Regional and network neural activity reflect men’s preference for greater socioeconomic status during impression formation

Evidence from social psychology suggests that men compared to women more readily display and pursue control over human resources or capital. However, studying how status and gender shape deliberate impression formation is difficult due to social desirability concerns. Using univariate and multivariate fMRI analyses (n = 65), we examined how gender and socioeconomic status (SES) may influence brain responses during deliberate but private impression formation. Men more than women showed greater activity in the VMPFC and NAcc when forming impressions of high-SES (vs. low-SES) targets. Seed partial least squares (PLS) analysis showed that this SES-based increase in VMPFC activity was associated with greater co-activation across an evaluative network for the high-SES versus low-SES univariate comparison. A data-driven task PLS analysis also showed greater co-activation in an extended network consisting of regions involved in salience detection, attention, and task engagement as a function of increasing target SES. This co-activating network was most pronounced for men. These findings provide evidence that high-SES targets elicit neural responses indicative of positivity, reward, and salience during impression formation among men. Contributions to a network neuroscience understanding of status perception and implications for gender- and status-based impression formation are discussed.


Methods
Participants. Participants were recruited from the Chicago area via ads posted online, on public transportation, and fliers. In order to participate in the study, participants were required to meet the following criteria: (1) identify as White; (2) between 18 and 35 years old; (3) right-handed; (4) lived in the U.S. for at least 5 years; (5) good command of the English language; (6) no history of drug abuse; (7) no history of serious head injury; (8) no color vision problems; (9) no current acute illness; (10) not currently taking psychotropic medication; (11) no diagnosis of developmental disorders; (12) no diagnosis of a chronic disease that compromises mental, neural, or autonomic function; and (13) pass a standard MRI safety screen. Two-hundred fifteen people completed the initial screening procedures (see Supplemental Material S1). Of these 215 individuals, 69 eligible participants completed all parts of the study. Scanner data from three participants were corrupted during MR image reconstruction and therefore could not be used. One participant with an extremely low recorded response rate (9%) on the impression formation task indicated a lack of diligence and was removed as a sample outlier (i.e., exceeding 3 standard deviations from the sample mean recorded response rate: M = 93.75%, SD = 16.89%). This resulted in a final sample size of 65 participants (M Age = 23.844, SD Age = 4.334). Although a larger number of men (n = 37) than women (n = 28) were included in our final sample, the difference was non-significant, X 2 (1, n = 65) = 1.246, p = 0.264. Of the 63 participants who provided subjective status scores, women trended lower in subjective status (M = 6.571, SD = 1.665) than men (M = 7.257, SD = 1.482), but this difference was non-significant (see Supplemental Text S4 for full statistics).
Ethical guidelines and informed consent. This research was approved by the University of Chicago Social and Behavioral Sciences Institutional Review Board (IRB14-1232). All research was conducted in line with the University of Chicago's ethical guidelines. All participants provided informed consent prior to participation.
Protocol. Procedures and measures are outlined below. Further details on procedures and measures are provided in Supplemental Texts S1 and S2, respectively.
Gender identification and online pre-testing session. Upon passing the initial prescreen, participants were instructed to complete a series of surveys prior to arriving for their scanning appointment. In addition to reporting their gender ("Male", "Female", or "Other"), participants also completed several exploratory measures in this online pre-testing session (see Supplemental Text S2 for a full listing). No one in the analyzed sample indicated "Other" as a gender identity.
It bears mentioning that participants were asked to report their gender, which is a social construct, but the provided response options were for biological sex (male, female) instead of gender (man, woman). Base rates for these attributes are extremely correlated. Accordingly, we conflate sex and gender in this report. Although Scientific Reports | (2020) 10:20302 | https://doi.org/10.1038/s41598-020-76847-z www.nature.com/scientificreports/ differentiating the effects of sex from effects of gender is beyond the scope of the present work, we acknowledge this as important future direction.
fMRI session. Eligible participants were scheduled for a two-and-a-half-hour appointment at the imaging center. Approximately thirty minutes of the appointment was spent on the pre-scan procedure (see Supplemental Text S2). The next hour of this session involved time in the scanner, and the last hour was spent completing the post-scan surveys and tasks (see Supplemental Text S2).
Arrival and preliminary measures. Participants first read and signed consent forms. After completing a brief survey (see Supplemental Text S2), participants received instructions regarding the different tasks. Participants learned that they would be completing two in-scanner tasks: one involving passively watching videos of police arrests (see Supplemental Text S1) and the other involving forming impressions of faces. Prior to the scanning session, participants learned to use the button boxes to indicate they formed their impression of each target (see below for details).
Scanner set-up and preliminary scans. Upon entering the scanner, participants were fitted with a button box on each hand. Once adequately situated in the scanner, participants completed two functional runs of an unrelated fMRI task (~ 13 min in total). This task involved viewing videos of police officers making arrests of civilians (see Supplemental Text S1 for details).
