Wednesday, January 15, 2014

Hasty Genderalizations

Gender schemas are largely non-conscious hypotheses we all have about the different characteristics of males and females. We see females as nurturing, as communal, and as doing things out of concern for other people. And we see males as capable of independent action, doing things for a reason, and getting down to the business at hand. [The male gender schema includes negations of the female gender schema and vice verse.] We have schemas about everything, every social group defined by race, age, sex, social class, and roles. So students have schemas about what it is to be a professor. And people have schemas about what it is to be a scientist. And for most professions, the schema that people have for being a professional person overlaps much more with the schema for being male than it does with the schema for being female. So we take requirements to be successful for most fields as being capable of independent action, doing things for a reason, and getting down to the business at hand.
-  Virginia Valian in an address to Chairs and Senior Administrators at the City University of New York
Our beliefs about the relative rationality of men and women are importantly problematic regardless of whether our beliefs about men and women in general are by and large correct.

Suppose that a random male raised by gender-blind robots who pass the Turing test is, on average, significantly less likely to end up more nurturing, communal, and likely to do things out of concern for other people than is a female raised under similar circumstances. And suppose both sexes vary greatly along those dimensions, such that men who are innately at least as nurturing etc. as the average woman are fairly common. When you meet a new person, your use some model of them to predict their behavior, and that model has only your prior beliefs about people with the characteristics you immediately observe, such as their appearing male or female.

If your priors are in favor of men in general being non-nurturing (and they're accurate on average), your implicit model of any specific randomly chosen man will also predict that he is non-nurturing. It will take extra evidence for you to update to expecting the man to be nurturing. So at this point, you're already going to end up with a gender imbalance in professions that require the characteristics of female gender schemata, such as teaching kindergarten, social work, and nursing.

If the vast majority of professions require the characteristics of the female gender role, then even given only the things I've mentioned so far, you're going to end up with at least a mild case of women ruling the world and men being second-class citizens.

Now suppose people are actually not so great at Bayesian updating--their beliefs have huge amounts of inertia due to confirmation bias and related phenomena. If your (implicit, unconscious) priors have grown to be strongly in favor of men being non-nurturing, non-communal, and doing things out of self-interest rather than a concern for other people, then any given man will have to exhibit the characteristics of the female gender schema much more overtly than a random woman before you believe that he is in fact nurturing etc. Due to cognitive biases we already know about, a slight gender imbalance in innate tendency to exhibit the nurturing etc. characteristics required by the vast majority of professions could easily lead to an overwhelming, horribly oppressive case of women ruling the world and men being second-class citizens. If you add to that a long history of people in power liking power and wanting to keep it and have more of it, this scenario is even bleaker.

In reality, this is exactly what the world looks like, except that the vast majority of professions require the characteristics of male gender schemata instead--most professionals benefit from being seen as agenty, having reasons for their actions, and working efficiently. There are some exceptions: Grade school teachers, social workers, and nurses benefit from being seen as nurturing, communal, and doing things out of concern for other people.

But so far the model I've described only obviously explains things we've already observed. Does it make risky testable predictions as well?

You bet!

For one thing, it predicts the following of people working in a profession that emphasizes characteristics of the male gender schema. Suppose you hand people equal evidence of the professional competence of two candidates. Then you tell them that one is a bio of a male, and the other the bio of a female. The model I've described predicts that the man will be rated as more highly competent. Why? Because the raters will need to encounter more evidence of professional competence for the female to overcome the rater's priors against her. If this doesn't happen in real life, it's strong evidence against my model.

Furthermore, it doesn't predict that men and women would differ in their ratings of the candidates. A difference would be evidence against my model. Competing hypotheses--anything along the lines of "gender inequality happens because men dislike women more than women dislike men"--do predict that the ratings should differ according to the sex and gender of the raters.

In 1995, the ratio of admitted/rejected male applicants for postdoctoral fellowships at a certain medical school was twice that of female applicants. Wennerås and Wold investigated. They came up with a system for determining the "impact points" of professional academics. The points were awarded according to number of journal publications, prestige of the respective journals, number of articles in which zer name is listed first among the authors, and number of citations zer article received in a one-year period. They then used this system to determine the impact on their field of applicants for postdoctoral fellowships to a certain medical school in Sweden.

Ordinarily, the results of admissions reviews are not made public. Due to an unusual court case, the committee reviews for this particular round of medical students were, and the reviews included an overall "competence rating". From their article in Nature:
Did men and women with equal scientific productivity receive the same competence rating by the MRC reviewers? No! ... The peer reviewers gave female applicants lower scores than male applicants who displayed the same level of scientific productivity. In fact, the most productive group of female applicants, containing those with 100 total impact points or more, was the only group of women judged to be as competent as men, although only as competent as the least productive group of male applicants (the one whose members had fewer than 20 total impact points).
Wennerås and Wold controlled for the applicant's nationality, education, field, university affiliation, evaluation committee to which the applicant was assigned, postdoctoral experience abroad, letter of recommendation, and affiliation with members of the evaluation committee. Perceived gender continued to matter. Lots.
According to the multiple-regression model based on total impact, female applicants started from a basic competence level of 2.09 competence points (the intercept of the multiple regression curve) and were given an extra 0.0033 competence points by the reviewers for every impact point they had accumulated. Independent of scientific productivity, however, male applicants received an extra 0.21 points for competence. So, for a female scientist to be awarded the same competence as a male colleague, she needed to exceed his scientific productivity by 64 impact points (95 per cent confidence interval: 35-93 impact points). 
So how much work does that amount to?
This represents approximately three extra papers in Nature or Science (impact factors 25 and 22, respectively), or 20 extra papers in a journal with an impact factor of around 3, which would be an excellent specialist journal such as Atherosclerosis, Gut, Infection and Immunity, Neuroscience or Radiology. Considering that the mean total impact of this cohort of applicants was 40 points, a female applicant had to be 2.5 times more productive than the average male applicant to receive the same competence score as he ((40+64)/40=2.6). [Emphasis mine.]
Let me repeat that. A female applicant had to be 2.5 times more productive than the average male applicant to receive the same competence score.

Sandstrom and Hallsten replicated this study in 2008.

There were not enough women on the review committees (5 out of 55 in 1995) to determine whether women equally favored male candidates. There are plenty of other studies, however, demonstrating that there's no significant difference between men and women in how they rate other men and women. Both genders and sexes seem to be equally subject to gender bias. Example: A study by Norton, Vandello, and Darley on how we rationalize favoring men.


I'm not ready to advise on what we should do about this. But here is the main update I'd like you to make: The women you meet are probably more agenty, rational, and efficient than you think they are, especially if you don't know them well. The men around you are probably more nurturing, communal, and compassionate. Your beliefs about them affect your interactions whether you're aware of it or not.


Unknown said...

This seems somewhat related to the statement I've heard said in discussions about IQ and race correlations - that even though a significant difference between the average IQ of two races may exist, your personal experience and information about any individual will screen off any information their race gives you about their intelligence

Anonymous said...

Based on a reading of the replication's abstract, it appears as though the replication came to the opposite conclusion.

healing said...

his appears to be fairly identified with the assertion I've heard said in conversations about IQ and race connections - that despite