The Conundrum of Bias – Understanding Gender Differences in Academia

Bias in academia is often difficult to define and separate from differences. Responding to a recent call from the journal Nature to establish new principles for research on race and ethnicity, Vincent A. Slow And Waltman gameDescribe how the concepts of causation can clarify approaches to the study of gender bias in higher education.

Equity, diversity and inclusiveness are pressing issues for research and higher education institutions. To ensure that research does not unintentionally harm underrepresented groups, Nature recently introduced new rules for research on issues of race, ethnicity, or gender. Nature encourages researchers to reflect on these concepts and clarify how they are defined and measured. For example, if gender is included in a study, is it self-reported or inferred using machine learning algorithms? Besides, Nature asks researchers to explain how they controlled for confounding variables. Why is it so important to control mixed variables and why is it so difficult to do it right?

Let’s look at a case of gender bias in academia. Researchers regularly observe gender differences in favor of men in various academic settings such as fewer women in top academic positions, fewer publications, lower citation rates And less funding women. However, researchers also note differences in favor of women, such as more women are elected to NAS, more favorable peer review And higher funding rates for women. Drawing conclusions about gender bias from these differences is very difficult, not least because it often remains unclear what is meant by “bias”. This raises difficult questions about confounding variables and how to control them.

What do we mean by bias?

IN our recent workwe propose to define bias as unjustified direct causal effect. This is a rather technical definition based on ideas from the field causal inference. The essence of this definition is to prejudice people with a certain characteristic, such as a certain gender, if they are unjustifiably treated differently because of that characteristic. For example, we talk about gender bias in funding if the exact same grant application, with the same resume and the same track record, would be funded if the proposal was written by a man and not a woman (or vice versa). ). If the same proposal is priced the same regardless of the gender of the applicant, there is no gender bias.

Gender bias may lead to other issues in the future, as shown in the figure below. Suppose there is a gender bias in the recruitment of new research staff at a prestigious research institution. Men are hired more often than women. Let us also assume that researchers at prestigious institutions are more likely to receive external funding than researchers of similar qualifications at other institutions. This will result in fewer women receiving funding. While there is no gender bias in funding, gender bias in hiring at a prestigious institution indirectly negatively impacts women’s chances of obtaining funding. To distinguish this indirect effect from the direct effect, we refer to the indirect effect as gender disparity in funding rather than gender bias.

A prestigious research institution (left) is recruiting new research staff. There are 100 researchers applying for jobs, 50 men and 50 women. On average, candidates of both sexes have the same qualifications. The prestigious institution recruits 20 women (10 highly qualified and 10 qualified) and 30 men (10 highly qualified and 20 qualified), showing bias against women in the recruitment process. Rejected candidates find employment elsewhere, at a less prestigious research institution (right). All 100 researchers then apply for external funding. There is no gender bias in funding decisions, but decisions are influenced by the prestige of the applicant’s institution, in addition to the qualifications of the applicant. All highly qualified researchers receive funding. Among qualified researchers, 70% of those who work in a prestigious institution receive funding, and only 20% of those who work in a less prestigious institution. As a result, funding was provided to 23 women and 28 men. Even though there is no gender bias in funding, women are indirectly affected by gender bias when they are hired by a prestigious institution. This gender bias in recruitment leads to gender disparities in funding. Taking into account the results of funding only in a prestigious institution, 17 out of 20 women (85%) are funded compared to 24 out of 30 men (80%). This higher level of funding for women is a consequence of gender bias against women in employment, resulting in relatively more highly qualified women working in a prestigious institution.

Difference does not mean bias

Just as correlation does not imply causation, our definition of bias emphasizes that gender differences do not imply gender bias. This is why it is necessary to control for confounding variables in order to draw correct conclusions about gender bias. At the same time, taking into account the wrong variables leads to wrong conclusions.

Consider again the case of gender bias in employment mentioned above, where women have to pass a higher bar than men in order to be hired by a prestigious institution. A consequence of this gender bias is that women hired by prestigious institutions are, on average, more highly qualified than men. As a result, relatively more women than men will receive funding in this institution, as shown in the figure above. This gender difference in funding does not imply a gender bias against men in the competition for funding. This is simply a consequence of gender bias against women in employment.

Providing sound policy advice

Distinguishing between gender differences, gender inequality and gender bias is critical to enable researchers to provide sound advice to policy makers. For example, suppose the researchers erroneously conclude that in the example shown in the figure above, there is a gender bias in funding. When they inform politicians of their findings, they can try to improve the funding process. This will be inefficient as there is no gender bias in funding. There is a gender disparity in funding caused by gender bias in hiring at a prestigious institution. So instead of trying to improve the funding process, policy makers should focus on improving the recruitment process at that institution. Addressing gender bias in hiring is not only important in its own right, but will help address gender disparities in funding.

However, a deeper analysis may show that there is in fact no gender bias when applying for a job at a prestigious institution. Instead, gender differences in hiring may show up elsewhere. For example, suppose that, on average, women spend more time caring for children than men, and as a result women post less than men who spend evenings writing articles. If having more publications increases the likelihood of being hired, we will see more men being hired. This is not a gender bias in hiring, but a result of the difference between how men and women spend their time between childcare and scientific publishing.

Now suppose that politicians impose a hiring quota for women. How will this affect those who are hired? The introduction of a quota may actually have an effect that goes against what was intended. If hiring decisions continue to be heavily based on publications, the introduction of a quota will increase the chances of women being hired, but it will benefit women in particular, who prioritize publications over childcare. The quota will lower the chances for men, especially those who prioritize childcare over publishing. While the gender balance will improve, the quota will only reinforce the idea that time spent caring for children hurts career opportunities, especially for men.

Towards a solid evidence base

Effective policies to improve gender equality and gender diversity require careful analysis of the causes of the gender patterns we observe. Observational studies inevitably run into the problem of causal inference. Experimental studies can alleviate some of these difficulties, but they are generally difficult to carry out under real conditions. Qualitative research can address some of the complexities of causation but faces other limitations. Triangulating data from these different approaches seems to be our best way to develop a robust understanding of gender bias.

The terminology we propose offers a way to carefully distinguish between biases and inconsistencies, which are direct and indirect causal relationships, and differences, which are not necessarily causal. This terminology is intended to facilitate a clearer discussion of issues related to gender, race, and ethnicity. We hope that researchers will use our terminology and the underlying causality to implement Naturenew guidance on the control of confounding variables.

The content created on this blog is for informational purposes only. This article represents the views and opinions of the authors, but does not represent the views and opinions of the Impact of Social Science blog (blog) or the London School of Economics and Political Science. Please see our comment policy if you have any concerns about posting a comment below.

Image credit: adapted from Pavel Chervinsky via Unsplash.com.