The ventromedial prefrontal cortex is particularly responsive to social evaluations requiring the use of person-knowledge

Humans can rely on diverse sources of information to evaluate others, including knowledge (e.g., occupation, likes and dislikes, education, etc.) and perceptual cues (e.g., attractiveness, race, etc.). Previous research has identified brain regions supporting person evaluations, but are evaluations based on perceptual cues versus person-knowledge processed differently? Moreover, are neural responses consistent when person-knowledge is available but unnecessary for the evaluation? This fMRI study examined how the use and availability of person-knowledge shapes the neural underpinnings of social evaluations. Participants evaluated well-known actors based on attractiveness or body of work (i.e., person-knowledge) and unknown models based on attractiveness only. Analyses focused on the VMPFC, following research implicating this region in positive evaluations based on person-knowledge. The VMPFC was sensitive to the (1) availability of person-knowledge, showing greater responses as ratings became more positive for actors (but not models) regardless of rating dimension and (2) use of available person-knowledge, showing greater activity as ratings for likability based on body of work became more positive for actors versus models rated on attractiveness. These findings indicate that although brain regions supporting person evaluation are sensitive to the availability to person-knowledge, they are even more responsive when judgments require the use of available person-knowledge.

As anticipated, actors were rated as significantly more familiar than models: MActor=4.255, MModel=2.085, F (1,17) For facial expression, we selected stimuli that at least 70% of raters categorized as having a happy or neutral expression. Facial expression did not statistically differ from the expected hypothesis of 2/3 neutral and 1/3 happy faces in each occupation-by-gender condition, χ 2 (7, n=90)=5.400, p=.611.
We also calculated a facial expression intensity score for the male and female actors and models.

Pre-Scan Procedure
Cover story and pre-scan survey. After consent forms and imaging center paperwork, the experimenter took each participant's picture using a digital camera and tripod. It was explained to the participant that he would later take part in a trust game unrelated to the current study for a chance to earn more money. The trust game included responses from real participants paired with their photographs. The experimenter further explained that the participant's responses in the trust game would be paired with his own picture and used in a future study with other participants. Participants then completed the State and Trait Anxiety, the Beck Depression Inventory II, and stress/anxiety sliding scales (see Supplementary Material S3).
Race-status task training. The race-status task training was based on training procedures used in previous work 1,4 .

Actor-Model Task Design and Exclusions
Design. As mentioned in the main text, the actor-model task consisted of an event-related design with two functional runs. The equated selection of 60 actors and 30 models was divided into two sets of 30 unique actors and two sets of 15 unique models, respectively. These unique sets of faces were divided across two functional acquisition runs, with 30 unique actors and 15 unique models in each run. Run assignment (first vs. second) for each unique set of faces was counterbalanced across participants. In each run, participants rated half of the actors on attractiveness and the other half of the actors on body of work. The halves of each actor set to be rated on attractiveness versus body of work were counterbalanced across participants. In summary, each participant rated each of the 60 actors once on a single dimension and each of the 30 models once on attractiveness. Finally, key assignment for responding was counterbalanced such that half of participants responded on a scale of 1 (very attractive/likable) to 4 (very unattractive/unlikable) and in reverse assignment for the other half.
In addition to counterbalancing the assignment of face set to runs (run 1 vs. run 2), the assignment of individual faces to evaluative rating conditions (attractiveness vs. body of work), and the assignment of response keys (ascending vs. descending), we also counterbalanced the order in which participants completed the three evaluative rating blocks within each run. The first run always used a different block ordering (e.g., actors rated on attractiveness, actors rated on body of work, and models rated on attractiveness) than the second run (e.g., models rated on attractiveness, actors rated on attractiveness, and actors rated on body of work). Specifically, the block orders of the two runs were completely orthogonal (i.e., no blocks were presented in the same location of the block sequence across runs). Following these constraints, we generated six orthogonal pairs of block orders. Taking into account all of the above factors (i.e., assignment of face set over two runs, the assignment of faces to two evaluative rating conditions, two response key assignments, and six block orderings) resulted in 48 versions of the experiment. Each participant completed one version of the experiment, with as few repeated versions as possible in the final participant sample.