Status-color association training. Before completing the impression formation task, participants completed a self-paced status-color association training lasting approximately 10 min. Participants were told they would be forming impressions of individuals that vary in status, defined as SES 8 : "Those who have the highest social status tend to have the most money, the most education, and the most respected jobs. Those who have the lowest social status tend to have the least money, the least education, and the least respected jobs or no job". Because some perceptual antecedents (e.g., strength, height) are thought to convey status more readily for men than for women 41 , we opted to convey status information via knowledge-based cues. To this end, we trained participants to associate high and low SES with the colors blue or orange (see Supplemental Text S2) [21][22][23]42 . Status-color associations were counterbalanced across participants such that, for about half of our participants, blue represented low status and orange represented high status, and vice versa for the remainder of participants. Participants categorized: (1) blue and orange silhouettes as either high or low status (M = 95.57% for all trials); and (2) the words "high status" and "low status" as represented by either blue or orange (M = 90.14% for all trials). Participants had an average of 92.87% accuracy across these two training components. Finally, participants practiced forming impressions of two unique faces per gender that were not used in the experimental trials, each framed with blue or orange. Before proceeding to the fMRI impression formation task, participants had to report the correct status-color association back to the experimenter when prompted. Only one participant incorrectly recalled the status-color associations prior to the start of the first task run. The researcher corrected the participant and asked the participant to repeat back what they just learned. This participant also correctly recalled the associations prior to the second run. Impression-formation fMRI task. Having learned to associate different colors with different levels of status, participants then completed two functional runs of the impression formation task, each lasting approximately 7 min. Prior to the start of this fMRI task, participants learned that they would simply be forming overall impressions of faces. We guided participants further by indicating that there were no correct impressions. For each face, participants were asked to focus on their subjective thoughts and feelings based on all the information available. When they had formed an impression, they were instructed to respond by pressing both pointer fingers at the same time 21,23 . Each face was displayed for 1,500 ms, followed by a fixation cross with a jittered duration of 500, 2500, 4500, or 6500 ms (see Fig. 1). Participants were encouraged to form their impression prior to or shortly after the disappearance of each face stimulus. After entering timely responses during four practice trials (prior to initiating the scanner), participants began the first of two runs (98 TRs each). After the first run, the experimenter checked in with the participant and confirmed that they remembered the status-color associations. Once the participant correctly recalled the status-color associations, the experimenter reiterated the initial instructions and the second run began.
Stimuli. During each run, participants formed impressions of 28 unique faces distributed across the four gender-status groupings. These faces were equated on attractiveness, likeability, emotional intensity, actual age, and perceived age. We selected faces that were predominantly categorized as their documented gender (> 90% accuracy). However, men's faces were rated with greater confidence as members of their documented gender and as more gender typical than were women's faces (see Supplemental Text S3 for stimulus equating). Because these dimensions are less directly relevant to perceived status than other attributes such as likeability, age, and attractiveness, they were not prioritized during stimulus equating. Each unique face was presented twice per run. Each run corresponded to one of two unique fMRI-optimized trial sequences generated in optseq2 43 . The selected sequences met the constraint that they did not present more than three consecutive repetitions of a given condition.
Paradigm rationale and prior validation. We chose an unconstrained impression formation task because it allows participants to form their impressions anonymously based on all available information, thereby mini-Scientific Reports | (2020) 10:20302 | https://doi.org/10.1038/s41598-020-76847-z www.nature.com/scientificreports/ mizing pressures to conform to any perceived expectations from the experimenter or society more generally 22 . Although this approach may minimize socially desirable responding, it also limits the experimenter's ability to verify that participants are indeed forming impressions. Fortunately, a body of work using similar impression formation tasks has repeatedly found them to elicit activity in brain regions supporting social cognition (e.g., MPFC) when contrasted with non-social tasks 44,45 . More recent work using this impression formation task to focus specifically on status-based evaluative responses has consistently identified preferential VMPFC responses to high (vs. low) status along several social dimensions including moral status 23,24 and SES 22 . These responses are consistent with other neuroimaging work focusing on status-based evaluations, which generally elicit greater VMPFC responses when participants evaluate high-status (vs. low-status) individuals in an array of paradigms beyond impression formation 6 .
Additional scans. After the final functional run, we acquired a resting state scan and an anatomical (MP-RAGE) scan over approximately 12 min, time permitting. The entire scanning session, including the unrelated fMRI task, was just under one hour long.
