Can data change the way we think about gender and research?

An Elsevier technology director talks about ‘invisible bias,’ the pitfalls of AI and why gender balance is crucial in research

By Karin Hilton - April 16, 2020
Karin Hilton EWIT Tom Leishman
Karin Hilton, Senior Director of Technology at Elsevier, speaks at the European Women in Technology summit in November. (Photo by Tom Leishman)

We all know that data has become ubiquitous, but have we realized just how much this has changed the way we look at research? And how much light it has shed on our erroneous ways thinking that we fail to challenge?

Research has shown us that diverse thinking comes from diverse groups – and that inclusion is important for driving innovation and coming up with ideas that might otherwise have been missed.

When we published our second gender report in 2017, we discovered where gender diversity was lagging. In Life Sciences research in the US, for example, 50 percent of doctorates are awarded to women. Yet when we go up the academic chain and into the world of research, the percentage of female professorships drops to 21 percent. And by the time we get to the highly influential position of department chair, that number slips to 15 percent.

As we see in our latest report, published in March, these trends continue.


The Researcher Journey Through a Gender Lens - report

The researcher journey through a gender lens

An examination of research participation, career progression and perceptions across the globe

Download the report


When we look at the number of researchers cited as first authors (by convention the junior scientists who are doing the legwork), women are 15 percent more likely to be cited as first authors yet 20 percent less likely to be cited as principal investigator – that is, the person who decided the question and was granted the funds to investigate it. This is a big dichotomy that mirrors the drop we see in the institutional numbers. And it impacts the way people think.

The first part of this slide illustrates while 50 percent of doctorates in the US are awarded to women, just 21 percent of professorships are held by women and 15 percent of department chair positions. The second part shows that while women are 15 percent more likely to be cited as first authors, they are 20 percent less likely to be cited as principal investigators.

Invisible bias

We know that (invisible) bias permeates the world around us. It comes with one’s background and experience. I'm going to give you an example from my alma mater, the world of archaeology: let’s look at how gender assumptions have influenced the way research findings have been presented to the world – and just how sticky these assumptions really are.

This slide shows the remains of the Birka Viking Warrior (left) and the ‘Red Lady of Paviland’ (right). Karin Hilton, Elsevier’s Senior Director of Technology, used it to show the lingering power of gender stereotypes at the <a href="https://www.europeanwomenintech.com/" target="_blank">2019 European Women in Technology summit</a>.

In the image above, you see Birka Viking Warrior at left. The skeleton was found in Sweden in 1870, surrounded weapons and shields, and immediately identified as a man. This claim was challenged in 1970 and again in 2014, when osteological analysis indicated that the bones were those of a woman. In both cases, the hypothesis was rejected, as the grave goods clearly indicated a man. It was only when DNA analysis conclusively proved the skeleton to be a woman’s that the record was changed — in 2017.

In fact, gender assumptions go both ways: the Red Lady of Paviland, at right, was excavated in 1823. The bones were painted in ochre and the skeleton was surrounded by jewellery, like a prostitute or a witch. It's only with positive logical analysis and lithographic studies that anyone thought to challenge the fact that this lady is in fact a man. Yet today, we still call this the Red Lady of Paviland.

Personalized medicine?

Let’s go back to the health world. In 2011, African Americans and Hispanics comprised 12 percent and 16 percent of the US population, respectively, but only 5 percent and 1 percent of trial participants were African Americans and Hispanics. Other research indicates that studies of cardiovascular disease are particularly male-biased. A review of 19 randomized controlled cardiovascular trials found that only 27 percent of the participants were female and that only 13 of the studies presented sex-based analyses of the data. This has real-life life-and-death impact.

Researchers in the 1970s and 80s made some massive breakthroughs in the way we treat heart attack and strokes and had the ability to change the survival rates dramatically. Campaigns of public information encouraged people to recognize the symptoms. But the number of women dying of heart attacks and strokes did not fall. Why? Because women were not in the initial research trials, and we failed to realize that the symptoms women present are often very different. As a result, their heart attacks were often not being recognized until too late.

The lack of diversity also encompasses ethnicity. In the US, fewer than 8 percent of cancer drug trials reported representation from the four major races (white, Asian, black and Hispanic).