Trial-and participant-level exclusions.
A total of 56 trials were eliminated because no ratings were provided. We then performed the familiarity exclusions (see main text) to ensure that each participant had sufficient knowledge of the actors from which to make body-of-work evaluations and no existing knowledge about the models, which were intended to be unfamiliar. After these trial-level exclusions, three participants were ultimately eliminated for having fewer than six trials per rating dimension (i.e., actors rated on attractiveness, models rated on attractiveness, and actors rated on body of work). The mean post-scan familiarity scores of the In order to test for differences in variance of in-scanner ratings within-participant by condition, we computed a repeated measures ANOVA on the standard deviation scores by subject for each of the three conditions: SD ~ Condition + Error(Subject/(Condition)). The mean standard deviations for all three conditions were not significantly different from one another,

Race-Status fMRI Experiment
Following the actor-model fMRI scan, participants completed a race-status impression formation task in the scanner 1,2 .

Post-Scan Procedure
Following their scan, participants completed explicit rating measures including a social distance task, likeability ratings, status recall, the trust game, and the actor-model familiarity emotions, the PANAS, the Perceived Stress Scale, a measure of chronic stress, and a measure of social value orientation. See Supplementary Material S3 for further details on these measures.
Then after their final saliva sample, participants were paid and debriefed.

Saliva Sample Acquisition
Saliva samples were collected for the race-status experiment 1 . Upon arrival, participants received water to facilitate saliva collection. Ten minutes after, the experimenter collected the first baseline sample. An additional baseline sample was acquired immediately prior to scanning. We first conducted one-sample t-tests to compare each parametric predictor to zero to explore whether activity in each exploratory ROI changed as a function of ratings (i.e., increased as evaluations became more positive) for each of the three conditions. Using the R function lm, a second analysis focused on the relative impact of each parametric predictor on activity in each exploratory ROI. Specifically, to compare the differences between parametric predictors, we conducted three separate linear regressions each using a unique set of contrast codes: (1) parametric predictor for actors' body of work = .5, parametric predictor for actors' attractiveness = 0, and parametric predictor for models' attractiveness = -.5; (2) parametric predictor for actors' body of work = 0, parametric predictor for actors' attractiveness = .5, and parametric predictor for models' attractiveness = -.5; and (3) parametric predictor for actors' body of work = .5, parametric predictor for actors' attractiveness = -.5, and parametric predictor for models' attractiveness = 0). However, for some of the exploratory ROIs reported here (i.e., bilateral NAcc, right STS, and bilateral TPJ), the contrast codes for the linear regressions differed from what was used for the VMPFC because the pattern of parametric modulation differed from the observed linear pattern in the VMPFC. In other words, parametric modulation for actors rated on body of work was not always larger than parametric modulation for actors rated on attractiveness, which was not always larger than for models rated on attractiveness. Any contrast coding that is different from the coding used for the VMPFC in the main text is explicitly stated in this section.
Do evaluations based on person-knowledge or attractiveness modulate exploratory ROI activity? We first examined whether increased positive evaluations led to greater exploratory ROI activity when those evaluations were based on (1) perceptual cues without person-knowledge (i.e., models rated on attractiveness), (2) perceptual cues with available person-knowledge (i.e., actors rated on attractiveness), and (3) person-knowledge (i.e., actors rated on body of work). Separate analyses were conducted for DMPFC, bilateral NAcc, bilateral STS, bilateral TPJ, precuneus, and bilateral amygdala. We present these results to spur future research and for meta-analytic purposes; therefore, below we report uncorrected p-values.
However, readers should note that when aggregating across laterality, the Bonferroni-corrected p-value to achieve significance for these exploratory analyses would be p<.008.
NAcc. All effects in the right NAcc were non-significant, |t(54)|<0.879, p>.382. We observed significant left NAcc involvement for actors rated on likability based on body of work, t(54)=2.515, p=.015, such that left NAcc activity increased as ratings of likability based on body of work became more positive. However, we did not observe significant left NAcc involvement for actors rated on attractiveness or models rated on attractiveness, |t(54)|<0.519, p>.605. Taken together, these results indicate that only the left NAcc may be sensitive to positive evaluations during the use of available person-knowledge.

STS.
All effects in the left STS were non-significant, |t(54)|<1.588, p>.117. We observed significant right STS involvement for actors rated on attractiveness, t(54)=2.261, p=.028 i , such that right STS activity increased as ratings of likability based on body of work became more positive. However, we did not observe significant right STS involvement for actors rated on likability based on body of work or for models rated on attractiveness, |t(54)|<1.943, p>.056.
Taken together, the left STS did not show preferential activity regardless of rating dimension while the right STS was sensitive to positive evaluations when person-knowledge was available but not when it was relevant to evaluations.
TPJ. All effects in bilateral TPJ were non-significant, |t(54)|<1.733, p>.087. The use or availability of person-knowledge did not reliably affect bilateral TPJ activity.
Precuneus. We observed significant precuneus involvement for actors rated on likability based on body of work, t(54)=2.304, p=.025 i , such that precuneus activity increased as ratings of likability based on body of work became more positive. We also observed significant precuneus involvement for actors rated on attractiveness, t(54)=2.385, p=.021 i , such that precuneus activity increased as ratings of attractiveness for actors became more positive. However, we did not observe significant precuneus involvement for models rated on attractiveness, t(54)=0.958, p=.342. Taken together, these results indicate that the precuneus may be sensitive to positive evaluations when person-knowledge is available, irrespective of whether that person-knowledge is directly relevant to evaluations.
i Note that this contrast does not survive Bonferroni correction for multiple comparisons: α = .0167.

