Why Lee Jussim is wrong about stereotype "accuracy"

It has long been fashionable for academics pushing right-wing views to claim that they are merely trying to pursue honest, objective science and bravely standing up to the academic establishment--one which is invariably assumed to be composed of liberals whose views are based on their political ideologies rather than on scientific evidence. Recently, Rutgers psychologist Lee Jussim has been playing this game with respect to the specific subject of the "accuracy" of stereotypes. He claims that stereotypes are generally very accurate, and that there is a widespread assumption in both academia and the general public to the contrary that is simply objectively false (e.g. Jussim, Crawford, & Rubinstein 2015). 

That's right, a professor at a prestigious American university actually wants to hold up people's beliefs about entire groups of human beings as being highly accurate representations of reality. So many obvious questions abound: is Dr. Jussim saying that black people are lazy violent criminals who always eat fried chicken and watermelons, and that Asian Americans are bad drivers who are good at math, and that women love pink things? After all, these are all stereotypes, and he is saying that many stereotypes are very "accurate". So it follows that many stereotypes that people are very familiar with, and often highly offended by, must be very "accurate", according to Jussim. This would, one imagines, include some or all of the stereotypes mentioned above. But to be fair, let's look closer at Jussim's argument before painting him as some sort of racist/sexist simply for trying to legitimize people's intergroup beliefs. After all, maybe the really offensive stereotypes like those previously mentioned are also the ones that aren't accurate (not least because they can't really be accurate: see below).

So what's wrong with this line of argument? First of all, it assumes that people's beliefs can accurately be represented in quantitative terms. Before explaining why this is a highly problematic assumption, I will explain how researchers in the past have assessed the (in)accuracy of stereotypes. As Jussim himself wrote in 2012, "There are many different ways to test for the accuracy of stereotypes, because there are many different types or aspects of accuracy.  However, one type is quite simple -- the correspondence of stereotype beliefs with criteria.  If I believe 60% of adult women are over 5' 4" tall, and 56% voted for the Democrat in the last Presidential election, and that 35% of all adult women have college degrees, how well do my beliefs correspond to the actual probabilities?  One can do this sort of thing for many different types of groups." 

In other words, asking people what % of members of group B (hereafter simply B) have characteristic X, and comparing it to the actual %, assumes that the answers to the % question will accurately represent how people normally perceive others as a result of holding stereotypes about the groups to which they belong. That is, such an approach to assessing the "accuracy" of stereotypes assumes that people's beliefs are statistical (e.g. 30% of B have X) rather than generic (all B have X). But this assumption may be wrong.

Another problematic assumption inherent in any claim that stereotypes are to some degree accurate is that stereotypes contain content that can be accurately operationalized in a way that conforms to reality. That is, the "stereotype accuracy" argument assumes that a stereotype being described as accurate is the result of things that members of the group to which it applies actually do. But of course, there are some stereotypes that, by their nature, certainly bear no relationship to objective reality. For instance, if it is believed that African Americans are apelike and mentally inferior to white people, then this is a stereotype that clearly cannot be assessed on its (in)accuracy, insofar as "inferiority" and the state of resembling an ape are both entirely subjective characteristics. Notably, this is clearly also true with respect to many other characteristics of stereotypes, many of which (e.g. laziness, athleticism, etc.) Thus such stereotypes cannot be the result of an accurate perception of reality, but must instead be artifacts of prejudice or other cognitive biases. 

But the biggest problem by far with claims of stereotype "accuracy" is that they ignore this point: Most people in a group won't fit the stereotype, however accurate it may be. By this I simply mean that if the stereotype is that members of group B are X, researchers like Jussim who want to play up its accuracy will frame it as "(Number) % of members of group B are X" or "members of group B are (number) % less likely than members of group A to be X". Then the number in these formulations can be compared to the actual quantity to assess how "accurate" the stereotype is. In reality, however, my point is that even for the most "accurate" stereotypes, the stereotype will actually be highly inaccurate in that most people in group B will not be X. As we shall see, this point is a lot less controversial than you might think. 

The examples given by Jussim of the % of women who voted Democrat or are more than 5' 4" tall seem of little or no relevance to actual stereotypes people actually hold about groups. What about more abstract and common stereotypes of women, like that they are more emotional, accommodating, and better equipped to do housework than are men? Is this actually true of most women (to the extent that such a question can even be answered objectively)? Fundamentally, the problem here is that even the most "accurate" stereotype will often lead to conclusions that are not just wrong, but highly offensive, when used to make judgments about individuals. 