Post-scan measures and procedure. Participants exited the scanner and returned to the separate testing room to complete a series of exploratory measures and behavioral tasks. These post-scan tasks included exploratory measures of stimulus likeability and status recall for all face stimuli presented during the two scan runs (see Supplemental Text S1). After completion, participants were debriefed, paid, and excused.
fMRI acquisition, pre-processing, and GLM. We used a Phillips dStream Achieva 3 T system and 32-channel head coil to acquire BOLD, T2* contrast-weighted EPIs. With a 2000-ms repetition time (TR) and a 25-ms echo time (TE), we acquired 34 oblique slices using an interleaved z-shim acquisition protocol 46 . Slices were 4-mm thick with a 0.5-mm gap, a 3 mm 2 in-plane resolution, 77° flip angle, and a 192 × 134 × 192-mm field of view. Slices were aligned to the AC-PC axis of each participant 47 . As in prior work 21 , EPIs from each participant's two runs were pre-processed and analyzed using SPM8 and SPM8w (for details, see "Data and code availability"). We first implemented slice-time correction 48 , using the 17th slice acquisition as the reference. Subsequently, we integrated the four repeated z-shim slices 46 . The resulting images from each participant were then unwarped and realigned to the participant's mean EPI to correct for motion and motion-by-distortion interactions 49 . Images were subsequently normalized to the MNI template and smoothed with an 8-mm FWHM kernel 50 .
To estimate the BOLD responses for each condition, each trial was considered as an event, and the stimulus time series was convolved with the canonical hemodynamic response function. A GLM modeled both scan sequences concatenated as a single session with regressors for each of the four conditions (ordered as follows: high-status woman, high-status man, low-status woman, and low-status man) followed by a standard set of regressors controlling for variance associated with various sources of systematic noise 21 . Contrast images reflecting the first-level effects of interest were used in second-level analyses.
Analyses. The present analyses focused on potential effects of target gender and SES during impression formation and how these effects may be sensitive to the perceiver's own gender. Our primary analyses focused Trial procedure for impression formation fMRI task. A high (or low) SES woman's (or man's) face would appear (in place of the gray silhouettes pictured above) for 1500 ms followed by an intertrial interval (ITI) consisting of a fixation cross of a jittered duration between 500 and 6500 ms before the next face would appear. To indicate that they had formed their impressions for each face, participants responded by pressing simultaneously buttons assigned to each index finger. Participants were encouraged to finish forming their impressions by the time the face disappeared or shortly thereafter. Participants completed 112 total impression formation trials over two runs.
Scientific Reports | (2020) 10:20302 | https://doi.org/10.1038/s41598-020-76847-z www.nature.com/scientificreports/ on key regions of interest (ROI) identified above: the VMPFC, NAcc, and the amygdala. We also conducted exploratory analyses testing for any potential main effects of or interactions with the perceiver's self-reported subjective SES. Including effects of subjective status in our ROI analyses did not change the results reported here, nor did any significant effects of subjective status emerge. For full results from all supplemental analyses, see Supplemental Text S4.
In an attempt to connect activity within our primary ROIs with co-activating regions throughout the brain, we followed up on the ROI analyses with an exploratory seed PLS analysis that focused on the main effect of perceived SES in the VMPFC. We also conducted a data-driven task PLS analysis to detect coordinated activity across the brain as a function of target gender and SES, unconstrained by any particular contrast. Finally, to further probe for effects of perceiver gender, we also examined the contribution of participant gender to the network identified by the task PLS analysis. Average parameter estimates (vs. average signal response) were extracted for each condition from an 8-mm sphere (VMPFC) or a 4-mm sphere (NAcc, amygdala). The size of these ROIs is commensurate with differences in anatomical size between these regions and with ROI volumes from our previous work 21 .
ROIs were analyzed using the lme4 package for linear mixed-effects models 53 in R 54 . Degrees of freedom were estimated using Satterthwaite's approximation, provided by the package lmerTest, version 2.0-36 55 . Face stimulus coding for all models was as follows: women = − 0.5, men = 0.5, low status = − 0.5, and high status = 0.5. Perceiver gender was coded as follows: women = − 0.5, men = 0.5. We allowed for between-participants variance in intercepts (i.e., random intercepts) to account for participant-level variations in average neural response. For each ROI, we examined whether target gender, target status, perceiver gender, and all possible interactions predicted neural activity. We used Bonferroni correction to set our alpha level for the omnibus models (each of the five ROIs) to 0.01. To follow up on significant Target Status × Perceiver Gender interactions in each ROI, we tested the simple effects of status at each level of perceiver gender and the simple effects of perceiver gender at each level of status using dummy coding for target status and perceiver gender. These models accounted for variance associated with target gender by conserving the original contrast coding for target gender. For the follow-up models, we maintained a 0.05 alpha level.
PLS analysis. For this analysis, we used an existing MATLAB-based PLS analysis toolbox (https ://www.rotma n-baycr est.on.ca/index .php?secti on=84). In brief, PLS relates two blocks of data to one another. For our seed PLS analysis, one data block represented each participant's fMRI contrast image for the contrast (high-SES faces > low-SES faces), and the other data block represented each participant's average signal difference in the VMPFC ROI for high-SES faces minus low-SES faces. For our task PLS analysis, one data block represented whole-brain Blood Oxygenation Level-Dependent (BOLD) activity in the form of each participant's beta maps for each condition versus the session mean, and the other block represented the study design (i.e., the four face conditions) 38 . The goal of these analyses was to find weighted patterns (i.e., latent variables-LVs) that best explain the covariance between blocks (i.e., "cross-block covariance"). These latent variables are computed via singular value decomposition. For further details on the specifics of PLS analysis, see previous work by McIntosh and colleagues 40 .
