Diving into the EPSRC diversity data looking at gender in grant applications as PI
On June 24th, UKRI released diversity data for the past 5 years. The published data finally shed some light onto various diversity characteristics such as gender, ethnicity, age or disability of funding applicants and recipients by UKRI councils. For the first time, the data addressed not only success rates (now called award rates), but also the amounts of funding awarded to Investigators with varying characteristics. This is an exciting step forward, which allows more in depth analysis than has previously been possible, and we hope that the additional inequities it reveals will spur UKRI into action. The data available covers the five years from 2014/15 to 2018/19. Here we look at data from the EPSRC council with a focus on the gender of the principal investigator (PI).
It is worthwhile to point out that the data published by UKRI has been “harmonized”. Data harmonization is the process of gathering data of varying file formats, disparate sources and naming conventions and molding it into one cohesive data set. The aim is to provide a comparable view of cleaned, sorted and aggregated data from various sources. Here, the harmonization is between the separate Research Councils that make up UKRI, and whilst it is in some ways welcome, it comes with the loss of individual narratives for each Council addressing the specific disparities in their area. Hence, it becomes even more vital for the research community to address council-specific issues.
In this blog post, we’re going to be thinking about the impact of gender on EPSRC applications. Before we do, it’s worth noting that there are three categories in the UKRI gender data: “male”, “female” and “not disclosed”. The data on gender thus fail to address the existence of non-binary genders, and there is also no information on how rates of funding differ between cis- and trans- people. In writing this post, we choose to use the words “men” and “women”, rather than “males” and “females”, and since we acknowledge that trans men are men, and trans women are women, we anticipate that difficulties in grant funding affecting women, affect both trans and cis women. We acknowledge that within the available UKRI data we can’t address issues particular to non-binary or trans people, and we hope that future data sets may allow better consideration of this issue.
As a first step, we take a look at the average grant value awarded to women and men as PIs (Fig. 1) by EPSRC.
We can notice a substantial difference in the amount of grant income women and men receive on average. For example, in 2014–15, the grant value awarded to women PIs was on average £354k whereas a typical proposal led by a man received £779k. It’s also worth noting that when the EPSRC were challenged on the proportion of funding awarded to women following an FOI request for the 2016–2017 data they suggested “the 2016–17 figures were skewed because of a number of very large grants with male principal investigators”. We can see here that in fact the 2016–2017 figures are actually pretty typical of the last 5 years. We can visualise this difference by plotting the average grant value of women PIs as a fraction of those awarded to PIs who are men in the same year (Fig. 2).
Averaged over the last five years, the grants received by women have been only 70% of the size of those received by men. By consequence, women PIs need to win on average nearly 1.5 grants for every grant a male PI wins to receive the same research income. This difference matters for several reasons. Firstly, at least for responsive mode grants, it takes the same amount of effort and time to write a grant proposal whether the grant is comparatively high or low in value. So if a man, let’s say, has to write and win 2 grants per year, a woman would need to write and *win* 3 grants per year to receive equal funding. Given average success rates for EPSRC of around 1 in 3, this means that a woman would have to write 9 proposals, compared to 6 for the man. This is not sustainable.
This problem gets further compounded for researchers who have “multiple disadvantages’’, like for example PIs that are both women and BAME. UKRI states “We have started work on other strands such as intersectionality”, but currently there is no data available addressing these intersectional impacts. Furthermore, it is already known that COVID19 has had a disproportionate impact on the BAME community and women which is why we are anxious to see this data coming out for 2020/2021.
But *why* do women get on average smaller grants than men? Do women ask for less? Are grant values reduced by the funder? The latter is likely not the case, because EPSRC does not generally reduce the amount of the grant that was requested by the applicant. Occasionally, when a budget does not appear to be justified during the review process, the proposal might be sent back to the applicant with a request to revise the budget, but this is rather rare. Consequently, the awarded grant values should match the grant value that was requested by the applicant pretty well .
It is therefore possible that women ask for less in their grant applications. Let’s have a look how much women have asked for in their grants as PI and normalize this amount by the average amount requested by men in that year (Fig. 3).
Here we can see again a substantial difference. For example, in 2014/2015, on average women requested grants that amounted only to 48% of the value on average requested by men. The absolute values are presented in Fig 4:
Here, in 2014/15, the average grant application by women was for £381k compared to an average grant value of £793k by men. We can also note that the similarities between Figure 2 and Figure 3 suggest that the difference in the size of grants men and women take home reflects the differences in what they apply for. It seems that women ask for less money in their grant applications as PI, as can be seen when averaging over the 5 years of data available to us (Fig. 5):
Let’s also bear in mind that the displayed distributions appear symmetric because we have averaged the 5 yearly averages available from the UKRI data set which means that we have lost all information about the distributions in individual years which could well be more skewed. In fact, they most definitely *are* more skewed because there are sometimes huge discrepancies between the median and the mean (Fig. 6):
For example, in 2017/18 there is a £411k discrepancy (Fig. 6 and Fig. 7) between the median and the mean which means the available data points towards a very skewed distribution. In other words, one (or a few) very large grants compensate for many more grants of much smaller value. Inversely, a mean value which sits fairly close to the median value indicates a more symmetric distribution and that not many high-value grants have been won by women PIs. The range of money awarded would help contextualise where the means and medians fall. Frustratingly, we are not being given access to the range of the data, the mode of the distribution or, better yet, every individual data point!
