Bias and Racism: A Social Learning Perspective

Abstract

Social learning is thought to be an integral part in bias formation. The implicit association test is considered the holy grail of implicit bias, and yet, there are many criticisms of the IAT. Social learning of bias is one intersection of these two topics that while investigated, fails to consider all possible options. In this proposal I present the current literature on the social learning of bias, review some of the issues with the IAT, and propose a novel investigation of the social learning of bias using a social learning intervention paired with the implicit association test. I present a proposed analysis plan, review the paper as whole and present potential outcomes of the study, and discuss limitations of the proposed study, methodology, and the work as whole.

Bias and Racism: A Social Learning Perspective

Racism and bias are issues covered in varying degrees among varied academic disciplines. Whether its psychology, sociology, anthropology, or others, the topic of bias is deeply investigated on many levels in the literature. To begin, I first want to address what exactly is meant by bias and racism. First, bias is “a personal and sometimes unreasoned judgement, or an instance of prejudice” and is not always negative (Merriam-Webster, n.d.). Racism on the other hand is a specific type of negative bias in which a person or people are biased against specific racial categories, usually not their own. Racism generally involves not only the belief that other racial categories are inferior to another, but also involves actions and systems that perpetuate that belief (Merriam-Webster, n.d.).

Social learning is the process of learning through social means such as observation, or instruction, and is an efficient process for learning for both human and non-human animals (Golkar, Castro, & Olsson, 2014). The literature on social learning covers a wide array of research. For instance, research on social learning provides evidence that fear responses to things such as certain sounds, shapes, or situations is socially learned through observation and instruction in both human and non-human animals (Cook et al., 1985; Hygge & Ohman, 1978; Olsson & Phelps, 2004; Jeon et al., 2010; Kavaliers, Colwell, & Choleris, 2005). Further, social learning has been expanded upon to include not only evolutionary systems such as fear responses, but also group processes such as the ingroup vs outgroup dynamic of social categorization. Research on this topic has found that ingroup members are taken more seriously and are learned from better than outgroup members (Golkar, Castro, & Olsson, 2014).

The goal of this paper is to introduce some of the literature on racism and bias, and how they relate to social learning, to provide some research ideas on the social learning of bias, and to present some discussion points for consideration on the topic.

Implicit Bias

Literature on racism and bias is very diverse. One of the most recognizable topics comes from the field of psychology: implicit bias. Implicit bias is an unconscious process that reflects both conscious and unconscious ideas that we hold about something, someone, or somewhere (Greenwald & Krieger, 2006). The notion of implicit bias follows the ideas of implicit cognition in which many mental processes are considered to function implicitly, or outside conscious awareness. Psychology as a science has long been concerned with the unconscious, or implicit, mental processes beginning in Freud’s work on psychoanalytic theory and continuing today with empirically reached conclusions about mental processes (Greenwald & Krieger, 2006).

Implicit bias is most often thought of in the realm of racism, or race-based bias, especially in recent years. Literature on race-based biases has covered a great deal of ideas and conceptualization of implicit bias. Research by Dotsch et al. (2008) found that mental templates of out-group faces tend to be perceived as less trustworthy by participants with high levels of implicit bias. Further, differences in emotion recognition occur as well with those individuals higher in implicit bias rating black faces as more “angry” than white faces (Hugenberg & Bodenhausen, 2004).

However, implicit bias research is not just racial in nature. For instance, research by Megreya (2015) found that the categorization of faces differs when wearing headscarves. The results showed that women wearing headscarves were categorized at a faster rate, and men wearing headscarves were categorized a drastically slower rate than without them. The implicit idea that “women wear headscarves, men do not wear headscarves” was put to the test and was found to be a strong determinant of facial categorization in this instance. Further evidence of implicit biases in instances that are not racial in nature comes from a systematic literature review by Robb and Stone (2016). The review determined that implicit biases are extremely prevalent against people with mental illness and that implicit bias interventions showed no real effects on improving the biases. Even more evidence of the extension of implicit bias to nonracial categories comes from Hansen et al. (2019) who found that men and women resident physicians in the U.S. showed implicit bias favoring men as leaders in the medical field, and that differences in the level of bias was found depending on what type of medicine (emergency medicine vs OB/GYN) the participants worked in.