The LVs reflect linear combinations of voxel activity throughout the brain that may be differentially instantiated for each experimental condition. To test the significance of each LV, we generated a set of 2,000 permuted samples using the same procedures and parameters reported in previous work 22 . The reliability of the original LV (i.e., its p value) is calculated as the proportion of the permuted singular values that exceed the singular value for the original LV. The reliability with which each condition contributes to the LV was determined using a bootstrap procedure. Specifically, we tested the reliability of condition-specific seed-brain correlations (for seed PLS) and brain co-activation scores (for task PLS) using 95% confidence intervals (see Figs. 2, 3). These confidence intervals were generated using a 2000-sample bootstrapping test. Because the top and bottom bounds of the confidence intervals are derived from a bootstrap distribution 22,56-58 , it is common for these bounds to be asymmetric relative to their corresponding estimates e.g., 22,57,58 . The reliability with which each voxel contributes to the overall multivariate pattern captured by the LV (i.e., the voxel's bootstrap ratio-BSR) is determined with a set of 2,000 bootstrapped samples 22 . For descriptive purposes, mapped BSR values were thresholded to approximate a 95% confidence interval, corresponding to BSR ≥ 3 or BSR ≤ − 3 for Task PLS. For the seed PLS analysis, we used a more stringent threshold of BSR ≤ − 5 in order to more clearly depict the resulting seed-based network, which is necessarily characterized by very large BSR values with increasing proximity to the seed coordinates. We used the standard cluster reporting function from xjView 97 (https ://www.alive learn .net/xjvie w), to extract and report clusters of 20 or more contiguous voxels containing BSRs that survived the above thresholds. This was not implemented for the purpose of statistical analysis, but rather to illustrate the peaks and extent of co-activating brain regions that characterize a significant latent variable.
In the present report, we used a seed PLS analysis to follow up on the univariate ROI analyses that showed main effects of perceived SES. Specifically, we examined the degree to which preferential responses to high-SES (vs. low-SES) faces in the VMPFC were associated with the extent of co-activation in participant-level wholebrain contrast images reflecting greater activity for high-SES (vs. low-SES) faces. We also used task PLS analysis to explore coordinated patterns of neural activity related to target gender and status across the whole brain without the constraint of any particular contrast (as in the seed PLS analysis). The task PLS analysis identified linear combination of voxels that optimally varied by gender and status of the faces. In our design, we had four Scientific Reports | (2020) 10:20302 | https://doi.org/10.1038/s41598-020-76847-z www.nature.com/scientificreports/ viewing conditions: high-SES woman, high-SES man, low-SES woman, low-SES man. Finally, to explore effects of perceiver gender (which emerged in our ROI analyses-see "Results"), we also tested post hoc the magnitudes of each participant's task PLS co-activation scores by condition as a function perceiver gender.
Data and code availability. Raw data and analysis scripts are available for download (https ://osf.io/pnqe7 ). These data were pre-processed and analyzed using SPM8 (www.fil.ion.ucl.ac.uk/spm) and a custom suite of scripts for fMRI analysis (https ://githu b.com/ddwag ner/SPM8w ). PLS analyses were implemented using the PLS Significant simple effects are indicated with asterisks, p < .05. Error bars indicate one SE unit above and below condition average. For full contrast statistics, see Table 1. Lateral, ventral, and medial views are displayed separately for the right hemisphere (top three images) and left hemisphere (bottom three images). Voxels were thresholded at BSR ≤ − 5. Note that the directionality of brain co-activation (B) needs to be interpreted in conjunction with the bar graph (A); lighter colored voxels indicate greater co-activation as a function of increasingly preferential VMPFC responses to high versus low SES.  Whole-brain regressions by perceiver gender. Complementing the ROI analyses, we tested for clusters across the brain that differed between men and women participants during impression formation for faces varying in gender and SES. Participant-level contrast images for the effects of target gender (woman > man), target SES (high SES > low SES), and the Target Gender × Target Status interaction (high-SES woman − high-SES man > low-SES woman − low-SES man) were analyzed at the group level using independent-samples t-tests to test for differences in these effects by perceiver gender. For these whole-brain exploratory analyses, we estimated a cluster extent threshold using AlphaSim. After 10,000 Monte Carlo simulations of all voxels at a voxel-wise false-positive rate of p < 0.001, we arrived at a cluster extent threshold of 52 voxels. It should be noted that cluster-level thresholds generated by AlphaSim do not adequately control for false positives in all conditions 59 . Therefore, these whole-brain analyses should be interpreted with caution, pending future replication.
