Below is the short version of my full research essay – Inconvenient truths on gender inequality in STEM – How differences in choices explain disparities better than the discrimination narrative

I recall reading Manglin Pillay’s article on women in engineering with initial approval as I was a bit familiar with the research he alluded to. However, I found myself baffled by the public’s general negative reaction to it, which culminated in his removal as SAICE CEO. At that point I had to take a pause and reflect on the source of my own bafflement. Was Pillay sexist and if I failed to see it was I sexist as well? Did he get the science and research wrong or was the research not accepted because it did not fit into a particular narrative? As a fellow practicing engineer this question was not simply an abstract exercise but one that directly involved my own actions and attitudes which were responsible for creating a particular kind of environment for women who are a minority. The aim of the essay is an answer to my own set of questions. I will demonstrate that the main idea which I agreed with Manglin Pillay on, namely that gender inequality in STEM professions is not caused by discrimination; remains true. The first part of the essay will argue that the discrimination narrative; in both its implicit and explicit forms cannot explain why there are gender disparities in STEM. The second part of the essay will be a bit more ambitious and difficult and demonstrate that different choices result in disparities. The differences in choices is influenced by three factors. Firstly by gender differences in particular cognitive abilities of verbal, quantitative and visuospatial as well as the academic strengths of an individual. Secondly, differences in vocational interests with men in general being more thing-orientated and women being more people-orientated. Thirdly, gender differences in lifestyle preferences and values.

The term STEM is too broad because if we looked at bachelor pass rates in STEM as rough measure of inequality we would find there would be none. Data from the US for bachelor passes in 2014 showed that females earned 50% of all science and engineering degrees. Similar results in South Africa were found by examining the 2008 matric cohort; researchers found females earned 63.2% of natural science degrees. So the problem of STEM inequality, atleast at the bachelors level as a starters, is ill defined because inequality exists in certain STEM fields and not others. The female dominated STEM fields are the life and health, social and psychology sciences (abbreviated as LPS); the male dominated fields are geo-sciences, mathematics, computers, physical science and engineering (abbreviated as GEEMP). Females in the US in 2010 earned just over 60% of bachelor degrees in LPS fields and just over 25% in GEEMP fields. So the question is why are women under-represented in GEEMP fields?

The discrimination narrative captures the broad general sentiment that bias against women is the root cause for disparities in male-dominated STEM fields. The discrimination is often thought to exist in both explicit and implicit forms; and in whatever form it takes the end result is a clear signal to woman that they are not welcome in those GEEMP fields.

In a series of articles analyzing gender inequality in science, the prestigious science journal, Nature commented that, “Why has progress stalled…unconscious bias which exists even in those that support stem”. So then implicit bias is considered a key factor for gender inequality in STEM. But what is it? Implicit bias is thought to be the unconscious biases and negative stereotypes we have against particular groups which affect our behaviour towards them. One of the popular psychological tools for measuring implicit bias is the Implicit Association test (IAT). Research conducted across 34 countries and with half a million participants by Nosek and others found that there was a positive correlation between the male advantage in 8th graders’ math scores and implicit association scores which associate men more strongly with science than with women. Studies such as this are presented as evidence that negative stereotypes that women do not have the ability to succeed in STEM lead to both men and women making choices that then conform to these stereotypes. The conclusions are unwarranted. Even if we conceded that the IAT is measuring stereotypes it does not necessarily mean that it is inaccurate and negative as psychologists have argued that stereotypes are generally accurate. Secondly, psychologists have also found that people use stereotypes (defined as features about groups) rationally; when given particular information about an individual we generally rely on that rather than our stereotypes to form judgements. So if we take as an example that I believe that more men are scientists than women; that does not mean I believe that it should stay that way and will act in ways to keep it that way.  Stereotype threat is the phenomena where knowing a certain stereotype about a group I belong to causes me to conform to it. An example would be a woman being made aware of a negative stereotype such as “women are bad at math” and that awareness causes her to perform much more poorly than she would have in the absence of knowing about the stereotype. A meta-analysis of stereotype threat found that its effects are negligible and therefore cannot have an impact on gender inequality in STEM.