Consider the conclusions Jussim et al. (2018) have recently reached on this subject, namely, that "race, gender, and age stereotypes tend to be moderately to highly accurate." What does this mean? Does it mean that judgments based on stereotypes of someone of a given race/gender/age will usually be correct? What % of the time will such judgments be correct? 

To their credit, Jussim et al. appear to anticipate this obvious criticism, writing (p. 7), 

This line of reasoning’s suggestion – that all stereotypes are inaccurate because most members of a group fail to fit a stereotype – is partially justified. It is true that most members of a group will fail to perfectly fit a stereotype. This, however, does not mean that the stereotype is inaccurate. To understand why requires understanding how this reasoning confounds two different levels of analysis and how considerably greater conceptual clarity can be brought to understanding stereotype accuracy by clearly distinguishing among these levels of analysis. [Emphasis mine.]

So there it is: the emperor has no clothes! They admit the obvious fact that most stereotypes aren't really accurate in the sense that people can't use them very accurately to make individual level inferences. What does this mean? Among other things, it means Jussim et al. acknowledge that most blacks aren't athletic or lazy, and that most Asians aren't bad drivers but good at math, and that most women aren't emotional and submissive, and that this holds for all other stereotypes as well. But the fundamental point is that stereotypes are oversimplified because they attribute the same characteristics to all members of a group, which leads to inaccurate conclusions most of the time, because such an attribution ignores that most people in the group don't have those characteristics. This is the basic shit about how stereotypes have long been defined that Jussim and his colleagues love to bitch about, yet in the above passage they admit that it's actually true--but, but, it, uh, doesn't count as real accuracy!

Here's how they dismiss this problem later in the same paper (p. 9):

Absolutist stereotypes – beliefs that all members of a group have some attribute – will indeed almost always be false, because there are almost always wide variations among individuals. A single exception invalidates an absolutist belief. Just as a belief that the temperature in all locations in Alaska is always below freezing will be disconfirmed by a single reading of 33 degrees Fahrenheit in Juneau on July 15th at 1pm, a belief that all Germans are efficient will be disconfirmed by discovery of a single inefficient German. The vast accumulated empirical evidence on stereotypes, however, has yet to report a single person who holds absolutist stereotypes. Instead, the evidence indicates that most stereotypes are quantitative and probabilistic, not absolute (Citations omitted).

Really? Most stereotypes are in the form of quantitative values? Most people's stereotypes are that a given percentage of members of group B have trait X? That seems quite hard to believe, since even most statisticians, much less normal non-experts, think in quantitative terms on a regular basis in their day-to-day lives. Aside from being contradicted by the Hammond et al. paper I previously cited in this post, the claim that most stereotypes are quantitative, not absolute doesn't appear to be well-supported by the sources cited. The Hammond et al. study aimed specifically to assess the cognitive structure of stereotypes, but the 3 sources cited by Jussim et al. (2018) did no such thing. For instance, McCauley & Stitt (1978), one of these 3 sources, just proposed "...a quantitative and individual measure of stereotyping, based on defining stereotypes as probabilistic predictions that distinguish the stereotyped group from others." So they defined stereotypes as probabilistic rather then empirically demonstrating it, so that clearly doesn't support Jussim et al.'s views at all. What about Judd et al. (1995)? This study assessed white and black Americans' stereotypes of each other's groups. The study aimed "...to examine theoretical issues in stereotyping and to describe the current state of ethnic interrelations among young people." Their findings were: "Throughout, the samples of African Americans demonstrate interethnic judgments that are consistent with existing work on stereotyping and ethnocentrism. White American students, however, reported judgements that replicate neither the out-group homogeneity effect nor ethnocentrism." So...white people don't perceive black people (the out-group) as all being similar (homogeneous). But black people do hold such perceptions of white people. Interesting, but doesn't support the claim about most stereotypes being non-absolute. Lastly, there is Swim (1994), which "assessed the accuracy of people's stereotypes about gender differences in 2 studies by comparing perceptions of sizes of gender differences with meta-analytic findings." But this obviously assumes that numerical answers given to researchers' questions about "what percent of women have college degrees" or other such questions are accurate reflections of stereotypic beliefs.

I will close by quoting some important points previously made on this subject by Ryan (2003), who made several observations very similar to those I make in this post: 

"The stereotypes that women are passive or that Blacks are athletic, for example, are no doubt erroneous if they are meant to imply that all women or all Blacks are so. And on what basis would one determine the actual passivity of women or the actual athleticism of Blacks anyway? Further, the notion that stereotypes can be accurate seems to imply that group attributes should be applied to individual group members so long as those attributes (or stereotypes) are accurate. This implication seems highly offensive in a society that values the individual and his or her unique merits."