Target gender effects. No significant clusters were observed.
Target SES effects. For the high SES > low SES target contrast, we observed five clusters where men relative to women exhibited more activity to high-SES than low-SES targets (see Table 2). Participants identifying as men exhibited greater activity for high than low SES in various areas of the prefrontal cortex; the ventral anterior cingulate cortex; the orbital frontal cortex; the cerebellum; and the middle occipital gyrus. There were no significant clusters showing more activity to high-SES than low-SES targets for participants identifying as women.
Target gender × target SES interaction. For the high-SES woman − high-SES man > low-SES woman − low-SES man interaction contrast, we observed no significant clusters for either men or women in our sample.

PLS analyses. Seed PLS.
Using seed PLS, we first explored the relationship between each participant's preferential response to high SES in the VMPFC and co-activation across the brain in contrast images reflecting high SES > low SES. Results from this seed PLS analysis revealed a significant LV (p < 0.001) that captured a relationship between the average preferential VMPFC response to high (vs. low) SES (i.e., the seed) and coactivation of an extended network in the univariate high SES > low SES contrast images. This relationship was similar for faces depicting men and faces depicting women (see Fig. 3A). The regions that contributed most to  24. We next plotted all voxels contributing most strongly to this LV (see Fig. 4B). Specifically, we observed increased neural co-activation when participants formed impressions of high-SES (vs. low-SES) faces in a broadly distributed network consisting of frontoparietal and insular brain regions, inferior/superior temporal regions, cingulate cortex, and cerebellum, together with extensive co-activation in regions like the pons and midbrain (see Table 4). The regions constituting this network largely coincided with those involved in salience and attention [60][61][62] , possibly encompassed by more extended domain-general networks supporting focused task engagement [63][64][65] . We revisit this network in greater detail in the Discussion. Following up on effects of perceiver gender in the ROI analyses, we explored whether the status-sensitive network of brain co-activation captured by the LV may be modulated by perceiver gender. To do this, we extracted participant-level design salience scores from the task PLS analysis and submitted those scores to a 2 (Target SES: low, high) × 2 (Target Gender: woman, man) × 2 (Perceiver Gender: woman, man) mixed ANOVA (Type III), with perceiver gender as a between-participants factor. Results again revealed a significant main effect of target SES, F(1, 63) = 10.79, p = 0.002, CI 95% = [19.657, 73.719]. Importantly, this effect of target SES was significantly modulated by perceiver gender, F(1, 63) = 7.44, p = 0.008 (Fig. 5). All other effects were non-significant, p > 0.25. Analyses of simple effects revealed that the effect of target SES on participant-level design salience (i.e., brain co-activation) scores was reliable for men in the sample, t (

Summary of stimulus ratings and responses.
To provide greater context on the faces on which participants formed their impressions, we assessed likability ratings, feeling thermometers, and SES recall performance for all face stimuli presented during the two scan runs. Analyses predicted likeability, warmth, or recall as a function of the participant's gender and the four task conditions. We summarize the results from those tasks here, but full analyses can be found in Supplemental Text S4. For the likeability ratings and feeling thermometers, participants rated faces of women more positively than faces of men. This effect was significant for men and women participants but was especially pronounced among women for the feeling thermometers (i.e., perceived www.nature.com/scientificreports/ warmth). We found no other significant effects involving ascribed face SES nor participant gender. The status recall results showed above-chance recognition of ascribed SES levels for the face stimuli (65.15% across all participants), despite the lack of instructions to memorize face-status pairings. SES recall was not significantly impacted by face SES, gender, or participant gender. Finally, we tested the possibility that men and women were forming impressions of target faces at different speeds. RT data were analyzed using a mixed linear model that included fixed effects for target gender, target SES, participant gender, and all possible interactions between these terms (and random effects by participant for the main effects of target gender and SES and the interaction between these terms). Results indicated that all effects from this model were non-significant, p > 0.06. Additional simple comparisons confirmed that women and men showed no significant differences in impression formation response times for any of the four face conditions, p > 0.11. Taken together, our contextual analyses of the self-report data showed some limited evidence that participants' explicit evaluative ratings were sensitive to both perceived SES and gender, albeit only for recall and likeability/warmth ratings, respectively. Perceived SES did not bias any explicit post-scan ratings of the face stimuli. Although the likeability/warmth ratings did not parallel the fMRI findings, this could be explained by key differences between the tasks. Whereas impression formation is more holistic and likely integrates a number of dimensions beyond likeability/warmth (e.g., competence), likeability and warmth judgments are more narrowly focused. Additionally, as indicated in our rationale for choosing this paradigm, the impression formation task completed in the scanner was more private in the sense that participants' responses were not indicative of their impressions, unlike in the explicit ratings tasks. Accordingly, likeability and warmth ratings are more susceptible to socially desirable responding.