While the presentation of the above findings is interesting, it is important to note how we measure implicit bias. To measure this difficult and often challenging topic, we utilize the implicit association test, or IAT. The standard Implicit Association Test (IAT) is usually utilized to test implicit bias towards faces of different racial categories (i.e., white versus African American) (Nosek, Greenwald, Banaji, 2007). Essentially, the longer it takes for you to categorize certain face/word pairs, the more bias you have towards that type of face, and thus more implicitly biased you are.

While many researchers champion the IAT as the best measure of implicit bias attitudes, there has been much criticism of the IAT from its psychometric properties to its potential links to learning and conditioning (i.e., conditioning responses of certain kinds). Blanton et al. (2006) presented a review of the major criticisms of the IAT and supported these criticisms through an analysis of the IAT’s methodology and assumptions. They determined that the IAT is not a completely valid or consistent measure and that the IAT tends to make questionable assumptions and that the measure does not consider simple issues such as cognitive processing time in its assumptions and analysis. Further, psychometric properties of the IAT are called into question as evidence has been found that the IAT measure does not predict explicit attitudes nor real behaviors, and that reality often contradicts the IAT scores. Evidence for this comes from Schimmack (2019) who found that anti-black bias on the IAT actually predicted pro-black behavior in real situations. This runs contrary to the proposed ideas of the IAT that implicit attitudes are related to explicit and real behaviors that follow the implicit attitudes measured.

Social Learning of Bias

The foundations of bias and racism are thought to come, at least partly, from social learning (Hjerm, Eger, & Danell, 2018). Devine (1989) proposes that prejudicial socializing begins in childhood via exposure to prejudicial attitudes from socializing agents such as parents, teachers, and peers. A review from Over and McCall (2018) propose that social learning is valuable asset in socialization processes that also affects the learning of bias, stereotypes, and discriminatory behavioral patterns, but that it works in conjunction with cognitive processes and social categorization processes. However, despite the unsure nature of whether social learning is alone, or works in collaboration, to promote social learning of bias, there is ample evidence that shows social learning is indeed a factor in the learning of bias and discrimination.

For example, social learning surrounding threats and friendly behaviors are partially shaped by ingroup/outgroup ethnic face differences. Implicit biases have been found to have a much greater effect on avoidance of threatening behaviors from both ingroup and outgroup faces (Lindstrom et al., 2014). However, avoidance behaviors were stronger for outgroup faces than ingroup. Further, individual differences in racial bias play a role in reinforcement learning with higher racial bias associated with learning of avoidance for racial outgroup faces. More evidence from Golkar, Björnstjerna, and Olsson (2015) shows that previous exposure to out-group racial categories is an important factor in learned responses to racial out-group faces. Further, past experiences play a key role in responses with negative experiences leading to more negative future reactions and positive experiences leading to more positive future reactions.

Mosher and Scodel (1960), as well as Carlson and Iovini (1985), found that parents’ social distance from African Americans and ethnocentrism correlates with children’s aversion to the same ethnic backgrounds. Essentially, the results showed that parents who were more likely to isolate themselves from ethnic diversity tended to act more negatively towards the diversity, and their children demonstrated these same behaviors. Further research by Sinclair, Dunn, & Lowery (2005) have demonstrated a clear link between parents’ and children’s’ biases where greater prejudice in parents is directly related to higher levels of similar bias in children. Moreover, research by Skinner, Meltzoff, & Olson (2017) found evidence for social learning of bias in children. In experiment 1, nonverbal bias was demonstrated, and children showed explicit preferences for those who were treated more positively than negatively in the examples. Experiment two demonstrated that children generalized learned bias towards other people, not just the one shown in the demonstration.

