A Lower-Class Advantage in Face Memory

People remember what they deem important. In line with research suggesting that lower-class (vs. higher class) individuals spontaneously appraise other people as more relevant, we show that social class is associated with the habitual use of face memory. We find that lower-class (vs. higher class) participants exhibit better incidental memory for faces (i.e., spontaneous memory for faces they had not been instructed to memorize; Studies 1 and 2). No social-class differences emerge for faces participants are instructed to learn (Study 2), suggesting that this pattern reflects class-based relevance appraisals rather than memory ability. Study 3 extends our findings to eyewitness identification. Lower-class (vs. higher-class) participants’ eyewitness accuracy is less impacted by the explicit relevance of a target (clearly relevant thief vs. incidental bystander). Integrative data analysis shows a robust negative association between social class and spontaneous face memory. Preregistration (Studies 1 and 3) and cross-cultural replication (Study 2) further strengthen the results.


Demographics.
Participants in all three studies filled out a short demographic questionnaire including age, gender, and ethnicity.In order to assess ethnicity in Germany, we asked participants about their own and their parents' place of birth.

Additional Measures Study 2
Demographics.In addition to asking participants about their age, gender, and birthplace, we also asked participants about the zip code of the area where they are currently living and the zip code of the area where they spent most of their time while growing up.We also asked them about their political orientation and their religiosity.
Communal Orientation Scale.We asked participants to answer a 14-item communal orientation scale (Clark et al., 1987).The first author of this paper translated the scale from English to German.

Additional Measures Study 3
Object memory.We asked participants about their memory of objects and other peripheral details of the scene with these 7 questions: "What color pants was the thief wearing?", "There was a cup on the desk.Did it have liquid in it?","Was there a desktop computer on the desk?", "Were any of the windows open?", "Did you see a stapler on the desk?", "The thief carried one object out in their hand.Did they carry it in their right or left hand?", and "Did the person who passed the window come from the right or the left?"We created a continuous variable (called "object memory" in the main text) by summing all correct responses to the 7 questions (0 = no correct answers, 7 = all answers correct).
Confidence rating.After both target identifications (i.e., thief and bystander), we asked participants how confident they were in their identification with one item: "How confident are you that the person you selected is actually the thief/bystander in the video?" (1 = Not at all confident, 7 = Extremely confident).

Additional Analysis Study 1
Binary race covariate.Given the relatively small sample of non-white participants in Study 1, we reanalyzed the data using a dichotomous race variable (0 = non-white participants, 1 = white participants).The results remain substantively unchanged with this coding scheme (B = -0.032,SE B = 0.009, t = -3.52,p < .001).
Preregistered analysis.We preregistered an ordinary least squares (OLS) regression analysis utilizing a difference score instead of the mixed-effects logistic regression reported in the main manuscript (https://osf.io/pjxq6).First, we calculate an eyewitness testimony (EWT) difference score (as preregistered): EWTdiff = EWTbystander -EWTthief.As predicted, we find a negative association between social class and EWTdiff, controlling for participants' gender, participants' race, and participants' memory for objects, B = -.090,SE B = .045,t = -2.01,p < 0.05.While both analyses-the preregistered analysis and the analysis reported in the main manuscript-yield the same results, we decided to report the mixed-effects logistic regression because it is considered a more conventional analysis for data that is nested within participants.

Note.
Target was coded such that 0 = thief and 1 = bystander.Ethnicity and gender were dummy-coded with White and male acting as the reference categories.Object memory was coded as a continuous variable with higher scores equaling better memory.

Early termination of a replication study with a modified Berlin Face Test
We included our social class variables in a study that was designed to test the psychometric validity of a modified Berlin Face Test, in which we aimed to test the association between explicit/incidental face memory and social class.This modified Berlin Face Test included a new set of facial identities and faces were presented in color (previous test versions presented black and white stimuli).The new stimuli came from the Berlin Emotion Dataset and the Berlin Face Dataset (Wilhelm et al., 2014).
Unfortunately, the COVID-19 pandemic effectively halted our ability to gather these data in the lab, and we were forced to end data collection prematurely.This resulted in a smaller sample (N = 154 instead of a preregistered N = 250) than needed to test our hypotheses based on an a priori power analysis.The sample was also skewed in terms of education and age.
Participants were younger (ages 18 to 36) and more educated (92% of participants had received a college education) than the general population.

Materials and Procedure
Incidental

Results
We examined bivariate correlations between social class and the two types of face memory.No bivariate relationship emerged between social class and explicit face memory, r(151) = .11,p = .18or social class and incidental face memory, r(151) = -.06,p = .50.Including covariates and using SEM did not substantively change the results; thus, we will not discuss them in detail here.Although we did not find the predicted negative association between social class and incidental face memory, this may be due to both our reduced power to detect a reliable effect in this study and our use of modified versions of the standardized face test battery.
face memory and explicit face memory were assessed with a modified version of the standardized face test battery used in Study 1 and Study 2 (described above).The tests were programmed in Psychopy version 3.0.7,with other questions completed via Unipark.de,and testing was conducted in small group settings in a computer lab.Participants' social class was assessed with the same indicators used in Studies 1-3.

Table S1 .
Descriptive statistics and bivariate correlations between modeled variables in the SEM (Study 2).

Table S2 .
Results of the association between social class and eyewitness accuracy in Study 3 (adjusting for demographic covariates).