Amygdala.
We observed significant left amygdala involvement for actors rated on likability based on body of work, t(54)=2.677, p=.010, such that left amygdala activity increased as ratings of likability based on body of work became more positive. We also observed significant left amygdala involvement for actors rated on attractiveness, t(54)=2.103, p=.040 i , such that left amygdala activity increased as ratings of attractiveness for actors became more positive. However, we did not observe a significant left amygdala involvement for models rated on attractiveness, t(54)=-0.225, p=.823. Additionally, we observed significant right amygdala involvement for actors rated on attractiveness, t(54)=2.149, p=.036 i , such that right amygdala activity increased as ratings of likability based on body of work became more positive. However, we did not observe significant right amygdala involvement for actors rated on likability based on body of work or for models rated on attractiveness, |t(54)|<1.076, p>.286. Taken together, these results indicate that whereas the left amygdala is sensitive to positive evaluations in the presence of person-knowledge regardless of its relevance to evaluations, this is the case for the right amygdala only when that person-knowledge is relevant to evaluations.
Do person evaluations modulate exploratory ROI activity more when personknowledge is used or when it's simply available? To examine whether sensitivity to positive ratings was especially pronounced for a particular condition (e.g., use of person-knowledge: actors' body of work) relative to other conditions (e.g., mere presence of person-knowledge: actors' attractiveness), we next conducted a linear regression on parameter estimates to test for differences between all parametric predictors. As above, separate analyses were conducted for DMPFC, bilateral NAcc, bilateral STS, bilateral TPJ, precuneus, and bilateral amygdala.

Supplementary Analyses Controlling for Post-Scan Familiarity Ratings
As mentioned in the main text, supplementary analyses were conducted in order to determine whether the VMPFC and the exploratory ROIs were robust to post-scan familiarity ratings. In these supplementary analyses, we therefore controlled for post-scan familiarity ratings to ensure that any remaining variability in familiarity could not account for the effects.
Specifically, we re-analyzed the ROI data while accounting for individual ratings of familiarity as a parametric modulator in the level-1 GLM. In sum, one single GLM incorporating three regressors for each of the three conditions (i.e., non-parametric parameters), three regressors for ii This value was rounded up from p=.04975. the condition-specific parametric parameters, three regressors for individual ratings of familiarity, and additional regressors for covariates of non-interest (a session mean, a linear trend to account for low-frequency drift, and six movement parameters derived from realignment corrections) were convolved with a canonical hemodynamic response function and used to compute parameter estimates (β) for each condition at each voxel. Results from these analyses were consistent with those reported when not accounting for post-scan ratings in familiarity, except for bilateral amygdala.
Finally, in order to account for individual ratings of familiarity in the whole-brain exploratory analyses, we also re-analyzed these data including post-scan familiarity ratings as a parametric modulator in the level-1 GLM for each whole-brain analysis. Results from the whole brain analyses were also generally consistent.

Do evaluations based on person-knowledge or attractiveness modulate activity in the
VMPFC and exploratory ROIs? Again, we first examined whether increased positive evaluations led to greater exploratory ROI activity when those evaluations were based on (1) perceptual cues without person-knowledge (i.e., models rated on attractiveness), (2) perceptual cues with available person-knowledge (i.e., actors rated on attractiveness), and (3) personknowledge (i.e., actors rated on body of work). As in the previous section, we did this by conducting separate one-sample t-tests on parameter estimates for each parametric predictor (i.e., actors rated on likability based on body of work, actors rated on attractiveness, and models rated on attractiveness) compared to zero.

VMPFC.
Controlling for post-scan familiarity ratings, we found similar findings in VMPFC as reported in the main text. We observed a significant effect in the VMPFC for actors rated on likability based on body of work, t(54)=3.726, p<.001, such that VMPFC activity increased as ratings of likability based on body of work became more positive. We also observed significant VMPFC involvement for actors rated on attractiveness, t(54)=2.585, p=.012, such that VMPFC activity increased as ratings of attractiveness for actors became more positive.
However, we did not observe significant VMPFC involvement for models rated on attractiveness, t(54)=0.506, p=.615. Taken together, these results indicate that the VMPFC is sensitive to the availability of person-knowledge, irrespective of rating dimension.