Can bias be unlearned?

While it is clear from the literature that there is ample evidence for the social learning of bias, stereotypes, and discriminatory behaviors, what is not as clear is whether we can unlearn that learned bias. Literature on the idea is sparse but some does exist. For instance, research by Hu et al. (2015) show evidence that memory processes occur while sleeping that help consolidate recent experiences into our mental models. Results demonstrated that by pairing the learned information with a sound, then presenting that sound when individuals slept, the information was consolidated more readily during sleep, and participants were able to more easily recall and utilize that information. Further, the reduction in bias seen form the information lasted longer after the tone-reinforced consolidation.

Further evidence of the role that learning can play in bias comes from Livingston and Drwecki (2007). Their results demonstrated that some individuals respond differently to neutral conditioning of affective experiences. Nonbiased individuals were less likely to incorporate negative experiences into their mental representations, and more likely to incorporate positive experiences, than their biased counterparts, which demonstrates a distinct association between learning processes such as reinforcement to the learning of bias.

There is also the idea that implicit attitudes (as well as explicit attitudes) can be changed via counter-conditioning, in which we attempt to condition the opposite of the targets current attitudes towards something (such as conditioning a biased person to be less biased). In a study by Flint, Hudson, and Lavallee (2013), the counter-conditioning approach was tested towards reducing anti-fat biases. Results demonstrated that the biases held towards obesity and obese people were not able to be changed via counter-conditioning procedures, and in fact, the counter-conditioning procedures actually worsened the bias. Beyond this literature, a review of interventions to reduce bias by FitzGerald et al. (2019) presented 47 different interventions and their effectiveness. Most of the interventions were effective, however, few of the studies utilized learning procedures, and none of the studies reviewed attempted to use social learning to alter bias. This review demonstrates a clear methodological gap in the literature.

Method

The goal of this proposed research is to fill a methodological gap in the literature by asking the question: Can we unlearn a learned bias? To address this gap, a study (or series of studies) is proposed that will use social learning concepts to attempt to recondition an individual to act more positively towards a group or idea that is generally averse to them.

Participants

Participants will be collected as randomly as possible from a sample of undergraduate college students. It is important that participants have a demonstrated bias towards the same specific thing and so participants will be gathered based on sports team affiliation. Participants will be gathered that are fans of two different rival sports teams or not fans of either team (as a control). Participants will be as equal as possible in gender, race, sexual orientation, and any other demographics that may affect the results (to prevent effects based solely on these demographic factors). Participants will be recruited from local university research pools, social media posts, flyers, and word of mouth recruitment.

To ensure that our sample statistics come as close as possible to our population parameters and our statistical inferences are valid and consistent, we will gather at least 90 participants per condition, for a total of 360 participants. This number was found using the a priori procedure for sampling precision (Trafimow, 2017; Trafimow & MacDonald, 2017), using a .95 confidence level and a proposed precision level of .3, which is the minimum precision level necessary to be considered a precise sample.

Materials

Social Learning Effects- How do we measure the social learning of bias? Well, as mentioned earlier, the IAT is not a very good measure of implicit bias and the potential for biased behavior. However, there is evidence that the IAT measure can be a consistent measure of learning effects (Mitchell et al., 2003). Further research by Quek and Ortony (2012) found that the IAT can further demonstrate learning effects acquired via evaluative conditioning and other learning procedures. To test the effect of our social learning intervention, I propose the use of the implicit association test as a pretest and posttest measure of implicit bias. To be clear, not as a measure of the existence of the bias, but rather a measure of the effect of our social learning intervention (i.e., how different are they from time 1 to time 2?).

