Friday, May 7, 2021

relational frame theory on prejudice; implications for anti-racism

Rene Magritte, The Psychologist (1948)

Responding to Wittgenstein's Philosophical Investigations (1953) from the perspective of Skinnerian behaviorism was an exercise in assessing my comprehension of Skinner's Verbal Behavior (1957), which I read during the early stages of last year's lockdown. Probably the most substantial token of the exercise's success was my discovery that there were situations I couldn't give satisfactory accounts for, even after leafing through my copy of Verbal Behavior for an hour or longer and cross-referencing my notes with Science and Human Behavior (1953) and About Behaviorism (1976). I came to suspect the problem wasn't my understanding of the literature, but conceptual knots in the literature itself. The problem of reference, for instance: Skinner insists that verbal events do not refer to anything; the spoken word "chair" does not communicate a mental image of a chair, but is a conditioned response strengthened by a stimulus class. But in speaking of rule-governed behavior, he states that rules are verbal stimuli which specify contingencies—which seems to violate his own strictures against reference, and makes a functional description of rule-governed behavior a confused, onerous affair.

So I picked up Hayes et al.'s Relational Frame Theory: A Post-Skinnerian Account of Human Language and Cognition (2001) to see what progress the field made five decades after the publication of Verbal Behavior. Relational Frame Theory is not an easy read, but it's very exciting. At least from the perspective of a dilettante, RFT shows promise: it closely adheres to the main tenets of radical behaviorism, it starts from simple first principles which allow for the emergence of complexity, and it appears to be testable. What separates it from Verbal Behavior is its centering of derived stimulus relations, which began to be studied in the 1960s.

The core concepts of RFT are lot more difficult to synopsize than the "stimulus-response-reinforcement" of twentieth-century behaviorism, and take much of Skinner's work as a given. Rather than offer a summary here, I can point you to an excellent rundown of RFT written for the lay reader, and also a more technical overview if you'd like to chew on something with more gristle. 

For now, we'll just provide five basic points to bear in mind:

1.) RFT posits that human beings are unique among the animal kingdom in that we're capable of learning to derive relations between stimuli (without direct training) as an overarching operant. "While 16-month-old babies readily show robust forms of mutual entailment," the authors write, "even 'language-trained' chimpanzees show no such thing."¹ Deriving stimulus relations, they often point out, is a learned behavior that affects the learning process itself.

2.) RFT uses a specialized redefinition of verbal behavior as "framing events relationally." In these terms, language (which consists of "arbitrarily applicable relational responding") is an outcome of verbal behavior, not the other way around.²

3.) A relational frame isn't a thing: it is the behavior of framing events relationally (responding to events in terms of other events) under the control of particular contexts. A relational network doesn't exist outside of the organism, nor really inside of it either: it is a pattern of responding.

4.) Our environment acquires verbal functions as its constituents participate in verbal networks—or, rather, as we respond to objects and events relationally, and these "networked" responses become ingrained in our behavioral repertoires.

5.) A major component of RFT is the transformation of stimulus functions across relational frames. For instance, a hiker might read a sign at a trailhead: CAUTION: ALL THE SNAKES IN THESE WOODS ARE POISONOUS. The likely effect will be that her responses to snakes on the trail will be augmented prior to her seeing one. This is unremarkable at first glance, but no other animal on Earth (so far as we know) is capable of modifying its behavior to a stimulus class prior to direct contact with its members as a consequence of another stimulus with no nonarbitrary physical relation to it.

Now: a while back, I wrote about the problems of curbing racist sentiment through shaming tactics. Anyone who's taken behaviorism 101 understands that punishment merely displaces objectionable behavior (and only temporarily), and can result in insidious, undesired side effects. Today we'll be looking at a passage from Relational Frame Theory which explicitly examines prejudicial behavior, with worrying ramifications for the "anti-racism" discourse that has become prominent in educational and business settings over the last year. Emphases are my own.

Prejudice involves responding to people or events based on conceptualized groups, but in this case the conceptualized self shares few if any of the salient characteristics with the conceptualized others (i.e., the prejudiced person is not a member of the conceptualized group). From an RFT perspective, prejudice involves a derived transformation of the functions of individuals based on direct or verbal contact with the functions of a few members of a conceptualized group. For instance, if my car is vandalized in a particular neighborhood then the functions of other "members of that community" may be transformed in such a way that I also respond to them as vandals.³ This transformation of the community members is derived and this, in essence, is what makes behavior toward them prejudicial.

