Mind reading through stereotypes or projection?

The sort of mind reading referred to in Daniel R. Ames’ 2004 paper (Ames, 2004a) isn’t the mystical sort practiced by Jedi’s or hypnotists or con artists on eBay. It is actually the most boring, mundane, and intuitive kind of mind reading every one of us do in everyday of our lives–it is how we sense what the people around us think, feel, and want. A good friend cancels lunch without an apology, yet it doesn’t take a mind reader to understand that the good friend must’ve had a good reason. A co-worker cancels a meeting at the last minute and after a bit of mind reading, we come to suspect that the co-worker might be wagering some workplace power play. A romantic interest cancels dinner, and even after repeated mind reading, we still might never realize that she’s just not interested. After all, no one is a perfect mind reader, which is exactly why social psychologists like Ames and (cough) myself are concerned about how we make these mental inferences and judgments quickly and accurately.Coming from a philosophy background, I had always thought of the problem of mind reading in terms of Simulation vs. Theory. However, the two opposing social psychological models of mental inference that Ames reviewed, projection and stereotyping, seem to closely parallel the simulation/theory debate. One might mind read by projecting (=simulating?) their own mental states onto others, or perhaps by relying on his/her stereotypes (=theories?) about the mind reading targets. Ames’ literature review suggests that previous social psychology research programs have treated these two models as two mutually exclusive strategies. Ames goes on to suggest that projection and stereotyping might function as something like alternatives for one another. He explains it better than I do:

I suggest that when a target’s behavior is ambiguous, perceivers shift between stereotyping and projection as mind-reading strategies. Specifically, when perceivers see themselves as initially more similar to a target, they will rely more heavily on projection and less heavily on stereotyping than when they see themselves as less similar. In short, I claim that perceived similarity plays a moderating role in inferential strategies.

Ames conducted 3 separate studies to test his theory:

Study 1 articipants (=people serving as lab rats) first answered questions about themselves such as “Do you enjoy the television show South Park?”, “do you enjoy watching professional sports on television?”

Then participants were then shown a cue card about Alice, a fictional 28-year-old medical student. One version of the card read “Alice loves the television show South Park–she has even bought some DVDs. Alice doesn’t really enjoy playing or watching sports and only very rarely watches sports on TV.” The other, opposite version read “Alice says she can’t really stand the television show South Park. Alice loves playing and watching sports and frequently watches sports on television.” The matches (0 to 2) between the participants’ answers and the cue card version about Alice serves as one measure of similarity/dissimilarity. Participants also indicated their similarity to Alice by rating their agreement to the statement “I’m similar to Alice” on a 12-point scale. This serves as the 2nd measure of similarity/dissimilarity.

Participants then read a vignette involving Alice helping her professor with a bike problem.

Alice is walking across campus one day after classes and comes across a person crouching next to a bicycle. It’s obvious that the chain on the bicycle has come off and the person is struggling to repair it. The chain is greasy and dirty and the person is getting messy but doesn’t appear to be having any luck with it. Alice notices that the person is actually a professor in one of her most important classes. She pauses and stops and then goes over and begins helping. She finishes fixing the chain for the professor. The professor thanks her and Alice smiles and continues on her way.

After reading about the scene, participants rated Alice’s mental states by rating their agreement with 8 questions on a 12-point scale (1 = strongly disagree, to 12 = strongly agree). The items range from “Alice wanted to help any person in need,” “Alice hoped she would get a better grade for helping,” “Alice felt shy about interacting with the professor,” etc. This serves as the prediction of mental states for the target.

Next, participants were given the instruction: “What do you think you would have felt in this kind of situation?” Participants then answered the modified questions: “I would have wanted to help any person in need,” “I would have felt shy about interacting with the professor,” etc. This serves as the prediction of mental states for self.

This was followed by another round of rating, with the instruction “What do you think a typical medical school student would have felt in this situation?” The questions were again modified: “They would have wanted to help any person in need,” “They would have hoped they would get a better grade for helping,” etc. This serves as the prediction of mental states for the group to which Alice belongs.

To eliminate order effects, half of the participants completed first the self questions, followed by the group questions; the other half completed the surveys in the group-then-self order. Two other scenarios were also shown with similar procedure: one about Mark, a fraternity member, approaching a female student at a party; the other about Steve, a lawyer who privately reported his team’s work to the firm’s senior partner, on the same day when the team was to make a presentation on the same work.

Now comes the interesting part: how do you differentiate between projection and stereotyping from the survey results? Ames’ solution is simple: computing multiple regression for each participant and each scenario, and predicting (clarification: by “predicting” I’m referring to using one set of questionnaire results survey to predict the results of another, which is different than the person “assessing” Alice’s, his/her own, or the typical medical student’s mental state) target responses with self responses and group responses both taken together. The standardized beta for each were then used as measures of projection (the self response beta) and stereotyping (the group response beta).

So, the more a participant’s own responses predicted the target response (the higher the self response beta), the higher his or her measure of projection was. The more a participant’s group responses predicted the target response (the higher the group response beta), the higher his or her measure of stereotyping was.For the statistically disinclined, check out the figure on the left for more clues. For the statistically disgusted, there might be a separate post on multiple regression in this blog’s future.