Social Learning Intervention- The social learning intervention will be developed so that it demonstrates a positive interaction towards a specific sports team (depending on which teams are chosen for participant recruitment purposes). This intervention will be a video which demonstrates a positive interaction with towards the team that the participant is most biased against. The video will show a simple interaction between a neutral person and the opposing team’s fan. The purpose of this intervention is to demonstrate a positive interaction on which the participant can base social learning of positive interactions towards the team member. The interaction will involve the neutral individual to prevent any minimal group or ingroup/outgroup effects.

Design/Procedure

Participants will be recruited who are fans of two different sports teams as well as fans of neither sports team (preferably fans of no sports teams). Participants will provide informed consent and then will be placed into the condition that matches their respective sports team affiliation (i.e., sports team 1, sports team 2, team 1 control, or team 2 control). Each participant will take an implicit association pretest created for the specific sports teams used in the interventions. After the pretest, participants will then undergo a week-long intervention regiment in which they will be presented the social learning intervention video each day at around the same time, followed by a posttest IAT. After the week is over, participants will be presented with a final video intervention, a final IAT posttest, and then be debriefed.

Results

To analyze the data, I propose a mixed methods analysis of reaction time differences and conditions. Since participants will receive the IAT several times over the course of a week, the analysis would treat the IAT measures as repeated measures (i.e., time 1, time 2, time 3, etc.) and then treat the condition as between subjects. The goal of this analysis is to see if there is an improvement across the IAT scores based on the intervention. First, a comparison of the mean IAT scores would demonstrate a simple difference in whether the social learning intervention had the intended effect. Following this comparison, a repeated measures ANOVA would be conducted to examine differences between conditions across the various IAT test instances. Finally, post-hoc comparisons would be conducted to determine if the differences were for just certain conditions, or if for all (or none) participants.

Discussion

The goal of this paper was to present the current state of the literature on social learning of bias, and to propose a study that would allow us to fill a methodological gap in the literature. As it stands there is much evidence that social learning is an integral part of bias, discrimination prejudice, and their accompanying behaviors. To fill an existing gap in the literature, I proposed a study that would address the lack if social learning interventions in the learned nature of bias.

The literature itself showcases many instances of social learning of bias in both racial (Sinclair, Dunn, & Lowery, 2005; Lindstrom et al., 2014; Skinner, Meltzoff, & Olson, 2017) and nonracial examples (Megreya, 2015; Robb & Stone, 2016). Alongside this ample literature, there is much literature on interventions for reducing implicit bias (for a review of interventions see FitzGerald et al., 2019). However, the effectiveness of social learning interventions is not as well researched. To address that methodological gap this study sought to utilize a social learning intervention paired with the IAT as a measure of social learning effectiveness. The results of this study could come out several ways, but below I cover two of the more extreme possible outcomes.

First, let’s consider the idea that the intervention proposed in this study not only works, but works well, in that participants show a drastic improvement in their implicit attitudes towards a team that they are biased against. The implication for this is two-fold. First, this type of result would show clear evidence that bias is a social learning phenomenon. While research does indeed show that social learning is an important part of bias formation (Devine, 1989; Hjerm, Eger, & Danell, 2018; Over and McCall, 2018), the existing literature is confident in that it is only a part, not even necessarily a major part, just a part of bias formation. If this study works as intended, and participants show a drastic improvement in implicit biases as hoped for, then this would demonstrate the social learning plays a much larger role in bias formation than currently thought. Further, if the study works as intended, then evidence will be found that the implicit association test, which has been greatly criticized by many researchers for its apparent post-hoc rationalization of its results among many other issues (for a review see Blanton et al., 2006), the IAT can successfully be used to demonstrate the effects of social learning interventions.

Second, the implications if it does not succeed as intended should also be considered. Of course, there are some methodological limitations of this study on its own, however, if the study does not work as intended, and participants show little to no changes in their implicit attitudes as hoped for, then it would seem that social learning plays a smaller role in bias formation than we thought. Further, if the implicit attitudes get worse, then it would show that a social learning intervention would perhaps be harmful in these efforts and help to reinforce the bias or prejudice that exists.