Social psychologists have studied the phenomena of prejudice extensively. For example, using what has become known as the minimal groups paradigm, Tajfel, Billig, Bundy, and Flament (1971) found that British schoolboys at summer camp expressed a preference for other members of a temporary in-group to which they had been randomly assigned, allegedly based on their common fondness for the work of particular painters. Furthermore, when asked to distribute pocket money to other children identified only by code numbers and group membership, a bias toward in-group members was observed. In effect, simply being categorized as an in-group or out-group member is enough to produce ethnocentric behavior and inter-group competition.

From an RFT perspective, the social functions of subjects in the Tajfel et al. (1971) study were transformed by their participation in conceptualized groups. This transformation was based on derived relations between particular individuals and the group category (i.e., from the boys' perspective; "if my group is the preferred group then all members of my group are preferred individuals and all members of the other group are non-preferred"). Prejudice, therefore, is based largely on derived relations among individuals and their groups rather than upon direct experience. As such, RFT suggests that prejudice is a natural effect of human language, and that instructional approaches to reducing prejudice are unlikely to be successful. The effect is natural because the transformation of stimulus functions through relational frames is a common verbal event.⁵ It seems to be impossible, therefore, to prevent the construction of groups based on myriad features. Like the thought exercise in Chapter 3 ("How is a TV like a prostitute?"), similarity can be constructed between any two events, and seemingly be explained by the physical features that participate in arbitrary relations.⁶ Once a relation is formed, it is difficult to prevent the transformation of stimulus functions in terms of that relation.

Instruction is unlikely to reduce prejudice for three reasons. First, these instructions usually refer to group membership and thus increase group membership. For instance, the instruction "don't think bad things about black people, they have great abilities" increases the salience of "black people" as a conceptualized group and adds features to that group. This process is easy enough to detect when the conceptualized group involved is unusual. A person hearing the sentence "don't think bad things about short people with harelips, they can play the piano wonderfully" may form a conceptualized group ("short people with harelips") that had never been formed before, and verbally attribute features to the entire group. Even if the sentence is known to be a joke or an example, the verbal relation may still occur. The readers of this volume, for example, may indeed be somewhat more likely now to think "piano" if they ever come across a dwarf with a harelip.

The second problem is that anti-prejudice instructions may actually strengthen stereotypes in the name of denouncing them. Common stereotypes are already available in the verbal repertoire of most members of a culture, by definition. A simple priming study would probably show that stereotypical verbal relations are in the verbal behavior of saint and bigot alike. Just as "white" will prime "black," "Jew" will probably prime "money" whether the person has been instructed that Jews are more interested in money than others, or that they are not.

Finally, anti-prejudice instructions may increase the fearsomeness of groups, since they often add negative functions to the presence of prejudiced thoughts.⁷ As will be reviewed in Chapter 12, suppressing thoughts tends to increase their frequency. [See below.] Consider the instruction "It is horrible to think that disabled people are being punished by God." This instruction is likely to strengthen the salience of conceptualized group membership and the relatedness of the specified stereotype, as we have already discussed, but it also make it essential to avoid the very verbal relation that the sentence helps establish, since noticing it "is horrible." This is very likely to increase the frequency of the prejudiced thought, and to lead to avoidance of occasions that might give rise to it (such as spending time with disabled people). That precise implication has been examined by Harvard psychologist Ellen Langer.

In one of her studies (Langer, Bashner, and Chanowitz, 1985) children were shown a disabled person in a wheelchair through a one-way mirror. There were two conditions. In one, children were encouraged to say aloud all of the prejudicial things they had heard about disabled people. No punishment, correction, or support was provided——the statements were merely acknowledged. In the second condition, prejudicial statements were corrected and non-prejudicial statements were instructed. The children were then led into the room with the disabled person. The group that had received anti-prejudicial training instructions avoided going near the disabled person significantly more than the group that was allowed to notice prejudicial thoughts without confirmation or disputation.