Hopefully, by discussing Ames’ results for study 1, shown in figure 2, all this will become clearer. A person’s answers to his own mental state (projection) does a better job at predicting his answers to Alice’s mental state (higher self response beta) when the person is most similar to Alice (he matches Alice in both South Park preference and watching sports). Conversely, a person’s answers to the mental state of the typical medical student does a better job at predicting his answers to Alice’s mental state (higher group response beta) when the person is least similar to Alice (he matches Alice in neither South Park preference and watching sports). While I have no idea what Ames did with the one match data, these results are also consistent with the stated perceived similarity/dissimilarity to Alice (the 1 to 12 point scale “I am similar to Alice).

All this is just a convoluted way of confirming Ames’ theory that people use either projection or stereotyping to make judgments about what other people are thinking, depending on how similar other people are perceived to be. When I think that I am similar to Alice, I am more likely to guess what Alice is thinking/feeling by tapping the question “What would I do in the same situation?”; but when I think that I am dissimilar to Alice, I am more likely to guess what she’s thinking by asking “What would a med student do?”

Study 2: My summary for study 1 was perhaps a bit long-winded because all 3 studies follow similar logic, but with different caveats in experimental design. For study 1, the participants indicated their perceived similarity/dissimilarity to the target, and the study takes advantage of these professed matches and mismatches for comparison. For study 2, similarity and dissimilarity was actively manipulated, and the participants were randomly assigned to either a similarity condition or a dissimilarity condition.

Participants first were asked to imagine themselves being paired up with someone as part of a personality assessment workshop and then solving challenging puzzles together. Participants answered questions on a 1 to 5 scale about themselves in such team working situations (e.g. I would want my partner to have fun and be comfortable, 1=Not at all, 5=Very much), and then answered the same questions on what they thought a typical MBA students would answer (e.g. A typical MBA student would want his/her partner to have fun and be comfortable). Participants then filled out a 3 question survey about their personal preferences on Adam Sandler, watching sports, and creative arts, and then were given information about Brian, a hypothetical MBA student that either shared (similarity condition) or did not share (dissimilarity condition) their own preferences. Participants were then shown a 30s clip featuring Brian and another character David working on a logic puzzle. Finally, participants answered the same questions about Brian’s mental states (e.g. Brian wanted his partner to have fun and be comfortable)
Since the objective and procedure for study 2 is more ore less identical to study 1 except for the similarity manipulation, one would expect the results to be similar as well. And this was indeed the case (see figure 3 on the left), as participants in the similarity condition have higher projection beta and lower stereotype beta, and vice versa for participants in the dissimilarity condition. One additional effect that became more evident in the results of study 2 was the clear-cut negative correlation between projection and stereotyping.

Study 3: Study 3 adopts yet another caveat in experimental design. In both study 1 and 2, both projection and stereotyping revolved around the correlation in the reported contents of mental states (how much self/group responses correlated with target responses). Study 3 uses response time measures and a task facilitation approach (Klein, Loftus, Trafton, & Fuhrman, 1992) to measure more directly the effects of stereotyping and projection.

Ames explains the task facilitation approach:

To test for stereotyping and projection effects, Study 3 drew on Klein et al.’s (1992) task facilitation paradigm, which suggests that if information needed for a second inferential task is used in a prior task, the second task will be executed more quickly than if the prior task had not been recently completed. Thus, reaction times for a second task in light of a preceding one can shed light on whether the judgments rely on the same information.

This is perhaps best explained in terms of the methods and assumptions in Study 3, which follows a 2×2 design. Participants are randomly assigned to two conditions: 1) similarity vs dissimilarity (as in Study 2) and 2) self-target-group and group-target-self survey oder (the new caveat). This yields 4 experimental groups:

1. 1. Similar; self-target-group–a participant is randomly assigned to the similarity condition and answers questions about themselves first, followed by the target (Michael the investment banker), followed by the group (banker in general).
   2. Similar; group-target-self
   3. Dissimilar; self-target-group
   4. Dissimilar; group-target-self

For participants assigned to group 1 and 3, after first answering questions about themselves, they would both presumably have that projection data on “top of mind,” since data about their self was recently “accessed” or mental buffer, or mental RAM, one can go on with the computer analogies) If mind reading someone whom is judged to be similar makes use of projection, one would assume that participants in group 1 would answer the questions (primarily yes/no) much faster than group 3, since projection data is more readily available than stereotyping information. Group 1 would also be expected to have faster reaction time than group 2, since group 2 does not have projection data on top of mind, but rather stereotype data, which isn’t the primary tool when mind reading someone similar.

Likewise, participants assigned to group 4 would be expected to have faster reaction times than groups 2 and 3. This was indeed the result (see figure 5).

Whew. Just some final thoughts:

Admittedly, Ames’ findings seem pretty straightforward: the more I judge another person to be like me, the more likely I would put myself in his shoes to gauge his thoughts; the less I judge another person to be like me, the more likely I would use information about the group to which he belongs to read his mind. But the projection vs stereotype construct is probably the best social psychological equivalent to the simulation vs theory stuff that I’m interested in. Also, might Ames’ similarity contingency model (or SIMCON, as Ames likes to call it, as in “We’re going to SIMCON-4!”) have some legs when applied to cultural differences?

—

Ames, D. R., (2004) Inside the Mind Reader’s Tool Kit: Projection and Stereotyping in Mental State Inference. Journal of Personality and Social Psychology, 87, 340-353

Klein, S. B., Loftus, J., Trafton, J. G., & Fuhrman, R. W. (1992). Use of exemplars and abstractions in trait judgments: A model of trait knowledge about the self and other. Journal of Personality and Social Psychology, 63, 739-753