Then there are general problems with IAT as the most popular tool for measuring implicit bias. 1) Does it measure implicit bias? 2)Does it correlate with actual explicit behaviour? The answer to both these questions is no. The IAT rather than measuring prejudice is actually measuring widely shared cultural associations which people, more importantly, might not endorse. For example I might associate jail more with men than with women because criminals as a group are mostly men. However, that does not mean I endorse it and believe that things should be that way. Interpreting the test this way avoids the absurd conclusion that not only men, but women are unconsciously sexist against themselves because the IAT has claimed to measure equal amounts of implicit bias in both genders. Further problems with the IAT when it comes to measurement is that it has very low test retake reliability. I can take the test multiple times and getting widely varying scores which is surprising if it supposed to be measuring a stable and deep phenomena.

When it comes to whether implicit bias predicts behaviour two meta-analytic studies have found the following. Firstly, the IAT does not predict explicit prejudiced behaviour or attitudes. To put it another way – if Verwoed and Mandela were to take the IAT test it would not be able to predict who is likely to discriminate against black people. Secondly changing people’s implicit measures on the IAT does not lead to changes in actual explicit behaviour. So then the popular tool which is supposed to tap into our unconscious, even if it measuring an implicit bias, it does not mean that we will act in ways consistent with that implicit bias. Therefore, we can safely conclude that claims that implicit bias is causing women inequality are simply false.

When it comes to explicit bias; research has found that no systemic bias against women exists even though there are particular instances where bias has been found. More importantly, the instances of bias against women are insufficient as causes for the disparities in GEEMP fields.

 If we look at what has been called the leaky STEM pipeline, which is the idea that within academic science women are more likely to leak from the pipeline as they transition from the time they enter university up until they receive tenure as professors. The leaky pipeline is considered to be evidence that STEM has some bias and hostility which deters and causes woman to exit the pipeline. Research has found that from 1990 for US institutions; women were equally likely as men in persisting from bachelors to PhDs. This was true for male dominated STEM fields of engineering and physics as well. When looking at the transition from PhD to tenure as professor surprisingly the GEEMP fields (which are supposed to be more biased) show equal persistence rates; while the LPS fields show women having less persistence rates than men. The difference disappears once family factors and having children is controlled for; with research finding that single women in LPS fields persist at higher rates than men.

Furthermore fairness and a bias in favour of women was found when it comes to the hiring of faculty members for assistant positions and tenure positions. Ceci and Williams conducted a hiring randomized experiment where 873 faculty members had to evaluate and hire individuals with similar academic profiles for assistant tenure track positions. They found a 2:1 preference for women over men in all fields; this was for both stem and non-stem fields with economics faculty the only ones with no bias. William and Ceci also analysed three large scale studies of actual tenure-track interviewing and hiring in U.S institutions and found that there is either a bias in favour of females or no bias at all. One specific study they highlight examined six math-intensive fields in over 500 departments at 89 research intensive universities, with 1800 faculty members. The study examined actual tenure-track positions hiring data between 1995 and 2003 and found a slight bias in favour of women when it came to interviews and hiring of candidates. Although various experiments have found bias in particular instances, such as the well cited paper where bias was found against women when being evaluated for a lab manager position. The particular finding fails to generalize to science as a whole and therefore can be ruled out as a cause of gender disparities.

Other areas where bias is thought to exist is during the careers of academic scientists. Publishing, applying for grants, citation rates and salaries are often claimed to be biased against women. Williams and Ceci reviewed the literature and found no bias against women in reviewing and publishing of manuscripts, as well as in citation of published papers. When it comes to the awarding of grants, six large scale studies which independently assessed gender bias at the National Science Foundation (NSF); National Institute of Health (NIH); US Department of agriculture; Australian Research Council; German Foundation between 1985 and 1990; Economics Program at NSF between 1987-1990; concluded that no gender bias against females was present in the awarding of grants. When it comes to the gender pay gap in STEM various researchers have found it ranges from 14% to 28% in favour of men. However, once you control for a number of variables such as age, race, level of highest degree, marital status, presence of children, current job and work time, employment sector, years worked: the gap disappears and reverses for single women without children and declines significantly for married women with children. Single women without children have a pay gap advantage of 5.4% and married women with children have a gender pay gap disadvantage of 12.4%. Kahn and Ginther also found that research by the US commerce department showed that the gender salary gap in STEM is lower than in non-STEM occupations. A surprising finding if we expected that STEM fields have higher levels of bias against women which leads to them exiting STEM fields.