The key point is that RFT suggests that verbal relations are only amplified by direct verbal attack, since these relations are additive, not subtractive. A more effective method of reducing prejudice may be to create contexts that reduce the psychological importance of categorization per se, or the likelihood of a transformation of stimulus functions in terms of such categorization. Langer's "mindfulness" intervention is an example. Repetition of verbal statements until they become meaningless (e.g., Titchener, 1916, p. 425) may be another——a method that has been successfully used by sex therapists to reduce the fearsome quality of sexual terms. The creation of myriad overlapping categorical concepts ("white gay dwarfs with harelips——Native American women Baptist vegan millionaires") may be another. More basic work needs to be done on how to loosen relational networks, and avoid the transformation of stimulus functions through them. That simple process is the essence of prejudice and we have good reason to suppose that a front, logical attack on prejudicial behavior may strengthen some of the basic processes it is designed to undermine.⁸

On the effects of thought suppression:

When human beings are asked what they do when they face negative thoughts and feelings, suppressive strategies (e.g., "think about something else," "forget it," and "avoid thinking about it") are among the most common responses (Rippere, 1977a). Clinical populations show the same pattern (Brewin, Watson, McCarthy, Hyman, and Dayson, 1998). The belief that these are effective strategies seems to come from their immediate effect, but when a slightly longer time frame is examined, deliberately suppressing thoughts is a shockingly counterproductive coping strategy (Beevers, Wenzlaff, Hayes, and Scott 1999; Wenzlaff, 1993). Thought suppression leads to a temporary decrease in the frequency of the avoided event, followed by a substantial increase in the event (Wegner, 1994). As would be predicted by RFT, when thought suppression occurs in the presence of a negative mood, a bidirectional relation is established between mood, thought, and suppression. For example, Wenzlaff, Wegner, and Klein (1994) found that reinstating a mood that subjects were in while suppressing a target thought resulted in a return of the suppressed thought. The implications of this are immediately disturbing. The relational qualities of human language glue together mood and the literal content of verbal events. Thinking about a death, for example, will often induce sadness. This negative quality leads to overt attempts to reduce the thought. In the short term, this works, but in the long term suppression leads to more thoughts of death and more sadness. A vicious cycle is established. Depression, panic, and other extremely destructive forms of psychopathology could also be the result of similar behavioral sequences.

1. Mutual entailment is a contextually controlled quality of a relational frame. From the authors: "mutual entailment applies when in a given context A is related in a characteristic way to B, and as a result B is now related in another characteristic way to A in that context." The relation between the vocal event bird and a feathered biped with wings is mutual entailment at its simplest. 

2. "Arbitrarily applicable relational responding" isn't as scary a concept as it may appear. Nonhuman organisms are quite capable of responding to nonarbitrary relations between stimuli—choosing the larger fruit, chasing the slower antelope, and so on. The relations of language are arbitrarily applicable in that the word "bird" has no physical correspondence to a feathered biped with wings, and yet we learn to place the two stimuli in a "frame of coordination." When we see a feathered biped with wings, we're primed to say "bird." When we hear someone say bird or read the letters B-I-R-D, we're likely to respond by looking around for a feathered biped with wings, or by weakly "seeing" one. The relation is arbitrary—but for all practical purposes, its application is not.

3. The thing to note here is that a nonverbal animal wouldn't be able to derive any relationship between "members of that community" and "car with its windows smashed in" after just one incident, and without actually witnessing the vandals damaging the car.

4. Verbal behavior was apparently a prerequisite for the formation of human social groups for which Dunbar's number is not a limiting value. 

5. As soon as people are grouped together under a named category—"black people," "white gay cis men," "K-pop stans," etc.—stereotyping invariably follows. The implicit treatment of members of a defined group as a monolith (which is manifested as much as in the remark that a black person is "well-spoken" as in the concept of being "politically" black, gay, etc.) is another natural byproduct of relational frames. An abstract, general term for a member of a group numbering in the thousands or millions, by definition, abstracts and generalizes.* One thing that anti-racism discourse usually gets right is the extent to which the baggage of a racial label contributes to our behavior with regard to a person we identify as "wearing" that label. What it gets wrong, as this passage makes clear, is reifying the relation between individual and category through its mode of attack.