Fig. 7 shows that for both men and women the mean is always above the median (pointing towards a positive skewed distribution and highlights the year 2017/2018 where at least one big grant was won by a woman not only nearly doubling the average difference with respect to other years for women but also even surpassing the difference mean-median for men.
Let’s have a look at the proportion of EPSRC principle investigator awardees (Fig. 8) and applicants (Fig. 9) by gender. The number of women PIs was lower last year (75) than in 2014/15 (80) even though the number of applications by women went up from 260 to 300. On average, the application (14% ± 1%, mean ± 1 standard deviation, n=5) and the awardee proportion (14% ± 2%, mean ± 1 standard deviation, n=5) are below the share of women working in the academic population of EPSRC. According to HESA, the Higher Education Statistics Agency (2017/18), approximately 18% of the engineering and physical sciences (EPS) academic community are women. Hence, these data suggest that not only are women applying for smaller grants, they are applying for less grants than would be expected.
The big question is why is this the case? The risk here is that women will essentially be blamed for the low level of funding they receive, suggesting they “choose” to only apply for lower amounts. We suggest that a spotlight actually needs to be shone on the factors which limit their choices. TIGERS have produced a review, highlighting the barriers experienced by women in the grants application process, a preprint of which is available online. It highlights the impact of excessive administrative and teaching loads, caring responsibilities and institutional gatekeeping on women’s opportunities to apply for funding. Some of these barriers may be exacerbated by funder policies concerning for example demand management and assessment protocols. Overall, we must reform the system which limits women’s opportunities, rather than blaming women for lack of progress on gender equity.
Finally, let’s have a look at the power of representation: let’s see how much of the science actually funded by EPSRC is dreamed up and investigated by women. For this we can simply multiply the number of female awardees with the average value of their awards in this particular year and compare this to the corresponding value for men (Fig. 10):
Here, frankly the numbers are quite grim: in the past 5 years, EPSRC has paid a total of £2.2 billion to drive research led by men versus funding only £260 million worth of women-lead research. This paints a grim picture about scientific independence and the ownership of ideas of women working in scientific research under the EPSRC remit.
As a final comment, it’s worth noting that these analyses have been done by grant number, where all grants contribute equally to the analysis representing a single application regardless of scale or duration. As we were finalising this blog post, EPSRC brought out a report in which they evaluate the gender dimension by unique identifier (where individuals who have made multiple applications/ hold multiple grants will only contribute once to diversity statistics) and by value ( where the financial value of the grant is stratified and serves as a proxy for scale and length of the award). We’ll comment more on the light that report sheds on the data in future posts.
In summary, here we have found that:
- Women PIs request on average only 69.9% of the grant value of their male counterparts (average of £514.2k vs £747.8k) with a range of requesting only 48.0% to 88.6% of what men have requested. However, this statistic is really not as straightforward as it appears and we need to find out *why* this is the case. Is our research culture discouraging women from asking for more money? Are the differences due to requested salaries and reflect the known gender pay gap? In future analyses, it would be interesting to take grant duration into account and compare average values for money/month between women and men. Are grants awarded to menPIs higher value because their projects are longer? What other extra costs do men ask for? Maybe it is more highly qualified PDRAs, or more generous consumables and travel? What effects would these potential differences have on women’s scientific work? A future blog post will delve into the issue of gender disparity in requested grant costs. It is also worth pointing out that , while the success rate appears similar between men and women when looking at applications by numbers, this is not necessarily the case when looking at applications by value. A future blog post will analyse how award rates differ for women men PI when we look at grants by value.
- A smaller fraction of women are awarded grants than what is expected from looking at the percentage of women working in research & teaching employment falling under the EPSRC remit. Women PIs are under-represented in applications across the EPSRC portfolio.
- The power of representation is heavily skewed towards men by a factor of 8.5 (£2.2 billion for male-led research vs. £260 million for female-led research).
Let’s not forget that there is a lot missing from these data, and more information is needed to allow us to understand them better. How does intersectionality affect these statistics, e.g. for women of colour, transgender women or women with disabilities? We currently cannot answer these questions with the data provided and hope to be shedding some light how intersectionality affects funding in the future.