Prior research has shown a few things. First, men are more prevalent than women in STEM majors and careers. For example, in 1985, 37% of people who graduated with a bachelor’s degree in computer science were female, and by 2011, that number had dropped to only 18%.[i] In the summer of 2014, many of the largest technology companies released their diversity statistics, detailing the proportion of women on the technology side (often defined as jobs related to programming, product development, and user experience). The numbers are alarmingly low: women represent only 10% of Twitter’s tech employees, 15% of Yahoo’s, 17% of Google’s, and 24% of Ebay’s.[ii] Dropbox, a startup valued at over $10 billion, had 9 women on its engineering team of 143 people in 2013.[iii]
And second, gender bias against women tends to occur when people judge women in "masculine" contexts. When hiring for male-dominated jobs such as police officer, people tend to prefer a resume with a man’s name to one with a woman’s name, despite the resumes being otherwise identical.[iv] Some evidence suggests that raters judge women as less competent than men with identical backgrounds.[v] A recent study, Moss-Racusin et al. (2012) had science faculty rate application materials of a student, randomly assigned a male or female name, for a laboratory manager position. The only difference between the applications was the gender of the applicant’s name, yet both male and female faculty rated the male applicant as significantly more competent and hireable than the female applicant, and were more likely to offer mentorship to that applicant.[vi]
A realistic depiction of the hiring funnel, broken down by gender, might look like this. Women start with fewer and get cut down quicker.
But, we also know that implicit biases tend to have small effects. Specifically, sex bias effects are quite small, accounting for only approximately 1% to 5% of the variance in work performance ratings. Now, some might use that fact to argue that there is little bias in the field, or that unconscious bias in the earliest stages is unimportant. I would argue the opposite.
Even with a small effect size, a gender bias can have cascading effects. Martell (1996) ran a computer simulation of a company with a pyramidal management structure with an equal gender ratio at the lowest-level positions. The company filled its higher-level positions with people from lower level positions depending on performance ratings. Even when sex differences explained only 1% of the variance of performance evaluations, only 35% of the highest-level positions were filled by women.[vii] Prior research indicates that the gender bias against women in tech hiring has an effect size of around 3%. In line with Martell's research, I created a simulation of the hiring process with a built-in bias in order to understand the cascading effects.
The simulation depicted a tech hiring process. It started with an equal proportion of men and women -- 50% of the total applicants were of each gender. Each person was assigned a competency score on a 100-point scale. The competency scores of men and women started out normally distributed with the same mean (μ = 70) and standard deviation (σ = 10). To reflect the 3% effect size bias, "bias points" were added to the score of each male applicant. Individuals who ended up with a score above 100 were reassigned to 100.
There were five stages, representing the Application, Phone Screen, Skype Interview, Onsite Interview, and Offer stages. During a simulated interview, the competency score of each individual received a random jitter (perhaps they interviewed very well and their score increased, or they performed poorly and their score decreased). Then, all candidates were ranked by their competency score, and only a certain number would move forward. This process continued until stage 5.
In certain iterations of the simulation, the bias continued throughout the hiring process, meaning that at each stage, men received bias points proportional to an effect size of 3%. In other iterations, the bias only existed in the first step. In both cases, the effect of the bias persisted and intensified throughout the rest of the process:
Even when a very small bias only exists at the first stage of the process, the proportion of women accepted drops from 50% to 36%. When discrimination continues, the proportion drops to 6%. As Martell says, “The effects of male-female differences are best determined not by the magnitude of the effect but its consequences in natural settings,” And in this case, taking into account the relative scarcity of women in the tech side, bias hurts women a lot. We can see it in the simulation as well -- small biases in interview evaluations create a negative cascading effect on pyramidal structure of the hiring process.
[i] National Center for Women and Information Technology. “Women and information technology by the numbers.” NCWIT, 2013. Web.
[ii] Lafrance, A. “Tallying female workers isn’t enough to make tech more diverse.” The Atlantic, 11 Aug 2014. Web.
[iii] Wadhwa, V. “Dropbox’s hiring practices explain its disappointing lack of female employees.” Washington Post, 14 February 2014. Web.
[iv] Glick, P., & Rudman, L. (2010). Sexism. In J. Dovidio, M. Hewstone, P. Glick, & V. Esses. The Handbook of Prejudice, Stereotyping, and Discrimination. 328-344. Thousand Oaks, CA: SAGE Publications Ltd.
[v] Foschi, M. (2000). Double standards for competence: theory and research. Annual Rev Sociol. Vol. 26(1): 21-42.
[vi] Moss-Racusin et al. (2012). Science faculty’s subtle gender bias favor male students.
[vii] Martell, R.F., Lane, D.M., Emrich, C. (1996). Male-female differences: a computer simulation. American Psychologist. Vol. 51(2):157-158.