Discussion
Across complementary univariate and multivariate analyses, the present findings reveal consistent evidence of greater sensitivity to status (viz., SES) in men than in women. In line with previous findings 6, 28 , we observed greater activity in brain regions indexing positive evaluations of others 26 and social reward/salience (e.g., NAcc, amygdala) as men (but not women) formed impressions of high-SES (vs. low-SES) faces. Seed PLS analyses revealed that preferential responses to high (vs. low) SES in the VMPFC were associated with greater co-activation in contrast images reflecting high SES > low SES within an extended network involved in person evaluation and mentalizing, including the precuneus, temporoparietal junction, insula, and orbitofrontal cortex. This co-activation pattern was similar for women and men, suggesting that despite a greater VMPFC preference for high SES in men, both genders show similar patterns of functional connectivity when they do show a pro-high-SES bias in the VMPFC. Beyond our focal regions of interest involved in person evaluation, an exploratory task PLS analysis revealed a coordinated network of brain regions that was sensitive to high-SES (vs. low-SES) faces. Again, this pattern was only reliably observed for men in our sample. The functional network emerging from these analyses is consistent with greater relevance and/or engagement with status cues in men. Taken together, the findings provide evidence that high versus low SES is associated with neural responses indicative of positivity, reward, and salience during impression formation for men. The present findings are noteworthy for their contribution to a network neuroscience understanding of status perception but also for their implications regarding how gender and status interact to shape our impressions of others. We elaborate on these implications in separate sections below. Additionally, by examining both target and perceiver gender, this work provides greater The main effect of target status emerged as a significant LV in the task PLS. Task PLS co-activation scores (i.e., design salience scores, see y-axis) represent the strength of the relationship between the network of brain regions and the task conditions. The error bars represent confidence intervals computed through a 2000-sample bootstrapping procedure 56 . (B) Voxel-wise patterns that increased in co-activation for high-SES versus low-SES faces. Lateral, ventral, and medial views are displayed separately for the right hemisphere (top three images) and left hemisphere (bottom three images). Voxels were thresholded at BSR ≥ 3. Note that the directionality of brain activities needs to be interpreted in conjunction with the bar graph; warmer colors indicate greater co-activation when forming impressions of high-SES faces.
Scientific Reports | (2020) 10:20302 | https://doi.org/10.1038/s41598-020-76847-z www.nature.com/scientificreports/ representation in samples than is often used for psychophysiological studies and avoids the potential pitfall of assuming that women respond to status in the same way that men do 66 . Indeed, as the present and previous work suggest, this is only sometimes the case.
Toward a network neuroscience approach to social hierarchy perception. The present study is among the first to use a multivariate brain network approach to investigate status-based impression formation. This data-driven approach revealed a functional brain network responsive to perceiving high-SES compared to low-SES people, particularly in men. This network shows considerable overlap with a previously identified set of regions involved in status-based attention 6,67 and more broadly with networks believed to support salience detection 60,62 and attention 61,62 . However, the observed co-activation network extended beyond these core networks, possibly implicating broader domain-general networks supporting focused task engagement [62][63][64][65] . Complementing our ROI approach, which focused on brain regions previously shown to support status-based evaluations 6,28 , the preferential involvement of this large brain network was also driven primarily by men in our Table 4. Task PLS (n = 65) reveals a network that co-activates more for high-SES (vs. Low-SES) targets. R right, L left, BSR bootstrap ratio-BSR indexes reliability of each cluster. All BSR ≥ 3; all clusters ≥ 20 voxels. Cluster subregions are reported to illustrate the anatomical extent of the cluster beyond the peak BSR. Follow-up analyses revealed that this network was driven primarily by participants identifying as men (Fig. 5).  Fig. 5). Taken together, these findings suggest that men may be especially engaged when forming impressions of high-SES people and that these impressions are likely positive 14 .
It is worth noting that the inferior parietal cortex emerged as one component of the status-based attention network identified through task PLS analysis. This region has been implicated in computing social distances [68][69][70][71][72] , showing greater activity for difficult comparisons between two closely ranked individuals compared to easy comparisons between very differently ranked individuals 71 . The involvement of the inferior parietal cortex in the SES-based functional network suggests that men in our sample may dedicate greater attention to differentiating SES when forming impressions of others, particularly when those individuals are high in SES. This interpretation would be consistent with findings from the human and animal literature that high-status (vs. low-status) individuals more readily attract and/or re-direct attention 6,[73][74][75][76] and are more easily remembered 76 . Implications for status and gender of the perceiver. One key takeaway in the present study is that the participant's gender consistently altered sensitivity to perceived SES. In multiple regions thought to support status-based evaluations 6,25,28 , men more reliably showed neural responses that were more indicative of positive impressions for high-versus low-status targets. Men also showed greater coordination for increasing target SES in an extended set of regions implicated in salience, attention, and task engagement as discussed in the preceding section. In other words, men more than women showed evidence of positive evaluations and more deliberative engagement with high-SES faces during impression formation. This finding is consistent with evolutionary psychological accounts suggesting that men more readily display, attend to, and/or pursue higher status 18,77,78 .
