Posted by n.chuehong on 3 May 2016 - 11:22am

By Toni Collis, Director - Women in HPC.

"What is a good (or bad) percentage of female users of a scientific computing facility?"

In this article, originally published to accompany a talk given at the recent public opening of Grace, UCL's new computing facility, Toni Collis uses this question to look at the diversity of the community in general, focussing on the High Performance Computing (HPC) community that represents the large-scale use of scientific computing. By looking at other studies, including analysing data collected by the Institute about research software use and development in the UK, Toni calls out the things we still need to be aware of if we are to continue to honour the work of pioneering computer scientist Grace Hopper, inventor of the first compiler for a computer programming language.  

Toni Collis is Director of Women in HPC, and works with the Institute's Policy team as a collaborator on diversity and engagement campaigns.

I’d like to begin with a question. It’s the question that drives us.

What percentage of registered Grace users are women?

• A: 43%
• B: 22%
• C: 12%
• D: 5%

The answer is 22%.

So the question that drives us is – is this good or bad? What else do we know about the proportion of women in the broader scientific computing community? This is a question I have been tackling since the start of Women in HPC.

To start, we can look at other large-scale scientific computing systems to gather more information. ARCHER: the UK National High Performance Computing (HPC) facility, mainly performing science for EPSRC and NERC researchers, has 17% of its users that self-report as female. By this comparison Grace is performing better. Closer in line with the Grace figure 21% of attendees at ARCHER training events are women.

Looking beyond HPC systems to conferences, the Exascale Applications Conference (held in the UK in 2013 and again in 2015) had 9% and 15% of attendees that were female respectively. The 2013 Partitioned Global Address Space (PGAS), a more specialised HPC field, had just 5% women, the majority of whom were from the hosting institution. This is a lot lower than the proportion of ARCHER and Grace users that are women.

We see this replicated on the international stage as well. The largest HPC conference in the world – Supercomputing – held in the US each Autumn, with over 12,000 attendees, reported that 13% of attendees were women in 2015.  Perhaps these numbers are lower than those using our services because women are less likely to travel to conferences?

Let's look at where this community comes from. There is not (currently) an undergraduate degree in HPC or supercomputing, instead much of the community comes from physics, chemistry, biology, computer science, engineering, environmental science, and increasingly from social sciences and humanities though these numbers are still small. I started out in physics before moving into HPC during and after my PhD. So what should we expect there?

To look into the gender split within academia, we can turn to the Higher Education Statistics Agency. From this we get the following gender profile of academic staff in English Universities in 2013/14:

• Biological sciences – 45%
• Earth, marine & environmental science – 36%
• Mathematical sciences – 29%
• Chemistry and materials science – 27%
• Computer sciences – 22%
• Engineering & technology – 19%
• Physics & Astronomy – 17%

By this measure, the 22% of Grace users sits in the middle to lower range of the people that could potentially feed into the HPC community. However, all of these percentages except the 17% in Physics and Astronomy are higher than the proportion of ARCHER users and all are far higher than the proportion of women attending HPC conferences.

My final data point to help us establish the percentage of women in the HPC community is to look at scientific publishing. A 2011 study of ACM affiliated conference proceedings revealed that between 1967 and 2009 the proportion of female authors on ACM published conference papers rose from 7% to 23%. The 23% is in line with what we are seeing with Grace, though obviously this includes a broad range of computational areas, not just HPC. And as we have just heard there are more women in Computer Science than in some of the other fields such as Physics that are one of the largest users of HPC.

Preliminary findings of research being carried out by the Women in HPC initiative suggests that for HPC conferences the proportion of conference paper authors that are women is far lower, with the more specialised conferences having the lowest proportion of female authors, potentially as low as just 6% of authors.

With that context, having 22% of Grace users that are women appears to be at the higher end of our expectations of women in the HPC community, which is great news. But women make up 51% of the UK population. So as 22% is below a quarter of the population, sadly that now seems quite low.

Is this a bad thing? Is diversity desirable?

Presumably from the title of this article you know that I at least believe that diversity is indeed desirable. Let me just briefly discuss why.