* Consider the following. A workshop leader gives you a sheet of paper with three headings: "White People," "Black People," "Asian People." For each column, you are given thirty seconds to write down, without any self-censoring, all the attributes that spring to mind for each of the groups. If you're writing quickly and being honest, each column would be filled with stereotypes—many of which you may not have observed firsthand if you're from a racially homogenous region.

Now imagine a child from an isolated rural commune in North America is asked to perform a similar exercise. She and most of the people in the commune are white, but she knows two other children—let's call them Dan and Janet—who are black and East Asian, respectively. This young person has never left the commune, never read any books but the sect's doctrinal texts, doesn't have internet access, and has never watched television, listened to the radio, or seen a film. She understands that Dan and Janet are black and Asian, but has no conception of what those labels mean outside of the context of Dan, Janet, and their parents. Would the items she wrote in the "black people" and "Asian people" columns be qualities she ascribes to Dan, Janet, and their families from firsthand experience?

6. Context, for the curious: twenty seven items are arranged in three columns, each numbered one through nine. This first column is headed "how is a..." and a list of nine nouns follows. The second column consists of nine verbal relations ("like," "unlike," "better than," etc.). In the third column are nine more nouns. The reader is asked to pick three random numbers between one and nine, construct a sentence ("how is a TV like a prostitute" is what you'd get if your numbers were 9, 1, and 1), and then respond quickly to it. "The primary point," the authors tell us, "is that the contextual control over relational responding is quite arbitrary."

7. I'm far from the first person to suggest this, but one of my fears regarding the Calvinistic strain of anti-racism discourse (a'la Robin DiAngelo) is that by drilling into white people that they are actively participating in the oppression of people of color—all their lives, no matter what they do, regardless of their intentions—their interactions with people of color become occasions for judgmental self-scrutiny, conscientious performance, and anxiety. Because of this, it might become the case that a trained and dedicated anti-racist doesn't actually enjoy being around people of color at all, and doubles down on the rhetoric and and conspicuous signaling of their anti-racism to displace their perception of this fact. (Obviously people of color in majority-white workplaces and schools are already accustomed to self-discrimination, to feeling as though their guard must always be up and they must always be "on"—but is the solution to erect more barriers against candor and openness between colleagues of different races? Racism is poisonous to civil society, yes, but prescribed racialism can't be the antidote.)

8. I've suggested before that one (not at all workable) way of attenuating prejudice across racial lines would be to conscript citizens to two years' service working in a restaurant kitchen, a coffee shop, or a Trader Joe's—environments where getting through the day requires that they trust, rely on, be relied on by, and find things to like about coworkers of different races, religious groups, political convictions, and walks of life, with whom they serve on equal social footing. (An office setting typically doesn't nurture the same degree of camaraderie; collaborators on a project still spend most of their time working privately, and seldom with as much urgency as when there's a line from the counter to the door and there's just no time for anything but intense cooperation.) As the authors note, relational networks are more easily added to than loosened, and all of us would benefit from having a given group category (racial, ethnic, religious, straight/queer, cis/trans, boomer/millennial/zoomer, etc.) populated by people who wore the same uniform we wore, who have covered for us, who have thanked us for having their backs, who shared our disdain for an overbearing boss or a scrofulous customer, who have seen us screw up, and with whom we've shared a laugh about it after the fact.

Possibly a more realistic solution would be class struggle—which could account for why the "racialism as cure for racism" approach has been embraced by powerful corporations. Worker solidarity is made more difficult to achieve when an increasingly diverse workforce views itself as irreparably divided among racial and ethnic lines.

9. If a person we identify as belonging to a defined social category offends us, we're unable to prevent relational frames from entering into our overall response to them. We may think slurs we don't say, or ascribe the offensive behavior of the individual to the essential character of the group to which we recognize them as belonging. It's nothing to brag about—but we do it. We literally cannot help it. If, later on, we frequently recur to the event out of well-intentioned shame, or in an exercise to improve ourselves, what we may end up doing is strengthening the relation between the person, the event, and the unspoken slur or stereotype. 