DMPFC.
When controlling for post-scan familiarity, we observed similar findings in DMPFC as when we did not control for post-scan familiarity ratings. All effects were nonsignificant, |t(54)|<1.029, p>.306. The use or availability of person-knowledge did not reliably affect DMPFC activity.

NAcc.
When controlling for post-scan familiarity, we observed similar findings in NAcc as when we did not control for post-scan familiarity ratings. Specifically, we observed significant left NAcc involvement for actors rated on likability based on body of work, t(54)=2.418, p=.019 i , such that left NAcc activity increased as ratings of likability based on body of work became more positive. However, we did not observe significant left NAcc involvement for actors rated on attractiveness and models rated on attractiveness, |t(54)|<0.785, p>.435. All effects in the right NAcc were non-significant, t(54)<1.205, p>.233. Taken together, these results indicate that left NAcc may be sensitive to the use of available person-knowledge.

STS.
When controlling for post-scan familiarity, we observed similar findings in STS as when we did not control for post-scan familiarity ratings. All effects in the left STS were nonsignificant, |t(54)|<1.410, p>.163. We observed significant right STS involvement for actors rated on likability based on body of work, t(54)=2.032, p=.047 i , such that right STS activity increased as ratings of likability based on body of work became more positive. We also observed significant right STS involvement for actors rated on attractiveness, t(54)=2.295, p=.026 i , such that right STS activity increased as ratings of attractiveness for actors became more positive.
However, we did not observe significant right STS involvement for models rated on attractiveness, t(54)=0.643, p=.523. Taken together, these results indicate that the right STS may be sensitive to the availability of person-knowledge, irrespective of its relevance to evaluations.

TPJ.
When controlling for post-scan familiarity, we observed similar findings in TPJ as when we did not control for post-scan familiarity ratings. All effects in bilateral TPJ were nonsignificant, |t(54)|<1.702, p>.094. The use or availability of person-knowledge did not reliably affect bilateral TPJ activity.

Precuneus.
When controlling for post-scan familiarity, we observed similar findings in precuneus as when we did not control for post-scan familiarity ratings. We observed significant precuneus involvement for actors rated on likability based on body of work, t(54)=2.419, p=.019 i , such that precuneus activity increased as ratings of likability based on body of work became more positive. We also observed significant precuneus involvement for actors rated on attractiveness, t(54)=2.223, p=.030 i , such that precuneus activity increased as ratings of attractiveness for actors became more positive. However, we did not observe significant precuneus involvement for models rated on attractiveness, t(54)=0.823, p=.414. Taken together, these results indicate that the precuneus may be sensitive to the availability of person-knowledge, irrespective of its relevance to evaluation.

Amygdala.
When controlling for post-scan familiarity, we observed different findings in amygdala activity compared to when we did not control for post-scan familiarity ratings. All effects in bilateral amygdala were non-significant, |t(54)|<1.784, p>.079. The use or availability of person-knowledge did not reliably affect bilateral amygdala activity when accounting for individual familiarity ratings.

Do person evaluations modulate activity in the VMPFC and exploratory ROIs more when person-knowledge is used or when it's simply available?
To examine whether sensitivity to positive ratings was especially pronounced for a particular condition (e.g., use of person-knowledge: actors' body of work) relative to other conditions (e.g., mere presence of person-knowledge: actors' attractiveness), we next conducted a linear regression on parameter estimates to test for differences between all parametric predictors. As above, separate analyses were conducted for VMPFC (confirmatory) and DMPFC, bilateral NAcc, bilateral STS, bilateral TPJ, precuneus, and bilateral amygdala (exploratory).

VMPFC.
When controlling for post-scan familiarity, we observed similar findings in VMPFC as when we did not control for post-scan familiarity ratings. The results revealed Exploratory whole-brain analyses. Finally, in order to account for individual ratings of familiarity in the whole-brain exploratory analyses, we also re-analyzed these data including post-scan familiarity ratings as an additional parametric modulator in the level-1 GLM for each whole-brain parametric analysis that was reported in the main text. Separate whole-brain analyses at the second level were performed to examine changes in neural activity as a function of in-scanner ratings provided during each of the three conditions: (1) actors rated on body of work, (2) actors rated on attractiveness, and (3) models rated on attractiveness. Using the Monte Carlo simulations included in AlphaSim, the minimum cluster size required for a whole-brain correction at p<.05 with an uncorrected threshold of p<.001 was estimated to be 51 contiguous voxels.
Separate parametric analyses predicted increases in brain activity across all voxels as participants' ratings increased or decreased for each of the three conditions. Results from the whole brain analyses were also generally consistent with those presented in the main text. We summarize results from each of the whole-brain analyses below. The results for all analyses are reported in Table S1.