A few years ago, when a large pharmaceutical company launched a lung cancer drug (getfinitib/IRESSA) in the US, the results were disappointing. Some of their trials results were good, however, and they identified a common pattern: that the research was done in Asia and generated best results in Asian women who were non-smokers below the age of 40. The drug is still out there saving or prolonging thousands of lives in Southeast Asia because it targets a specific gene that is highly common in that population. This tailoring of research data towards very precise markers/patterns present in ethic/diverse populations is how we can move in the world of research towards personalized medicine — the holy grail of medicine.

We are making progress in moving the needle in including diversity in trials, and organizations like the FDA have significantly revamped guidelines to encourage diversity, but there is still a long way to go.

Because AI is blind …

It’s not just about having invisible data; it's also about the quality of the data and the way you use that data. There are plenty of examples in the world of AI showing that the data you feed to an algorithm impacts the way the machine responds, from the Google photo platform that recognized a black man as a gorilla, to the Amazon recruiting tool that deliberately downwaged women, to the American judiciary system tools which makes black African males more likely to be sent to jail and to have longer jail terms.

Twitter image - Google Photo platform

Increased diversity would bring a change, but the data shows us that is still very much a challenge. We see that AI papers that are co-authored by a woman tend to be more applied. Female authors have a broader social awareness, they think in a different way. However only 13.8 percent of women get doctorates, 25 percent of institutions report women in their AI courses, and 20 percent of attendees at AI conventions are women.

Elsevier supports and organizes conferences around the world. In  2015, we recognized that we were not good enough: delegate attendance was 60:40 male-female, and invited speakers were 85 percent male. Since then, we’ve taken significant steps to change. We’ve moved from 15 percent female speakers to 32%, with no all-male panels, or “manels.”

We're also looking at the way papers get accepted into journals. The higher caliber the journals, the more likely you are to be cited, and the more likely your research is to be recognized. So something we found in 2015 caused us to take pause. We looked a the editorial boards of our Earth Sciences and Energy journals, and only 6 percent of our editors were women. Now the percentage is above 17 percent.

Or take The Lancet, our flagship medical journal, where we have achieved 50-50 across all editorial boards. Diversity is not only about those who do research but those who are challenging and assessing the quality of that research.

How should we measure success?

Research papers by women have almost identical download impact and citations and are looked at almost as many times as those by men, as Elsevier has shown in its latest gender report. It’s important to bring more balance in all kinds of diversity, especially by funding research by researchers from diverse backgrounds. And yet male investigators are 1.4 times more likely to receive funding, and mentors are 5 percent less likely to comment favorably on female scientists in terms of their research potential and qualifications. As a result, women generally funded less than men, whether you look at institutions, funds types or disciplines. And because your second round of funding depends on how much you've gotten your first round, and the third round depends on your second, this becomes a compound challenge, and women get increasingly less funds.

It's not all doom and gloom, though. The world of research is changing, and we at Elsevier support that. We make tools that help individuals and funding organizations connect with experts in their area of interest to assess the value of those ideas and how they can make change. How else does Elsevier support this change? Are there two other things you could mention in passing?

Research is fraught with challenges, but it is indeed a world that we're changing because the prize is enormous. The prize is personalized medicine. The prize is tools that recognize each and every one of us and our needs.

This article is based on Karin Hilton’s talk at the European Women in Technology summit in Amsterdam in November.


Related resources

Gender report main image
Kumsal Bayazit quote
HDSI gender panel

Contributors


https://www.elsevier.com/__data/assets/image/0010/1002898/Karin_Hilton.jpg
Written by

Karin Hilton

Written by

Karin Hilton

As Senior Director of Technology at Elsevier, Karin Hilton is responsible for the technology team that build Mendeley Data, Elsevier’s platform to support research data management. Karin and her team are using technology to transform the way that researchers collaborate and work together to spark discovery and support innovative interdisciplinary thinking.

Karin has held technology leadership roles at companies across a range of industries. Before joining Elsevier, Karin was a technology director with eBay Classifieds Group, heading up core capability technology platforms on a global basis and challenging how they respond to mobile wave as an organization. She is based in Amsterdam.

US business students visit Elsevier to learn about Pride and I&D in the workplace
Increasing (gender) diversity on a journal: a case study

Comments


comments powered by Disqus