Thursday, March 13, 2014

Studying Gender Stereotypes in Science. Or We All Know That Women Can't Do Math


Are there fewer women in STEM-fields just because women don't choose them as often as men?  Or could it be that there are demand-side reasons for the relative scarcity of women?  Those "demand side" reasons mean that the people responsible for hiring and promoting workers might have (perhaps subconscious) prejudices about women which affect the likelihood that a woman is picked for a job or an educational slot which requires mathematical skills.

A new study, How stereotypes impair women’s careers in science, by Ernesto Reuben, Paola Sapienza and Luigi Zingales tries to answer the latter question, about the possible impact of our prejudices concerning mathematics and gender.   Bryce Covert summarizes the study findings:

Researchers from Columbia Business School, the Booth School of Business at the University of Chicago, and the Kellogg School of Management at Northwestern University conducted an experiment that had both men and women complete an arithmetic task that both genders, on average, perform equally well as potential job candidates. Then test subjects had to decide who to hire. “Our results reveal a strong bias among subjects to hire male candidates,” the researchers note, which was true of both men and women. When the prospective employers were only shown a candidate’s physical appearance, making their gender clear, they were twice as likely to hire a man than a woman. This was because women were expected to perform worse on the math problems, even though it was a task they were equally like to do well.
Women were still less likely to be hired even after the candidates told prospective employers how they did on the task “because men tend to boast about their performance, whereas women generally underreport it,” the authors write. Employers don’t take this into account, particularly if they went into the experiment with a strong bias against women in math.
Things improved when those doing the hiring were given full information about how the candidates did on the task, but even then discrimination wasn’t totally eliminated.



The study design is complicated* (sadly, the  article is behind a pay-wall, though I have read it).  I believe that this is because it had to satisfy several criteria.  For instance, the participants were not supposed to know that the study was about the impact of gender, and the study also tried to imitate the hiring markets for STEM candidates without revealing to the participants that they were playing the roles of job candidates and employers.  As an example, those two roles were called being "a contender" and being "an observer."  Yet the small payments the participants received from the study did vary in the way incomes and profits would vary in an actual labor market.

What matters for the purposes of standardizing the demand side variables is that the task the subjects were to be perform was not a) voluntary, once a person had agreed to be in the study and b) did not show statistically significant gender differences in either the mean score or in the variance (scatter of scores).  These two aspects of the study meant that it controlled for differences in desires to do the kind of job the study measured and for the ability to do that job.

The design of the study also means that in the absence of employer prejudices or sexism the candidates should have been hired at equal rates (thus meaning that 50% of the hires should have been female).

Yet this was not the case, even when "objective" information about the candidates' actual scores were provided by the researchers.  Indeed, if all the information the "observers" had consisted of just the genders of the "contenders," the percentage of women hired  was 33.9%.  When the "observers" were also given the "contender's" own estimate about his or her performance of the same task in the future, the percentage of women hired did not rise (33.7%), and even with the objective information, the percentage of women among the hired climbed only to 43.0%.

The main conclusion of the study is that existing stereotypes about mathematics and gender affected the results**.  Those stereotypes were measured using the Implicit Association Test (IAT).  The average scores on the IAT showed that both male and female study subjects linked mathematics and being male.  Thus, female "observers" were no more free of the stereotype than male "observers."

Which is not surprising, given that we are all exposed to the same sources of information and misinformation.
---
*This post doesn't aim to seriously scrutinize the study as a whole.  But I didn't spot anything that looked off to me (though I'm not an expert on psychological tests etc.), save perhaps for the use of undergraduates in a study of this kind.  It could be that people whose job it is to screen job candidates in STEM fields have gone through a process where they have become aware of their prior biases and can control them.  Or not.  But a study using students wouldn't be able to address that hypothesis.

**It's important to note that the study subjects were asked about picking a contender to do the same type of calculation again, not some totally different type of mathematical problem.  This controls for the possibility that prior beliefs about men being better at certain types of problems could have been informative.

Added later:  Another somewhat related study (which I haven't read) argues that investors prefer men to women, both pitching the same idea,  and that attractive men do better than not-so-attractive men.