Finally, let’s review some limitations of the proposed study. The applied, in-person nature of this study would work as a detriment in itself as participant attrition rates would be higher than for a simple survey. Further, the longitudinal nature of this study would also likely increase attrition rates. Next, we have issues in methodology. To my knowledge, no other studies have done this type of targeted research. If that is indeed the case, then some trial and error about what works best for interventions would of course exist. Beyond these simple issues, there is also the issue of recruitment. The study depends on participants having specific implicit (or explicit) attitudes towards specific sports teams. This of course is a potential limitation that would affect recruitment as well as results, especially if the participants are not fans of team 1 or team 2, and are actually a fan of a team 3 which hates both of the study teams.

On this note, it is also important to mention that the goal of the participants who are not fans of any team is to determine if the intervention affects a neutral party in the same way as someone who is against a team. Finally, the issues with the IAT are something that must be considered. The IAT has much criticism (Blanton et al., 2019), however, there is literature showing that the IAT can be used to show learning effects (Mitchell et al., 2003; Quek and Ortony, 2012). The limitation of this however is that is has not been used to demonstrate social learning effects, and as such, this may limit the interpretation of any findings.

Overall, this paper demonstrates the relatively important nature of social learning in bias formation. It has been shown that bias is not just racial, but also nonracial in instances such as gender, weight, and even mental illness. For these reasons, this proposed study would fill not only a methodological gap in the literature, but also a conceptual gap, theoretical gap, and a practical-knowledge gap as well. 

References

Blanton, H., Jaccard, J., Gonzales, P. M., & Christie, C. (2006). Decoding the implicit association test: Implications for criterion prediction. , 42(2), 192–212. doi:10.1016/j.jesp.2005.07.003

Carlson, J. M. & Iovini, J. (1985). The transmission of racial attitudes from fathers to sons: a study of blacks and whites. Adolescence, 20(77), 233-237.

Cook M, Mineka S, Wolkenstein B, Laitsch K. (1985) Observational conditioning of snake fear in unrelated rhesus-monkeys. Journal of Abnormal Psychology, 94, 591–610. doi:10.1037/0021-843X.94.4.591

Devine, P. G. (1989). Stereotypes and prejudice: Their automatic and controlled components. Journal of Personality and Social Psychology, 56(1), 5-18.

Dotsch, R., Wigboldus, D.H.J., Langner, O., and van Knippenberg, A. (2008). Ethnic out-group faces are biased in the prejudiced mind. Psychological Science. 19, 978–980.

FitzGerald, C., Martin, A., Berner, D., & Hurst, S. (2019). Interventions designed to reduce implicit prejudices and implicit stereotypes in real world contexts: a systematic review. BMC Psychology, 7(29) doi:10.1186/s40359-019-0299-7

Flint, S. W., Hudson, J., & Lavallee, D. (2013). Counter-Conditioning as an Intervention to Modify Anti-Fat Attitudes. Health psychology research, 1(2), e24. https://doi.org/10.4081/hpr.2013.e24

Greenwald, A. G. & Krieger, L. H. (2006). Implicit Bias: Scientific Foundations, California Law Review, 94(4), 945-967.

Hansen, M., Schoonover, A., Skarica, B., Harrod, T., Bahr, N., & Guise, J. M. (2019). Implicit gender bias among US resident physicians. BMC medical education, 19(1), 1-9. Doi: 10.1186/s12909-019-1818-1

Hjerm, M., Eger, M. A., & Danell, R. (2018). Peer attitudes and the development of prejudice in adolescence. Socius: Sociological Research for a Dynamic World, 4, 1-11.

Hu, X., Antony, J. W., Creery, J. D., Vargas, I. M., Bodenhausen, G. V., & Paller, K. A. (2015). Unlearning implicit social biases during sleep. Science, 348(6238), 1013–1015. doi:10.1126/science.aaa3841

Hugenberg, K.,& Bodenhausen, G.V. (2004). Ambiguity in social categorization: the role of prejudice and facial affect in race categorization. Psychological Science. 15, 342–345.

