The Leaky Pipeline

When I was first asked to take part in a discussion about women in physics, sometime in the late 1970’s when I was a graduate student, it was generally thought that the dearth of women in science was due to women not studying these subjects at school and university. This was called a pipeline problem: women were not becoming scientists because there were none in the pipeline. By the 1990’s, people noticed that while the numbers of women studying science were increasing the numbers of women becoming scientists were not increasing proportionately. In particular, the number of academic scientists at any particular stage was less than would have been expected from the number of undergraduates the appropriate time earlier, see for example figure 2.5 of the ETAN Report Science Policies in the European Union: Promoting Excellence through Mainstreaming Gender Equality, European Commission, 2000, http://cordis.europa.eu/improving/women/documents.htm. Another study in Italy looked at researchers at the Italian National research Council who all started in the same year and found that after ten years 26% of men and 12.8% of women had been promoted to research director. Staying Competitive: Patching America’s Leaky Pipeline in the Sciences, Marc Goulden, Karie Frasch and Mary Ann Mason, 2009 (http://www.americanprogress.org/issues/2009/11/pdf/women_and_sciences.pdf) documents the preferential loss of women in academic science between receiving a Ph.D. and achieving tenure. This preferential loss of women scientists is called the leaky pipeline since women started out in the pipeline but were subsequently lost. Confusingly, the phrase is sometimes also used to describe any circumstances in which there are fewer women at higher levels in an organization compared to lower levels regardless of the reasons for the situation.

The American Institute of Physics published a report in 2005, Women in Physics and Astronomy, 2005, http://www.aip.org/statistics/trends/reports/women05.pdf, which concluded that, in Physics and Astronomy in the US, ‘women are represented at about the levels we would expect based on degree production in the past.’ This may reflect the fact that women needed to be exceptionally committed to go into physics in the first place, particularly in the period 1967-1980, when the average percentage of Ph.D. awarded to women by US Physics Departments was 4%. (How accurately would you be able to measure a decline from 4%?)

So, is there data demonstrating the ‘leaky pipeline’ in the sense of demonstrating the preferential loss of women from academic science? Yes, there is. Do data of the form women are 47.9% of lecturers but 18.7% of professors (HESA Press Release 131, http://www.hesa.ac.uk/index.php/content/view/1397/161/, figures are for 2007/08) demonstrate a leaky pipeline? Not by themselves: together with the information that in 1970/71 42% of all full-time undergraduates were women, and, more relevantly for potential professors, 23% of all full-time postgraduates students and by 1980/81 these figures were 41% and 34%, respectively [Source: Social Trends 2009, http://www.statistics.gov.uk/downloads/theme_social/Social_Trends39/ST39_Ch03.pdf], they do, though perhaps not so dramatically as the contrast between 47.9% and 18.7% suggests.

Dramatized Data

When I was in the sixth form, rather a long time ago, my English teacher asked the class if anyone had read Rachel Carson's Silent Spring. I turned out to be the only person who had so the teacher asked me what I thought of it. I said that I thought the data were convincing. Her response was that she had been impressed by the power of the writing. I wonder if this reflects a widespread difference in how people react to data. Are there those who ask what do these data tell me and those who ask how do I react to these data? One group is puzzled by being presented with data that seem incomplete, irrelevant or lacking sufficient context to draw conclusions and the other can’t understand why people are pedantically pointing out the limitations of the presented data when to them the statistics are obviously appalling and clearly demonstrate that something should be done.

An example of data presented to elicit an emotional response rather than to diagnose a problem is the statement that 47% of chemistry graduates are women but only 6% of chemistry professors. What does this tell us? We would have expected the percentage of women among professors to match the percentage of women among new graduates only if the system was in a steady state and had been for around forty years. This is clearly not the case. Current professors are largely drawn from a pool of people who were undergraduates 25-45 years ago, i.e. from about 1965 to 1985. With no further information we cannot tell whether these figures show that few women studied chemistry 25-45 years ago or whether the ones that did were much less likely to become professors of chemistry than their male counterparts. In fact this statement is not even very successful at eliciting an emotional response since it is unclear whether we should rejoice that the proportion of women among those studying chemistry is much higher than it used to be or bemoan that the low proportion of women among chemistry professors indicates that many women in chemistry are not fulfilling their potential.

Another example is the institutional gender pay gap. In February 2009 the University of Cambridge published an Equal Pay Review [http://www.admin.cam.ac.uk/reporter/2008-09/weekly/6141/4.html] which noted that the % Difference between average pay for women and average pay for men was 32%. Sounds terrible, doesn’t it? Especially when the national gender pay gap (calculated using mean hourly pay, excluding overtime) is 16% [http://www.statistics.gov.uk/cci/nugget.asp?id=167]. The first thing to notice is that these percentages are not calculated on the same basis. The Cambridge figure tells us the mean pay for men is 32% greater than the mean pay for women while the Office of National Statistics figure tells us that the mean pay for women is 16% less than the mean pay for men. Comparing on the same basis gives either 24% (Cambridge) and 16% (national) using the mean pay for men as a reference or 32% (Cambridge) and 20% (national) using the mean pay for women as a reference. If it were the case that women were paid 24% less than men for doing the same, or equivalent, work, this would, indeed, be appalling. But this is not the case. The analysis by grade shows minimal differences with gender. The information content of the Cambridge gender pay gap is that the University pays its 1500 or so, predominantly male, academic staff rather more than it pays its 1500 or so, predominantly female, clerical and secretarial staff. Even if the academic staff were 50:50 male and female this still would not remove the pay gap because of the very high percentage of women among lower paid support staff. Redressing this imbalance will require an intelligent analysis of the issues and a long-term, sustainable strategy, not knee-jerk reactions to statistics.