Notably, the effect of perceiver gender on neural responses to target SES was not further modulated by target gender. Based in part on predictions derived from theories of sexual selection 19 , it has been argued that women and men may prefer different dimensions of status, both for themselves 77,78 and for potential heterosexual mates 19,20,79 . Despite some support for this theory in the behavioral literature, it receives little support from the extant albeit sparse neuroimaging literature on gender and status 36 . One important caveat is that no fMRI study to date has explicitly invoked a context relevant to mating.

Implications for status and gender in person perception. Operationalizing target status. Status
can be conveyed along multiple dimensions ranging from commonly studied attributes like dominance 36 and finances 23,24,80 to social categories such as race 21,22,81 and gender 36 . One important aspect of this study is its focus on SES instead of dominance, which has perhaps received greater attention so far in neuroimaging studies of gender in hierarchical contexts 36 . As a measure of social rank, SES comprises an individual's standing in terms of education, income, and occupational prestige 82 . Although SES may convey dominance in some contexts, these two constructs are not the same 6,83 . Accordingly, neural responses to perceived dominance may not reflect responses to status when it is operationalized in terms of SES. Indeed, previous work suggests that VMPFC responses to high status depend on the dimension of status in question, with greater responses for high moral status than for high financial status 23,24 .
In addition to the dimension of status in question, it is also important to consider the means by which status is conveyed. In the present study, status was conveyed through colored cues previously paired with high or low SES. This approach is particularly important for neuroimaging studies of status due to potential confounds that exist in more naturalistic and subtle cues of status such as facial cues 84 and clothing 81,85-87 . For example, clothing can shape impressions of dominance 85 , competence 86,87 , and attractiveness 88 , all of which are imperfectly Figure 5. Men showed a larger effect of perceived SES on brain network co-activation. Participant-level design salience score estimates from task PLS that were averaged to create Fig. 4A were extracted and analyzed by participant gender. Results revealed that network sensitivity to target SES was driven primarily by men in the sample. Men showed a significant effect of SES in their brain co-activation scores, as indicated by the asterisks, p < .001. Error bars indicate a 95% confidence interval in t-tests against zero, which represents the overall mean contribution to the latent variable across all task conditions and participants. In contrast with the seed and primary task PLS analyses (Figs. 3 and 4, respectively) which utilized bootstrapping to generate confidence intervals, here we generated confidence intervals from between-participants variance in condition-specific co-activation estimates for each participant, which were the units for this analysis.  83 . This is important for neuroimaging studies of status for two reasons. First, using such cues makes it difficult to distinguish whether purported effects of social rank may instead be due to related but non-equivalent attributes such as competence. Second, this problem would be compounded when comparing the perception of two groups that stereotypically differ in terms of competence 87 . The present study largely bypasses these limitations by relying on ascribed SES knowledge through color assignment rather than clothing or facial expressions.
Absence of effects of target gender. Although the gender of perceivers shaped neural responses to target status during impression formation, these responses did not differ based on the gender of targets. This is noteworthy for a few reasons. First, participants did explicitly rate women are more likeable than men after scanning, suggesting that they did attend to gender to some extent. Additionally, previous work has revealed distinct neural correlates of target gender, regardless of whether gender is explicitly processed 36,89 . Given the relative salience of social category cues such as gender 90,91 , we anticipated that gender might interact with target status. Indeed, Marsh and colleagues 36 observed greater responses in the VMPFC and right amygdala for increasingly dominant body postures of only the woman targets. However, as previously mentioned, the fact that status was based on different dimensions (SES vs. dominance) and antecedents (perceptual attributes vs. person knowledge) may explain these differences. Furthermore, neuroimaging work examining status-based impression formation in the presence of another salient social category (viz., race) also only found effects of status 21,22 . One possibility is that participants disregarded initial impressions based on gender information by focusing on the available status knowledge to form more individuated impressions 92 .
The absence of effects of target gender is also noteworthy from the perspective of some proposals arising from evolutionary psychology. Due in part to a lower minimal male investment in child-rearing 93 , evolutionary psychologists argue that males were frequently engaged in ancestral conflicts within and between groups, resulting in greater natural selection of physical and psychological characteristics associated with dominance in males than in females 18 . This relationship between sex and social hierarchy cues may also extend to social constructions such as the clothes we wear. For example, previous work has shown that men wearing high-status attire are more readily attended 94 and rated as more competent 87 than women wearing high-status attire. In contrast with this previous work, the present study showed no unique effects of SES for faces depicting men compared to women. One possibility is that conveying status through person knowledge rather than appearance eliminates potential gender bias in how status shapes social attention and evaluations 6 .