We know that diversity improves scientific output

The findings from the literature concerning the effect of gender diversity on team performance suggest benefits for team process or ‘team IQ’, are better for mixed teams. A study by Carli found that gender-balanced teams lead to the best outcomes for group process in terms of men and women having equal influence but they must participate at an equivalent rate. Thus having a token woman on a team can be detrimental and indeed can impede team IQ.

By increasing the proportion of women working in HPC we will be having a positive impact on the output of the community, advancing scientific discovery faster than if we do not improve diversity.

Business is also recognising the positive steps that come from increasing the proportion of women including increased turnover, increasing a countries GDP and expanding their marketplace by not excluding female purchasers simply because all of the company’s staff are male. This fits with studies in commercial companies that replicate the findings in scientific innovation; homogenous groups under-perform compared to diverse groups. Industry also has a skills gap problem.

But this does not come without its challenges. Managing diverse teams is generally viewed as more difficult than managing homogenous teams; managers report having to work harder to provide team integration. And we begin to recognise that standard practice such as having meetings in the pub at the end of the day are not great for those that need to get home for caring responsibilities (affecting men as well as women).

But we have another problem as well: If we know that improving diversity is important, what can we do to improve the situation in HPC and increase participation above the 5-22% I’ve just described to you?

A survey of Russell Group Universities in 2014 by the Software Sustainability Institute, found that 92% of the surveyed researchers used software for their research. Of these, 56% reported developing their own software. But when we look at this broken down by gender, 70% of the men who used software also develop their own software but only 30% of the women who use software also develop their own software. This is a problem – if women are not developing software they are potentially less likely to be using what are seen as advanced systems such as HPC.

The survey also highlighted that there was a gender difference when considering the preferred operating system. 26% of the survey’s respondents reported using Linux, but of these Linux users, 88% were male, and only 12% were women. This represents a significant hurdle when the majority of HPC machines use Linux.

We see a similar issue in training. The survey also found that women were less likely to have received training in software development than men: 69% of the male survey respondents had received some form of training but only 39% of women. We see this replicated in HPC. Although 21% of the ARCHER training participants are female, when the courses are broken down into three levels, beginner, intermediate and advanced, we find that women are less likely to attend the advanced courses, with participation dropping from 30% at the introductory level to just 11% on the advanced courses.

Clearly we need to improve the uptake of training if we wish to make a meaningful positive change on the proportion of women using HPC.

We know that one of the biggest challenges we face is improving retention of the workforce with people leaving technology jobs across Europe at alarming rate. But women are even more likely to leave than men. We therefore also need to recognise HPC as a valid career path. The recent recognition that there is a career path for research software engineers is an important step to providing non-traditional career routes. Such non traditional routes are likely to have a direct impact on women. A recent study in the US found that the top reason cited for women leaving technology is working conditions including no advancement, too many hours, or low salary. Therefore recognising career routes such as research software engineers, providing career advancement and recognition of talent is an important step to retaining our workforce, of both genders.

And yes, for those who are already thinking that the problem starts in school, you are right, it does. So we also need to improve the image of science and technology careers to girls. Many of you will know that we have an image problem in STEM. If you do an internet image search for “scientist” you will most likely find that majority of the top images are of white men dressed in lab coats or hovering over a dusty keyboard. There are very few women in these pictures. We need to change this image.

It is therefore important to tell those who might consider a career in STEM what we do. For example, I get paid to play with the fastest supercomputers in the world every day. What could be better than that? Well for some that is not sufficient, conjuring up an image of sitting at a dusty keyboard in a dark room. What we need to be telling the general public is what I would say really makes me get out of bed in the morning, the idea that I might just be helping to make the world a better place. I started out in HPC simulating atom by atom peptides in the body and DNA strands. I have helped design the casing of Nuclear Fusion reactors and worked with teams that design the most effective spacing of wind turbines. The list goes on: weather modelling, climate change, predicting the path of Hurricane Sandy that saved lives, designer medicines, simulating emergency evacuation procedures, modelling earthquakes to identify where to send emergency vehicles first in the event of a natural disaster. This is the message we need to be getting out to schools.

But the problem then returns to recruitment and retention in the workforce which is the current focus of the Women in HPC initiative and what I personally have a great interest in.

