendency toward greater bias, whereas activity in the RVLPFC (implicated in emotion regulation and self-restraint) was inversely related to the consensus bias. Indeed, overall, the activity in these ROIs accounted for almost 50% of the total between-participant variation in consensus bias. These findings suggest, as some social psychologists have theorized (see, e.g., Morrison & Matthes, 2011; Sherman et al., 1984; Crano, 1983), that our tendency to project our own attitudes onto others is not simply the result of the greater accessibility intrinsic to our own perspective. Indeed, these neuroimaging results are congruent with the notion that projection is (at least in part) motivated, perhaps reflecting the need to affirm the normativity of our attitudes within the broader community.
The results of this study are also consistent with a number of cognitive and social cognitive findings concerning related phenomenon. A very similar pattern of motivated projection and regulatory restraint has been observed previously with the “belief” bias in syllogistic reasoning. The “belief” bias results when individuals are presented with a valid logical argument that results in an untrue conclusion. Consider the following argument:
No addictive things are inexpensive.
Some cigarettes are inexpensive.
Therefore, some cigarettes are not addictive.
This argument's conclusion is generally thought to be untrue; however, it is also a valid conclusion because it follows logically from the premises. Fewer than half of individuals identify this argument as logically valid (Evans, Barston, & Pollard, 1983), while showing almost perfect accuracy on trials where the participants' beliefs were not at odds with the argument's conclusion.
An fMRI study examined the “belief” bias (Goel & Dolan, 2003), including the critical trials during which participant beliefs were likely to be at odds with the validity judgment. When participants fell prey to the “belief” bias and projected their beliefs onto the validity decision, rather than preventing their own beliefs from interfering, the only region of the brain that was relatively more active was VMPFC. This is analogous to the greater VMPFC activity we observed to the extent that our participants erroneously projected their own attitudes onto the consensus estimates of others' attitudes. In contrast, when participants overcame the “belief” bias and correctly identified the valid, but untrue, conclusions as valid, the only brain region that was relatively more active was RVLPFC. This again is analogous to our finding that reduced consensus bias was associated with RVLPFC activity.
Within social cognition, VMPFC has been associated with motivated social cognition (Hughes & Beer, 2012; Beer & Hughes, 2010). A number of studies also suggest that RVLPFC plays a key role in detaching from one's own perspective or existing beliefs to consider additional information or perspectives. For instance, when first impressions, which are notoriously difficult to change, are successfully updated, this change is associated with RVLPFC activity (Bhanji & Beer, 2013; Mende-Siedlecki, Cai, & Todorov, 2013). In addition, in our own work, we have also observed that, when adolescents change their own attitudes to be more like those of a parent or a peer, there is greater activity in RVLPFC, relative to trials when less of an attitudinal shift occurred (Welborn et al., 2016).
Perhaps, most compelling is a case study of a patient with damage localized to RVLPFC (Samson, Apperly, Kathirgamanathan, & Humphreys, 2005). As long as the patient had no antecedent beliefs or preferences relevant to a perspective-taking task, the patient showed perfectly preserved performance. However, when the patient had his own perspective or preference, he could not help but project this onto others, showing childlike egocentrism. If a game were being played between two teams that he did not care about personally, he could accurately assess how fans of each team would react if one of the teams scored. In contrast, if the game included the patient's own favorite team, he assumed other fans would have the same reaction as him, even if told someone was rooting for the other team.
All of the aforementioned phenomena (FCE, “belief” bias, person perception updating, and recognizing another's perspective when discrepant with our own) may be examples of a broader phenomenon known as naive realism (Ross & Ward, 1996). Naive realism refers to the (implicit) belief that we see the world objectively and that other reasonable people should thus see it the same way as we do. If they fail to see it our way, we rarely consider how our perception or understanding might be wrong or only one of several possible points of view. Although we often fail to overcome our own initial way of seeing things and assume others see things the same way as we do, as evidenced by self-projection in the FCE, sometimes, we are able to detach ourselves from our own perspective. Across these various studies, including the current FCE findings, RVLPFC appears to play a role in overcoming naive realism and appreciating information beyond our initial intuitive perspective.
Given that naive realism is generally believed to be both entrenched and socially problematic, identifying neural dynamics that support even temporary detachment from this state of self-certainty and self-projection is very important. Because of naive realism, we tend to overestimate others' susceptibility to biases while underestimating our own (Pronin, Gilovich, & Ross, 2004; Pronin, Lin, & Ross, 2002). This pronounced asymmetry in perceptions of bias between self and others has been shown in a variety of important domains, including interpersonal perception (Pronin, Krueger, Savitsky, & Ross, 2001) and intergroup conflict (Robinson, Keltner, Ward, & Ross, 1995). Thus, if RVLPFC plays a central role in those occasions when naive realism is overcome, then this may serve as a point of focus for future investigations and interventions. For instance, a recent study observed that self-control training enhanced RVLPFC responses in a region very close to the one identified in the current study (Berkman, Kahn, & Merchant, 2014). It is possible that training regimens that focus on enhanced motor self-control would also produce benefits for overcoming nonmotor impulses as well, like those that must be restrained when we are under the sway of naive realism (Berkman, Burklund, & Lieberman, 2009).
Although the results of this experiment are consistent with motivated projection as a cause of consensus bias, the diversity of function associated with the brain regions in question (especially the VMPFC) means that other contributing factors should also be considered in future work. The VMPFC has often been implicated in self-related cognition (Jenkins & Mitchell, 2011; Tamir & Mitchell, 2010), and both the VMPFC and the NAcc have been associated with social influence processes (Welborn et al., 2016; Zaki, Schirmer, & Mitchell, 2011). Such processes are not inconsistent with motivated projection, but their involvement could be clarified by direct comparisons of self-related cognition, influence, and consensus estimation in a single sample.
We should also note a number of crucial limitations regarding causal inferences based on correlational evidence, such as the fMRI results presented in this article. Statistical models of hemodynamic response at best reveal associations between neural activity and bias but do not uniquely specify the causal relationships between brain regions and behaviors. In addition, there is considerable uncertainty about the timing of the psychological processes associated with consensus estimation in this paradigm. Consensus estimation trials evolved in a relatively unconstrained manner, with no clear demarcation enforced by the experimental design between the period during which participants were making judgments and the period of scale manipulation. Indeed, for many participants, these periods may have been overlapping. Thus, it is possible that other processes, besides motivated projection and regulatory restraint, are responsible for the association between the regions specified and consensus bias. In light of previous work on the FCE and the neuroscience literature on reward and regulatory processes, we feel that an account of consensus bias in terms of motivated projection and regulatory restraint is most consistent with the observed results. Nevertheless, other causal relationships are plausible and ought to be explicitly examined in future work. For example, activity in putative reward regions may be elicited as a response to or an effect of attitudinal projection, rather than as an antecedent cause of bias. Future research might assist in clarifying with greater precision the causal mechanisms involved in consensus bias.
The present research has explored the neural correlates of the FCE with respect to contemporary social, political, and personal issues. The results of this work are consistent with social psychological accounts of consensus bias in terms of motivated reasoning and suggest that regulatory mechanisms may offer hope for attenuating bias in the face of social feedback. Further research may profitably understand the circumstances and limits of individuals' capacities to overcome bias as well as investigate their neural mechanisms.
Reprint requests should be sent to Matthew D. Lieberman, Department of Psychology, 4611 Franz Hall, UCLA, Los Angeles, CA 90095-1563, or via e-mail: firstname.lastname@example.org.