Hygge S, Ohman A. (1978). Modeling processes in acquisition of fears: vicarious electrodermal conditioning to fear-relevant stimuli. J. Pers. Soc. Psychol. 36, 271–279. doi:10.1037/0022-3514.36.3.271

Jeon D, Kim S, Chetana M, Jo D, Ruley HE, Lin SY, Rabah D, Kinet JP, Shin HS. (2010) Observational fear learning involves affective pain system and Cav1.2 Ca2þ channels in ACC. Nat. Neurosci. 13, 482–488. doi:10.1038/nn.2504

Kavaliers M, Colwell DD, Choleris E. (2005). Kinship, familiarity and social status modulate social learning about ‘micropredators’ (biting flies) in deer mice. Behav. Ecol. Sociobiol. 58, 60–71. doi:10.1007/s00265-004-0896-0

Megreya, A. M. (2015). The effects of a culturally gender-specifying peripheral cue (headscarf) on the categorization of faces by gender. Acta Psychologica, 158, 19–25. https://doi.org/10.1016/j.actpsy.2015.03.009

Merriam-Webster. (n.d.). Bias definition & meaning. Merriam-Webster. Retrieved April 17, 2022, from https://www.merriam-webster.com/dictionary/bias

Mitchell, C. J., Anderson, N. E., & Lovibond, P. F. (2003). Measuring evaluative conditioning using the Implicit Association Test. Learning and Motivation, 34(2), 203–217. https://doi.org/10.1016/s0023-9690(03)00003-1

Mosher, D. L., & Scodel, A. (1960). Relationship between ethnocentrism in children and the ethnocentrism and authoritarian rearing practices of their mothers. Child Development, 31, 369-376. doi: 10.1111/j.1467-8624.1960.tb04972.x

Nosek, B. A., Greenwald, A. G., & Banaji, M. R. (2007). The Implicit Association Test at Age 7: A Methodological and Conceptual Review. In J. A. Bargh (Ed.), Social psychology and the unconscious: The automaticity of higher mental processes (pp. 265–292). Psychology Press.

Olsson A, & Phelps EA. (2004). Learned fear of ‘unseen’ faces after Pavlovian, observational, and instructed fear. Psychol. Sci. 15, 822–828. doi:10.1111/j. 0956-7976.2004.00762.x

Over, H. & McCall, C. (2018). Becoming us and them: Social learning and intergroup bias. Social and Personality Psychology Compass, 1-13. doi:10.1111/spc3.12384

Quek, B., & Ortony, A. (2012). Modeling the Effect of Evaluative Conditioning on Implicit Attitude Acquisition and Performance on the Implicit Association Test. Proceedings of the Annual Meeting of the Cognitive Science Society, 34. Retrieved from https://escholarship.org/uc/item/4nd6r6nb

Robb, J., & Stone, J. (2016). Implicit bias toward people with mental illness: A systematic literature review. Journal of Rehabilitation, 82(4), 3-13. Retrieved from https://nmsua.idm.oclc.org/login?url=https://www.proquest.com/scholarly-journals/implicit-bias-toward-people-with-mental-illness/docview/1861771559/se-2?accountid=38048

Schimmack, U. (2019, November 30). Anti-black bias on the IAT predicts Pro-Black bias in behavior. Index. Retrieved April 27, 2022, from https://replicationindex.com/2019/11/24/iat-behavior/

Trafimow, D. (2017). Using the coefficient of confidence to make the philosophical switch from a posteriori to a priori inferential statistics. Educational and Psychological Measurement, 77(5), 831 – 854. https://doi.org/10.1177/0013164416667977

Trafimow, D., & MacDonald, J. A. (2017). Performing inferential statistics prior to data collection. Educational and Psychological Measurement, 77(2), 204-219. doi:10.1177/0013164416659745