Alternative interpretation. Prior to concluding, it is worth considering an alternative interpretation of the present work. Given that the impression formation task did not include a control task condition, one could argue that our findings are not related to the status information ascribed to each target during impression formation. In this perspective, one could imagine obtaining similar findings if the faces of women and men were grouped based on other social information than social status such as their preference for different kinds of food. If so, then our findings might mean that men are more sensitive than women to social information in general rather than social status, per se.
Notwithstanding existing research suggesting that, in some instances, women may be more sensitive to social information than men 95-97 , we believe it is more parsimonious to focus our interpretation in the context of status sensitivity, rather than sensitivity to social information more generally, for three reasons. (1) Given that the face stimuli were equated across various dimensions and counterbalanced across participants within gender, we think it would be prudent not to over-generalize our findings to other social information. (2) The present findings are consistent with the literature suggesting greater status-driven perception and behaviors in men 10,[14][15][16][17][18]98 . (3) Assuming that men were more sensitive to social information in general, it is unclear why men didn't also show greater neural sensitivity to perceived gender, a dimension that has demonstrable biological and psychological relevance as a function of perceiver gender 19,20 . The present study was focused primarily on SES as the perceived status dimension, finding evidence of greater neural sensitivity to high (vs. low) SES in men compared to women. Importantly, SES is but one possible dimension of social status, being composed of subcomponents (e.g., income, education, occupational prestige) that may or may not elicit the same kinds of evaluations 6,28,83 . Other social hierarchies that have been studied using fMRI include hierarchies based on dominance/ability/competence 31,99 , power 80,100 , and moral character 23,24 . With few exceptions 23,24 , these studies have focused on just one hierarchy dimension and often just one gender, usually men. In general, these studies frequently find the VMPFC and NAcc are responsive to high-ranking individuals. Other work by Kumaran and colleagues suggests that the VMPFC in particular may be implicated in updating social hierarchy knowledge in hierarchies involving the self 100 . However, it is unclear whether these effects are similarly driven by men as they were for SES in the present study. In future research, it will be important to replicate and extend the present work to determine whether men also show greater neural sensitivity to high status compared to women for other status dimensions as well as for social dimensions that are potentially ordinal but not frequently used to delineate hierarchies (e.g., age).
In sum, we believe our present findings can be interpreted as a gender-specific differential in sensitivity to perceived status. However, we do not believe that men's greater sensitivity to SES in this study precludes the possibility that women (or men) might be more responsive to other social information in other contexts. For example, it is possible that subsequent studies may uncover a preference in women for other dimensions of status such as communal/moral character 23,24,77 or that women may value high-SES individuals more than men in certain contexts. These questions among others are ripe for future inquiry and consistent with recent calls to explore how perceiver gender may differentially shape sensitivity to other dimensions of social information 66 . We hope that this study will pave the way for future work exploring gender differences in sensitivity to different forms of social information.

Conclusion
The social psychological literature is full of examples illustrating the barriers faced by women who seek to ascend the social hierarchy 88,101 . The present findings help to identify potential neurobiological constraints that may be accounted for in interventions aimed at reducing the barriers to and promoting women's participation in leadership 101 . Specifically, this study provides evidence that men may place greater emphasis on SES than do women when they privately form impressions of others. Specifically, men showed greater responses to high versus low SES in brain regions shown to index positive status-based evaluations, such as the VMPFC and NAcc 6,25,28 . Moreover, men showed greater coordinated activity toward high-SES faces in a functional network comprising brain regions implicated in salience, attention, and/or task engagement. These findings are consistent with proposals that men are more sensitive to status than women 10,17,18,77,102 . However, as with any research uncovering gender differences, it is important to interpret the present findings with caution 103 . Just because women showed less sensitivity than men to ascribed SES in the present context does not mean that women are generally insensitive to social status during impression formation. Indeed, similar seed PLS results for men and women in the present study suggest that both genders may use a similar functional connectivity network for evaluating SES during impression formation even if they may differ in their average VMPFC responses to SES.
Diversity statement. Recent work in psychology and neuroscience has identified a bias in citation practices such that papers from women and other minorities are under-cited relative to the number of such papers in the field 104,105 . Here we sought to proactively consider choosing references that reflect the diversity of the field in thought, form of contribution, gender, and other factors. We obtained predicted gender of the first and last author of each reference by using databases that store the probability of a name being carried by a woman 105,106 . By this measure (and excluding self-citations to the first and last authors of our current paper), our references contain 23.3% woman(first)/woman(last), 12.2% man/woman, 14.4% woman/man, 50% man/man, and 0% unknown categorization. This method is limited in that a) names, pronouns, and social media profiles used to construct the databases may not, in every case, be indicative of gender identity and b) it cannot account for intersex, non-binary, or transgender people. We look forward to future work that could help us to better understand how to support equitable practices in science.

Data availability
Raw data and analysis scripts are available for download from https ://osf.io/pnqe7 .