Inside the Gender Data Divide

SMITH BRAIN TRUST –  At a time when data can influence much of our lives, it turns out that the data that researchers collect, broadly speaking, tend to be based on men’s – more than women’s – experiences. This leads to designer and developer output that’s more effective for men than women in areas such as healthcare, automobile safety, office-building design and urban planning.

Maryland Smith’s John Bono and Lauren Rhue, in an interview recently with Maryland Smith’s China Office, add context to the phenomenon. Here is an edited excerpt from that conversation.

To what extent is the “data divide” between men and women changing?

Rhue: From my perspective, the gender data divide is a part of a larger phenomenon about the consequences associated with insufficient representation in data. In the era of big data, in which machines learn from historical data, problems of male overrepresentation can lead to products and services designed for men. Although the gender data divide may happen because researchers do not gather enough data from women, it can also occur because of historical data that reflect societal gender roles. For example, a language-categorization AI learned to associate computer engineer with man and homemaker with woman, a gendered categorization that mirrors the gender imbalance in those careers. There are statistical techniques to correct for the gender imbalance in data, so although the gender data divide persists, it is decreasing as researchers and others become more aware of the problem and more receptive to addressing it. However, because the gender data divide is often non-obvious, researchers must consistently examine their data collection methods, their datasets, their techniques and their outcomes for evidence of gender disparities to avoid the gender data divide.

What have been some significant manifestations of the gender data divide?

Rhue: The gender data divide is pervasive in society, affecting areas as diverse as technology, human resources, finance, and Wikipedia, and researchers are finding more and more examples of gender disparities arising in unexpected ways. For example, in late 2019, Apple Card was investigated after it assigned higher credit limits to husbands than to wives. Apple explained that the different limits occurred because they did not include gender as a variable in their data, but their credit product learned to assign lower credit limits to women than men, even when the couple had joint accounts. Again, to address the gender data divide, we must develop ways to identify and address these issues before products and services hit the market.

Although the gender data divide persists, it is a good sign that many organizations are taking steps to address it. For example, Amazon pulled its AI recruiting tool after it exhibited a preference for men. Microsoft added more representative samples after its facial recognition algorithm exhibited bias. Once organizations realize that their products and services are not equally efficient for men and women, organizations and researchers can change their data collection and their statistical methods to reduce the effect of male overrepresentation and reduce the gender data divide.

What are inherent strengths and weaknesses in big data in terms of the gender data divide?

Bono: On the one hand, more information can yield better insights into preferences, behavior and lifestyle. If data are collectible from a wide range of sources across populations, the potential use for analytics substantially increases. However, if the premise is ‘more data is good,’ one must look at the sources for all the data being acquired: websites, blogs, social media, sensors, and other devices. Each of these has two underlying themes in common: wealth and opportunity. A person must have the wealth to be able to acquire the tools to participate in these mediums and the opportunity to participate with relatively frequency. Here’s where the gender pay gap comes in, as multiple studies and statistics have shown women often make less, on average, for the same job as men. Those with lesser incomes may not have the tools needed to participate nor the desire or opportunity to engage in the same way others with more wealth might participate. Given the disparity in how the data are collected, more skewed data under these conditions will not necessarily be as helpful for solving the data divide.

Why is promoting gender equality significant in social development?

Bono: Aside from the obvious ethical and moral reasons, promoting gender equality is a gateway to economic and political development. An increase in social awareness of issues related to gender equality would help to reduce violence toward women and help increase women’s rights, especially in countries that do not treat women as equal to men. As primary caretakers in many families, women are responsible for the health and wellbeing of family members, and disruptions in family life, such as lack of food, causes many women to shoulder the burden to find solutions. These types of realities can distract women from their aspirations to become entrepreneurs, innovate, solve problems, join political causes, et cetera. Better promotion of gender equality would help ease some of these burdens and allow for greater access to social roles and opportunities traditionally more accessible to men.