Mapping "prejudice" research reveals its preoccupation with implicit bias

One of the many difficulties of doing social science is that the concepts that we study are often fuzzy.  Precisely defining concepts like "attitudes", "cognition", and the "self" can be challenging, which sometimes leads to dramatic differences in how scientists use the terms.

The challenges are only enhanced when the object of study is a politically charged concept like my chosen field of study, prejudice.  I believe this fuzziness in the definition of "prejudice" has exerted a distorting influence on research on the topic, affecting the questions researchers ask, the measures researchers use, and the interventions researchers develop.

Today, I'm going to focus on a small piece of this issue by answering the following questions:

1. When contemporary researchers choose to study "prejudice", how do they use the term?
2. What does contemporary researchers' use of the term "prejudice" reveal about their (often unstated) definitions of of the term?

Considering the sheer number of papers that are published on the topic every year, answering even these two questions is quite challenging.  Fortunately, the advent of linked indexing services like PsychINFO and Web of Science means that it is quite easy to retrieve a large number of articles about a specific topic, provided one's search terms are well-defined.

In addition, as long as a given indexing service provides information about the which publications cite which, we can use some items in the fast-expanding toolbox of network science to cluster publications by these citation patterns.  As long as we can reasonably assume that articles that are linked through citations use the term "prejudice" similarly, we can use the highly cited papers within clusters of articles to infer what researchers within the cluster mean by "prejudice".

Here, then, is an outline of what I did:

1. Search Web of Science using well-defined search criteria.  More specifically, I searched for journal articles published between 1989 and 2013 that that were categorized as "Social Science" and that contained the term "prejudic*".  This query resulted in a corpus of 4,967 articles, along with citation information about these articles
2. Create a citation network with the retrieved articles.  For each of the 4,967 articles in my corpus, I recorded whether that article cited each of the other articles in the corpus.  This resulted in a 4,967 by 4,967 adjacency matrix that represents the network of citation patterns in the article corpus.
3. Visualize the network.  In this step, I first pruned the network of articles that had been cited less than 20 times to remove articles that have been cited only a few times.  I then chose an algorithm well-suited to the visualization of a large network (more specifically, the igraph implementation of LGL in R).

Here is the resulting map of research on "prejudice":

Click the upper right-hand corner to pop out the image

In this map, the size of the dot is proportionate to the number of times a given article was cited in the corpus.  I have labeled the top 20 articles receiving citations by the name of the first author and publication date.

What can we learn from this citation map?

Before answering this question, it may be helpful to know the topics of some of the most-cited papers.  Below are the authors and publication dates of the top 10 highly cited papers, all of which are in the primary cluster of the visualization above.  For people who aren't as familiar with this field, I have also provided a verbal description of these papers.

1. Devine (1989).  The author of this paper, Trish Devine, is my advisor, so this is a paper with which I am deeply familiar.  The paper proposes that relatively automatic (unintentional) processes are different from relatively controlled (intentional) processes and applies that distinction to prejudice.  In future research, relatively automatic bias is often called "implicit bias", whereas relatively controlled bias is often called "explicit".
2. Greewald, McGhee, & Schwartz (1998).  Introduces the Implicit Association Test (IAT) as a method of measuring implicit bias (including racial bias).
3. Fazio, Jackson, Dunton, & Williams (1995).  Introduces semantic priming as a method of measuring relatively implicit racial bias.
4. Wittenbrink, Judd, & Park (1997).  Introduces the lexical decision task as a method of measuring implicit racial bias.
5. Plant & Devine (1998).  Introduces scales measuring the internal and external motivations to respond without prejudice.  There are many uses for these scales, but one of them is to identify who cares about their levels of implicit racial bias.
6. Dovidio, Kawakami, Johnson, Johnson, & Howard (1997).  Finds that implicit bias predicts behaviors that are difficult to control, such as non-verbal behaviors in interracial interactions, and contrasts this relationship with the relationship between explicit bias and things like jury decisions.
7. McConnell & Leibold (2001).  Finds that implicit bias predicts behaviors that are difficult to control in interracial interactions, such as errors during speech.
8. Greenwald, Nosek, & Banaji (2003).  Proposes a new way of scoring the IAT to better measure implicit bias.
9. Dovidio, Kawakami, & Gaertner (2002).  Finds that a person's implicit bias predicts how friendly others perceive the person to be, whereas, whereas explicit bias predicts how friendly people perceive themselves to be.
10. Dasgupta & Greenwald (2001).  Finds that exposure to positive examples of famous Black people (and negative examples of infamous White people) reduces implicit anti-Black (pro-White) bias.

We can notice a theme among these papers.  They are primarily about implicit (unintentional) race bias, how to measure it, who is likely to want to reduce it, and a potential method to reduce it.  In fact, of the 20 top-ranked articles, only rank 13 (Wright et al., 1997) and rank 20 (Pettigrew, 1997) are not in some way about implicit bias.  Significantly, both these articles are positioned pretty far away from the primary "implicit bias" cluster in the citation map.

From this analysis, one can draw the following conclusions;

1. The most influential articles that contain the term "prejudice" are primarily about implicit bias
2. When contemporary prejudice researchers speak of "prejudice", they primarily mean implicit bias

Why should we care if most contemporary prejudice researchers focus on implicit (unintentional) bias?

1. Many acts of discrimination are intentional.  If we take the most extreme examples of discrimination in the last several decades, such as the massacre of Jews during the Holocaust, the discrimination does not seem to be the result of unintentional bias.  Even if we focus on more modern events, it is implausible to argue that actions such as the Charleston church shooting were driven by unintentional bias.  If "prejudice" is defined to be unintentional, we will be ill-equipped to understand the psychological causes of these incidents.
2. If you have a hammer, everything looks like a nail.  If researchers define "prejudice" as "unintentional bias", then researchers who attempt to develop interventions to change "prejudice" will tune their interventions to affect unintentional bias.  This results in a paucity of research on how to change bias that is, in fact, intentional.
3. Most non-researchers understand "prejudice" to be intentional.  The common understanding of someone who is "prejudiced" is a person who has malicious, ill intentions toward people in a particular group.  In other words, most people believe that people who are "prejudiced" are assholes.  The gap between the common and researcher understandings of "prejudice" can create problems when researchers attempt to communicate their findings to the public.

Overall, the map that I created of research on "prejudice" reveals what I believe to be an excessive focus on implicit (unintentional) bias.