The common feature of both these examples is that they are both attempts to summarize a complex, multi-factor situation in one or two numbers. Perhaps this is necessary in some contexts but these ‘sound-bite statistics’ should be backed up by proper data and analysis.

Does it matter? After all, people do not generally take action based on a rational analysis of data. If quoting ‘dramatized data’ inspires action then surely that is a good thing. This approach is not without its dangers. One is that the response to ‘something must be done’ is ‘we must be seen to be doing something’ whether it is relevant and effective or not. The other is that those whose response to data is to ask ‘what does this tell me’ will conclude that the answer is nothing and that there is, therefore, no need to take any action. They may also conclude that if people are quoting unconvincing data it is because there are no convincing data. Which would be a pity because when you look at the percentage of women among academic staff in a school and discover that it has barely changed in a decade you can’t help thinking that there genuinely is a problem.

The Absence of Data

This is the first of a series of posts on the use of data in the area of women in science. I spent thirty years working as a physicist. My first reaction to any problem is to get a feel for what it means quantitatively. For example, how many women are there studying SET subjects, how many researchers, how many lecturers, how many professors? The next question is: what can be deduced from the data? Can we identify the critical points in a career in academic science where we are most likely to lose women? What steps should we take now? What further data would help us identify issues and further steps to improve the situation? The remarks in this and the next few posts are not intended to suggest that there is no problem. I spent six years working towards the improved representation of women in SET. I would hardly have done this if I did not believe there are genuine constraints on women’s participation. My motivation for the next few posts is a strong personal belief that progress in this area depends on properly interpreting relevant and meaningful data and information.

My first comment is about the absence of data. When I started working in the area of women in science, as opposed to being a women in science, I was struck by the number of times a Head of Department would tell me that there were x% women in his department and when I checked the actual numbers I would find it was (x-10)%. It seems to me that if you are genuinely interested in improving the numbers of women in science the very first thing you would want to know is how many you actually had in each grade. Presumably in their scientific work they don't just guess when the actual figures are easily obtainable?

University Gender Pay Gaps and Women in SET

I have sometimes heard the view expressed that addressing institutional gender pay gaps in universities is important for women in SET. I believe that, in fact, focusing on an institutional gender pay gap is at best peripheral and at worst actually detrimental to women in SET in universities.

A gender pay gap is generally calculated by finding the mean pay for women and the mean pay for men and calculating

( (mean pay for men)/(mean pay for women) – 1 )

or

( 1 – (mean pay for women)/(mean pay for men) )

assuming that the mean pay for women is lower. The former is larger and consequently sounds more impressive: the latter is the definition used by the Office of National Statistics in the UK (http://www.statistics.gov.uk/cci/nugget.asp?id=167). Sometimes the median is used rather than the mean. The median tends to be more typical since it is the point where half the population earn more and half less. Mean pay can be sensitive to a few individuals who earn much more than everybody else.

Mathematically, both the median and the mean depend on the relative frequency, which is the proportion of the population earning a particular salary. If analysis shows that women and men are paid the same for the same or equivalent jobs, then the gender pay gap is determined by the difference in the way men and women are distributed across grades. A gender pay gap results if women are more likely than men to be in lower paid positions and less likely than men to be in higher paid positions. The gender pay gap is independent of the overall proportion of women in the institution. An institution with a small overall proportion of women who are fairly evenly distributed between grades would have a small gender pay gap, for example, if every grade had exactly 10% women the gender pay gap would be zero.

The main contributing factors to the gender pay gaps in universities are

1. The large proportion of women among clerical and secretarial staff, who tend to be lower paid.
2. The small proportion of women among senior academic staff, for example, HESA figures show that in 2007-08, 19% of professors were women while 48% of lecturers were women (http://www.hesa.ac.uk/index.php?option=com_content&task=view&id=1397&Itemid=161).

Strategies to reduce the gender pay gap could include:

1. Increasing the percentage of women holding professorships, which should be good for women in SET.
2. Decreasing the percentage of women among clerical and secretarial staff, which would be irrelevant to women in SET. Unfortunately technicians have similar pay rates so we would also want to decrease the percentage of women among technical staff, which would be bad for women in SET.
3. Decreasing the percentage of women among research assistants and post-docs or decrease the total number of post-docs, which might be a good thing for women in SET or might not.

A particular concern is that parasite institutions who rely on poaching senior women from other institutions would be able to reduce their gender pay gaps at the expense of nurturing institutions who work to develop the careers of their staff.

It may also be that not all women want to progress through grades at the same rate as men. Some may prefer to postpone applying for promotion to more responsible and stressful jobs until their children are older. Too much focus on gender pay gaps could make life harder for women who are trying to balance career and family.

A further problem with any measure that reflects occupational and vertical segregation is that it is bound to change only slowly. This is partly because you can’t fire incumbent staff to make way for new staff of the other gender and partly because aspects of the distribution of women, for example, their concentration in clerical and secretarial grades, are the results of societal and cultural factors outside the control of the institution.

We do need some measure of how women are distributed within institutions. Claiming that nearly 50% of your work force are women does not mean much if they are predominantly in low paid and/or insecure positions. Perhaps rather than rely on a single ‘figure of merit’ (or demerit) such as the gender pay gap it would be better to focus on particular areas of concern. For example, if it is felt that the proportion of women among professors is an issue then measure the proportion of women among professors. Similarly, if the career progression of post-docs is an issue then monitor the destinations