Impact of person-knowledge use (actors rated on body of work only).
We observed greater activity in the calcarine sulcus and VMPFC as body-of-work ratings increased for the actors, but no reliable changes as body-of-work ratings decreased ( Figure S1). These findings converge with the ROI findings reported above showing that VMPFC activity was sensitive to increasing positivity in body-of-work ratings for well-known actors.

Impact of person-knowledge availability (actors rated on attractiveness only).
We observed greater activity in the lingual gyrus as attractiveness ratings increased for the actors, but no reliable changes as attractiveness ratings decreased.

Impact of percept-based evaluations without person-knowledge (models rated on attractiveness only).
No reliable changes were observed as attractiveness ratings increased or decreased.  Figure S1. Brain regions associated with increases in body-of-work likability ratings for actors. The results of this exploratory whole-brain analysis are displayed on a sagittal section, x = 6 mm. Increased body-of-work ratings for the actors were associated with increasing activity in the VMPFC (MNIx, y, z = [6, 63, 0]) (see also calcarine sulcus).

S3: Supplementary Measures
Individual differences were assessed for an unrelated studies on race and social status 1,2 and other unrelated exploratory analyses of resting-state data.

Current racial contact. 18
Childhood experience questionnaire. 13 Social dominance orientation. 19 Intergroup anxiety. 20 Feeling thermometers. Participants completed feeling thermometers separately for four groups of interest: high-status Black, low-status Black, high-status White, and low-status White.
Threat stereotype endorsement. Participants indicated their degree of explicit endorsement for threat stereotypes separately for four groups of interest: high-status Black, lowstatus Black, high-status White, and low-status White.

Social-cognitive measures for resting-state and anatomical analyses.
We also assessed a number of more general social-cognitive measures for large-scale analyses of resting state and structural imaging data.

Miscellaneous demographics.
Participants completed a standard battery of demographic items regarding age, sex, race/ethnicity, political affiliation, conservatism/liberalism, high school type, native language, citizenship, country of origin, and religious affiliation. Some measures of subjective SES (e.g., family social class) and objective SES (e.g., education level and occupational prestige, childhood postal codes, K-12 educational institutions, parental education, income and occupational prestige) were also included. Participants also completed a measure of identification with their ethnic group 21 .

Online Pre-Testing Part 2
Participants received links to actor online profiles and to a survey to familiarize themselves with the actors used in the fMRI experiment. Participants were instructed to study these actors prior to completing the online survey where participants had to correctly choose (50% chance of guessing) a movie in which the actor appeared or write in a different movie casting that actor.

Pre-Scan Measures
State-trait anxiety inventory. 22 Beck Depression Inventory II. 23 Single-item stress/anxiety scales. Participants responded to three questions about their current anxiety level, current stress level, and their stress level over the past month.

Post-Scan Measures
Social distance task. This task assessed the degree to which participants would place themselves in proximity to targets varying in race and status. At the start of the task, participants are re-introduced to four individuals pulled randomly from each of the four Race × Status conditions from the race-status fMRI experiment and paired with a name (see Supplementary Material S1). Participants were prompted to place themselves by mouse click anywhere inside of a blank white field. After clicking, the word "YOU" appeared in the clicked location.
Participants were then asked if they wished to adjust the position of the word "YOU". After any desired adjustments for the self, participants were prompted to place each of the four faces in the same field. All previous placements remained visible. Participants were given the opportunity to adjust each placement. After all five identities were placed, participants then had one final opportunity to adjust each one in turn before finishing the task. Participants completed this task three times.
Likeability rating task. Participants completed a measure of explicit likeability for each of the 60 face stimuli presented with the same status-associated colored backgrounds used in the scanner from the race-status fMRI experiment.
Status recall. Participants indicated to the best of their ability whether each of the 60 faces from the likability task, without any status-associated color backgrounds, was low or high in status.
Trust game. We adapted a trust game 24 .
Actor-model familiarity questionnaire. Participants indicated their familiarity with each actor and model used in the actor-model fMRI experiment (see main text for details).
Basic emotion states. Participants rated the degree to which they felt angry, happy, anxious, fearful, pain, and sad.