When zooming into the engineering workforce, which has often been called the “old boys club” and tends to have the lowest female representation amongst STEM fields with estimates ranging from 6% to 11% across the world, we find surprising results. The claim is that the male-dominated engineering workforce presents a chilly and hostile environment to women. This results in them leaving the engineering field altogether and acts as a signal to other women not to consider engineering as they do not belong, or fit in. Hence there is a gender retention gap of roughly 10% in engineering, meaning women are less likely to be retained than men. There are two similar but distinct questions which research has tried to answer.

The first is whether the retention gap in engineering is unique to that field alone, or do other non-science and engineering fields experience similar retention gaps? Hunt in her research provided an answer to this question and performed a series of regression analysis holding various variables constant and found that engineering field had excess exit rates (5.8%) compared to non-engineering fields. She also found that as the share of males in the field increased the female exit rates increased as well and this was true for all fields and not just engineering fields. She took this as evidence of a chilly climate – the more men in the field the more inhospitable and unfriendly and hence the greater the women exit. However her analysis did not take into account research by Lippa as well as by Su and Round which found the following. Firstly, the more an occupation is people-orientated the higher the female share in it, and the more an occupation is thing-orientated the higher the male share in it. Secondly, women in general tend to have vocational interests that are people-orientated and men have interests in vocations that are thing-orientated. Therefore the male share in an occupation is correlated with whether it is thing-orientated. Hunt’s research fails to include this fact and if included would explain why women, who tend to be more people-orientated, would exit thing-orientated jobs more.

The second question is why is there a retention gap at all in engineering? Kuhn and Ginther in their work found that the gender retention gap declines from 7.8% to 1.6% when individuals that are working part time or unemployed are excluded. It means then that the differences in retention is due to more women with engineering degrees working part time or leaving the labour force altogether. Kuhn and Ginther found that women with children are more likely to leave the engineering industry and the labour force entirely and account for a large chunk of the average retention gap; 72% of the gender retention gap is explained by family factors and women with children. Women without children have better retention rates in engineering than even men. The gap between women with children and women without children ranges between 15%-25%. The authors state: “…single women without children are actually more likely than men to remain in engineering. Children have the greatest effect pulling women out of the labor force and thus out of engineering jobs”. This conclusion has been confirmed repeatedly. Ceci and other researchers in their review found that PhD women in academic STEM are more likely to leave STEM and more likely to do so for family reasons as well. A recent study by Cech and Blair-Loy found similar results. New parents are far less likely than similar childless individuals to remain in STEM fulltime after the birth or adoption of their first child. New parents who continue to work fulltime are more likely to leave STEM and work fulltime in non-STEM field. It was also found that new mothers were twice as likely as new fathers to have left full time STEM work. So then combining family with work remains a significant challenge for women in STEM.

I want to return to the claim that high male share in a field leads to a male-dominant and hostile culture to women which prevents women participation from increasing in the field. The reason it fails highlights a general problem with other explanations about why GEEMP fields have gender inequality. It fails to explain how other professions were able to overcome the same problem and yet engineering could not. In the U.S. women made up 8% of pharmacists in the 1960s and yet today they make up 55% (Women reached the 50% mark by 1980). Clearly pharmacy then was an “old boys club” with such a low female share and yet that did not stop it from seeing women increase their participation in the industry reaching parity within 20 years. All STEM fields started with low female enrolments; in 1910 women earned 22% of bachelors in psychology and yet within 20 years that number managed to double (National Center for Education Statistics, 2017). Therefore the discrimination narrative and its associated consequences (negative stereotypes, lack of role models, low confidence) must explain historically why LPS fields like psychology managed to overcome the effects of bias (which I would assume was far more pervasive in 1920) and yet GEEMP fields like engineering have been unable to do that today. Is the kind of bias found in GEEMP radically different and more severe than what was present in all the other fields which have managed to achieve gender parity? I doubt it.

In conclusion then, the popular idea that gender inequality in STEM must be the result of bias and discrimination against women is simply not true. Discrimination, whether implicit or explicit, fails as a general explanation for the disparity seen in male-dominated STEM fields. The STEM is one where there is in general fairness when looking at a whole host of different aspects from hiring, citation, publishing and grant funding. You also find that once you control for a number of variables, particularly the impact of family, the differences between men and women persistence rates, retention, and salary gaps mostly disappears. We must let go of the false but popular narrative that where we see unequal outcomes in gender then bias must be responsible. It leaves us chasing figments of our imagination, placing blame where it should not and ultimately wasting valuable time and resources in solutions that actually do not address the root problem.