New research provides evidence that people generally have a positive implicit bias towards women and a negative implicit bias towards men, as well as a similar but less consistent implicit bias in favor of people from higher social classes. But the new research, which examined data from nearly 6,000 individuals, found inconsistent evidence for implicit biases based on race.
The findings shed new light on intersectional implicit bias and have been published in the Journal of Personality and Social Psychology.
Implicit Association Tests (IATs) are a type of psychological tool designed to measure implicit biases (unconscious attitudes) that people may hold towards certain social groups or concepts. IATs typically involve participants quickly categorizing stimuli, such as words or images, into different categories using a computer keyboard or mouse. The basic idea behind the test is that people will be faster and more accurate when categorizing stimuli that are more closely associated in their minds, compared to stimuli that are less closely associated.
These tests have provided evidence that people have implicit biases towards different social categories, like race and gender. But individuals often have multiple identities that intersect, like being a woman of a certain race or social class. However, studies on implicit biases usually only focus on one identity at a time. The authors of the new study sought to investigate implicit biases in the context of multiple intersecting social identities.
“I’ve always been interested in stereotyping and prejudice based on social class, ever since I was quite young,” said study author Paul Connor, a postdoctoral scholar with the Adversarial Collaboration Project at the University of Pennsylvania.
“Starting in grad school, this interest led me to develop a large database of full-body photographs of individuals displaying varying levels of perceived social class status, but also differing in terms of their perceived race, age, gender, and more. Having those images then led me to the idea that I could use them to investigate the relative extent to which people’s implicit evaluative biases are driven by different target-level variables.”
“For example, to what extent are biases driven by people’s race, gender, age, or perceived social class? I also realized that we didn’t know much about this question, despite it arguably being fundamental to understanding how implicit biases operate in real-world social situations,” Connor explained.
The researchers conducted four studies on intersectional implicit bias.
For their first study, the researchers gathered 130 full-body color photographs of Black and White adults and presented them to 1,788 U.S. adults recruited via MTurk. These participants were asked to rate the photographs on perceived yearly income, perceived age, and whether they perceived the targets to be Black or White.
Based on the photographs’ ratings, the researchers assembled groups of eight photos varying in race and income but matched in age, which were used in Implicit Association Tests completed by 840 undergraduate students. Connor and his colleagues found that participants showed a bias in favor of targets with higher perceived incomes. The implicit biases were not significantly affected by target groups’ race.
For their second study, the researchers expanded their examination of how people evaluate individuals based on social categories such as race, social class, gender, and age. They selected 54 images of Asian, Black, and White targets from a large database of 726 full-body target images. The database contained explicit ratings made by 3,311 U.S. adults on 24 different personality and demographic traits. The researchers then selected nine female and nine male targets of each race, varying in social class and age.
After testing 371 undergraduates, the researchers found that people’s implicit evaluations were mostly influenced by the combination of the target’s gender and social class. They observed that targets who were upper-class females received particularly positive evaluations.
For their third study, Connor and his colleagues sought to improve the experimental control over their target stimuli. They used 24 unique faces from the Chicago Face Database, varying in race, gender, and age, and swapped them onto 24 bodies that varied in gender, age, and perceived social class. There were no significant differences in perceived attractiveness between the race, age, or gender groups, and no significant differences in perceived age, social class, or income between the gender or age groups.
After examining responses from 1,527 undergraduates, the researchers again found that gender “exerted effects many times larger than” race, social class, and age.
The first three studies utilized undergraduate samples, which might limit the generalizability of the results. To address this issue, Connor and his colleagues recruited a nationally representative sample of 2,466 U.S. adults via Prolific. The researchers also employed three different tests of implicit bias: single-target IATs (ST-IATs), evaluative priming tasks (EPT), and affective misattribution procedures (AMP).
The researchers again found that gender was the most significant factor affecting implicit evaluations, with female targets receiving more positive evaluations than male targets across all three methods used to measure implicit evaluations. The study also found consistent but smaller effects of target social class, with upper-class targets receiving more positive evaluations than lower-class targets via all three methods.
The effects of race were found to be inconsistent, with participants favoring White and Asian targets over Black targets in ST-IATs, and Asian targets over White and Black targets in EPTs, while Asian and Black targets were favored over White targets in AMPs.
Connor and his colleagues then analyzed the data from Studies 2-4 to investigate if the patterns of results differed among different subgroups of respondents. They found that all subgroups of respondents displayed a pro-female and anti-male bias, as well as a pro-upper-class and anti-lower-class bias, with women showing a stronger bias towards social class. These results suggest that while the magnitude of implicit biases may vary across demographic groups, the basic directions of these biases remain stable.
“Our headline finding was that it was predominantly targets’ gender that drove implicit responses, not race, age, or social class,” Connor told PsyPost. “Specifically, we found that pro-female and anti-male biases explained much more variation in participants’ responses than any other kind of bias.”
“This was surprising, because there is scarcely any previous work in our field that would have led us to predict that gender would have dominated to the extent that it did. So, although we don’t think this means that implicit bias is actually mainly driven by gender, we do think it indicates that our field is still quite far from having a well-developed understanding of implicit bias, especially as it pertains to complex, multiply categorizable social targets.”
But why the dominance of gender in implicit evaluations? The researchers believe it might be related to heightened visibility. Gender was conveyed through both faces and clothing in the stimuli used in the study, while race was only conveyed through exposed skin. Even if this is the underlying mechanism, the researchers said, the results should still apply to real-world interactions as gender is typically visible through both faces and clothing in everyday contexts.
As far as limitations, Connor said that “one important caveat to recognize is that we only studied implicit evaluations, associations between social categories and positive and negative concepts. Implicit stereotypes, such as associations between social categories and competence, leadership, or specific social roles, comprise a whole area of research on their own, and our work did not touch on any of this.”
“Another is that this work is really still in its very early stages. I would love to continue to try to better understand the construct of implicit bias and how it operates in the world, but I think we have a very long way to go before we have a firm understanding in this area.”
“When I discuss this work with people, their reaction is often to ask if others within the academic community have responded negatively to the results,” Connor added. “And I understand the question, because ever since I started doing this work, I have felt some trepidation about sharing the results. But I have to say that virtually everyone, from editors, to peer reviewers, to academic audiences, has been genuinely receptive and willing to engage thoughtfully with these data, even if they might be surprising or run counter to some people’s priors. It has been a very welcome and pleasant surprise.”
The study, “Intersectional Implicit Bias: Evidence for Asymmetrically Compounding Bias and the Predominance of Target Gender“, was authored by Paul Connor, Matthew Weeks, Jack Glaser, Serena Chen, and Dacher Keltner.