I suppose that depends on the specifics of the experiment; the brief description above doesn’t really make it very clear, and the actual paper is paywalled.
When evaluating individuals, facts about the individual should screen off demographic facts. “Men are physically stronger than women” is a statement about central tendency. But an individual about whom you know they are female and a prize-winning triathlete is probably stronger than an individual about whom you know they’re a man who watches 14 hours of TV every day.
Or, let’s say that expert nuclear engineers are (for whatever reason) 75% men and 25% women. That means if someone tells you, “X is an expert nuclear engineer”, your prior for that person being a man is 3:1. However, if you are in a professional nuclear engineering context and meet an individual woman who is introduced to you as an expert nuclear engineer, you should not assign 3:1 odds that this description is wrong and that really she is an administrative assistant or schoolteacher (or an incompetent nuclear engineer; or, for that matter, a man).
Or even 1:1 odds.
Or, you know, 1:20 odds.
In other experiments on biased behavior in hiring — résumé evaluation and the like — the evaluator is presented with detailed facts about the individual, not just their demographic facts. They have a lot to go on besides the person’s gender or race or age or whatever. That’s how we can be pretty confident that what’s being detected is not accurate reasoning about central tendencies (as in your “men are physically stronger than women”) but inaccurate reasoning about individual data points.
This formulation of evidence completely disregards an important factor of bayesian probability which is that new evidence incrementally updates your prior based on the predictive weight of the new information. New evidence doesn’t completely eradicate the existence of the prior. Individual facts do not screen off demographic facts, they are supplementary facts that update our probability estimate in a different direction.
This is all true, but doesn’t seem relevant. The study description says:
participants were simply asked to rate a particular group
That sounds like rating the group, not individuals. It sounds like being asked about the validity of the stereotype itself. And I’m pretty sure the stereotypes mentioned as examples are in fact true:
a series of stereotypical characteristics, for women were: warm, family-oriented and (less) career-focused
The only question is the magnitude of the true stereotypical difference, and whether people estimate it correctly.
I don’t think it would be right even when applied to individuals. If someone tells you “X is an expert nuclear engineer” and you know that X is a woman, the prior for nuclear engineers being male no longer applies, because you can observe that X is a woman with 100% certainty. But in the resume evaluation example, what the resume evaluator wants to discover (how good a worker the applicant is) is not something that he can observe. It is true, of course, that the more detailed facts on the resume also should affect the evaluator’s result, but that just means that both the applicant’s race/sex and the other facts should affect the result. Even if the sex/race has a small positive correlation with being a good worker and the other facts have a larger positive correlation, the evaluator is better off using both race and the other facts rather than using the other facts alone.
I suppose that depends on the specifics of the experiment; the brief description above doesn’t really make it very clear, and the actual paper is paywalled.
When evaluating individuals, facts about the individual should screen off demographic facts. “Men are physically stronger than women” is a statement about central tendency. But an individual about whom you know they are female and a prize-winning triathlete is probably stronger than an individual about whom you know they’re a man who watches 14 hours of TV every day.
Or, let’s say that expert nuclear engineers are (for whatever reason) 75% men and 25% women. That means if someone tells you, “X is an expert nuclear engineer”, your prior for that person being a man is 3:1. However, if you are in a professional nuclear engineering context and meet an individual woman who is introduced to you as an expert nuclear engineer, you should not assign 3:1 odds that this description is wrong and that really she is an administrative assistant or schoolteacher (or an incompetent nuclear engineer; or, for that matter, a man).
Or even 1:1 odds.
Or, you know, 1:20 odds.
In other experiments on biased behavior in hiring — résumé evaluation and the like — the evaluator is presented with detailed facts about the individual, not just their demographic facts. They have a lot to go on besides the person’s gender or race or age or whatever. That’s how we can be pretty confident that what’s being detected is not accurate reasoning about central tendencies (as in your “men are physically stronger than women”) but inaccurate reasoning about individual data points.
This formulation of evidence completely disregards an important factor of bayesian probability which is that new evidence incrementally updates your prior based on the predictive weight of the new information. New evidence doesn’t completely eradicate the existence of the prior. Individual facts do not screen off demographic facts, they are supplementary facts that update our probability estimate in a different direction.
This is all true, but doesn’t seem relevant. The study description says:
That sounds like rating the group, not individuals. It sounds like being asked about the validity of the stereotype itself. And I’m pretty sure the stereotypes mentioned as examples are in fact true:
The only question is the magnitude of the true stereotypical difference, and whether people estimate it correctly.
I don’t think it would be right even when applied to individuals. If someone tells you “X is an expert nuclear engineer” and you know that X is a woman, the prior for nuclear engineers being male no longer applies, because you can observe that X is a woman with 100% certainty. But in the resume evaluation example, what the resume evaluator wants to discover (how good a worker the applicant is) is not something that he can observe. It is true, of course, that the more detailed facts on the resume also should affect the evaluator’s result, but that just means that both the applicant’s race/sex and the other facts should affect the result. Even if the sex/race has a small positive correlation with being a good worker and the other facts have a larger positive correlation, the evaluator is better off using both race and the other facts rather than using the other facts alone.