  1. I was reading something recently about how Freud essentially reinstated the idea of sin in a secular, "scientific" context by saying that human behavior is polluted with wrong desire on an unconscious level (I'm not strong on Freud so sorry if this is a bit off.) It seems like there's a parallel to be made between the idea of repressed sexuality and repressed racism, and so also between racism and sin. Replace references to racism in the descriptions of avoidance and self-fleggelation in the excerpt you posted with adultery or drink or what have you and you have a very familiar picture. Of course partly our society, or parts of it, tries to dissolve the problem of secular sexual sin by punting on it via sex positivity and saying "everything is permissible that doesn't hurt someone else." That seems similar to a possible solution the excerpt is gesturing to, you can confront your own racism non-jugementally in an environment where nobody is hurt by it. In both cases, though, it's hard to figure out where to draw the line between something harmful and something harmless. Of course if we go back to sin and Christianity, a Christian is certainly supposed to avoid sinning, but a Christian also can't purify themselves of sin, only God can purify them, through forgiveness. Likewise it seems that this excerpt says that becoming, by your own individual power, completely non-racist is impossible. So the failure of this approach is giving people a task they simply don't have the power to carry out. But then who or what is the secular anti-racist god that can forgive and purify us? It's interesting to me that the other possible solution gestured at seems to be basically a dissolution of categories/separation, which feels like a very Christian concept as well, like a vision of heaven where all that matters is that we're united in god. Of course, I'm extrapolating a bit... but the confusion of category bounderies seems to point that way to me; wouldn't a transcendence beyond language be like the ideal, but impossible, solution to the problem as they've laid it out?

    Anyway, this was a good read, I definitely agree that this punitive approach to anti-racism is of very limited use, especially in changing people's minds or controlling their unwanted thoughts. It does seem like sometimes just getting people to avoid other people whose day they might ruin if they were more confident is a pragmatic victory, a lot of racists don't shy away from their racism or from black people, instead telling or showing them exactly what they think of them, and it seems good to avoid that even if the racists are not perfectly purified this way.

    1. Reminds me of a monograph I read a couple years ago about how the secular concept of social and technological progress was a gradual evolution of the Christian belief in the Millennium.

      I forget who it was—if I recall, it was some NYT columnist who's probably on Substack these days—tweeted something about how a secular doctrine of sin and perdition that omits any possibility of redemption simply isn't tenable. I completely agree, and I wonder where this trend is headed.

      My from-the-hip answer to how prejudice might be mitigated if it's an inevitable product of verbal behavior is through an all-around economic leveling. Maybe a member of [group X] thinks that [members of group Y] smell funny (because of their diet), and no, that's not great—but as long as members of each group can typically look at each other and see a person on an equal social footing, who lives in the same neighborhood, whose kids play together after school, who is just as likely to be their boss as their professional equal or subordinate, etc.—these little prejudices are less likely to boil over into vicious inter-group resentments. (One would hope, anyway.)

      (The last item is a good reason to push for diversifying the professional class, but frankly I'd also like to see, for instance, the pay and perks of custodial work becoming tempting enough that a white male with a two-year degree in a reasonably affluent city might decide he'd rather mop floors than enter data. This is utopian thinking, I'm aware. But one cause of racial prejudice is the correlation, in many parts of the country, between low-wage jobs and people of color. When I worked at the museum, the custodial staff had only one white member; all the rest were black. There was an incident where a black member of the IT team visited the office of the Chief Operating Officer to have a look at her computer, and she assumed he was a janitor and nudged her waste bin towards him. Yes, really. I'm willing to argue that a lot of modern racism and white-collar microaggression is denigration of the "unskilled" working class (and unemployed poor) distorted through a racial frame, but I've already gone off the rails as it is.)

      Re: the second paragraph—the thing is, the kind of anti-racism that's in vogue these days isn't directed at the kind of person who openly despises POC. DiAngelo and Kendi are never going to filter down to the white working poor and aging boomers in the rural South with Confederate flags hanging over their stoops. They're not trying to reach NEETs on 4chan or the Stormfront mutants. This stuff is aimed at, and is predicated upon, educated white liberals who are already fairly well-meaning, probably don't believe they're racist, and certainly don't want to be racist. The undesirable side-effects suggested by the RFT excerpt are going to manifest among that crowd—if they actually manifest at all. I'm curious (and afraid) to see where this is going. Most white people tend not place their race at the fore of their conception of themselves, and encouraging them to do so, even for the nominal purpose of combating racism, is to sow the seeds of white identitarianism.