One route forward is to ensure we recognise the achievements of all, women as well as men. Unfortunately, we have a long history as a society of not recognising the contribution women have made – we are improving, but it is important that this effort does not falter as this is essential in improving the image of HPC and providing visibility of inspirational role models. It can be as simple as making sure we recognise the importance of Ada Lovelace alongside Charles Babbage.

One such example is Grace – naming this machine after Grace Hopper is an important step to raising the profile of the contribution women make. It is often challenging as universities often like to name buildings after famous staff, alumni or Nobel laureates. So if we don’t have many women in these positions, often because of the failures of society in the past to enable women to hold such positions, then we perpetuate the problem by claiming there are no women who can be put forward for such honours. Sometimes we need to take the brave step of changing our policies in order to reflect the shortcomings of our past.

The visibility of role models whether by naming buildings or machines such as Grace are important. But there is a wealth of research that supports the hypothesis that it is even more important to have a role model that you can directly relate to. So for undergraduates these role models may be PhD students, for postgraduates this is early career and so on. It is not sufficient to simply have famous names on the sides of buildings, as much as this is useful, we need women across the entire spectrum of the HPC community.

We also need to recognise that role models are not just of the same gender. It is therefore important that we allow a wide variety of personalities in our workforce. The first way we can do that is by encouraging everyone to bring their whole selves to work. This is a concept whereby people feel more comfortable with who they are work, and improves the representation of the variety of individuals that work in the community, rather than hiding the diversity by encouraging everyone to conform. This benefits all, not just women, bringing down barriers to enable everyone to belong in the community.

In relation to barriers we also often hold up inspiring women as the first who did X or Y, the hurdles they overcome and the achievements they had because of what they have overcome. While it is incredibly important to recognise the achievements of such people it is important to remember that these will always be an exception. Why should we all have to fight the barriers in our way in order to make our contribution to society? A person’s ability to fight against the odds does not necessarily directly affect their ability to make major scientific discoveries or technological breakthroughs. This applies equally to men and women. By encouraging everyone to bring their whole selves to work we remove the need to conform, remove barriers and facilitate ideas generation and innovation.

I’d like to finish with a brief discussion of four things that we all need to be aware of if we are to have a positive impact on diversity in the HPC community.

First: Recognising unconscious bias
Unconscious bias has become a popular buzz phrase in the last few years for good reason. It is one of the biggest challenges we face if we are to hire and provide career advancement to underrepresented groups. It is not sufficient to simply tell oneself not to be biased, this makes the situation worse! Indeed mentioning unconscious bias should not be an excuse for why the situation is not better. Instead we need to anonymize processes wherever possible, however difficult this may be, and we need to all find the space to think carefully in order to help us address our implicit assumptions.

Second: Consider your citation practices
A study of citation data from the four most prestigious philosophy journals over a ten-year period showed that 4% of citations from the previous decade were authored by women. But women made up at least 15% of professional philosophers during that period. It is therefore worth checking one’s own citation practices – is implicit bias creeping in? Particularly if indicated earlier, are women are less likely to be attending conferences, presenting their research and networking?

Third: We need to challenge conferences to address gender inequality
We know there are women in the community. These women should be represented in the chairs and keynote speakers at conferences. This both provides visible role models and gives women credit for the contribution they make to the community. I would encourage you to consider whether to accept invitations to conferences at which the invited speakers are nearly all white men.

Finally, we need to familiarise ourselves with stereotype threat
This is a psychological phenomenon whereby a person’s performance at a task can deteriorate merely by being made consciously aware that they are from a group that is stereotyped as performing poorly at that task. It does not matter whether or not people from that group really do perform poorly at that task; nor does it matter whether the person in question has evidence of their previous performance at such tasks; when the person is made consciously aware of their membership of this stereotyped group, their performance deteriorates.

Having a single woman or minority group on a panel does not combat stereotype threat, it can instead reinforce it.

We need to recognise this phenomenon, discuss it and help those who experience it (which is the majority of people in any underrepresented or minority group, and even those who aren’t obviously in such a group).

More than anything, the biggest contribution we can make in improving the diversity of the HPC community is to accept that however good it may be that 22% of Grace users are women, there is still work to be done and I ask that all of us take responsibility in addressing diversity in our workforce.