For gender pay equality, universities must let social scientists see the data

The average gender pay gap in UK higher education stood at 18.3 per cent in 2021. This is roughly three percentage points higher than in 2017, when compulsory reporting was introduced, and is also higher than the 14.9 per cent average for all UK industries and services combined. We must do better.

Historically, universities have been at the forefront of investigating gender pay gaps and advocating for change. In pioneering work in the 1970s, Elizabeth Scott, a University of California, Berkeley statistics professor, convinced her employer to part with granular payroll data. Having painstakingly cleaned it, Scott used regression to model the relationship between pay, gender and a host of other variables to understand the nature and causes of the gender pay gap.

Today, gender differences in pay and employment command significant attention in economics, sociology, gender studies, anthropology and elsewhere. Between them, these disciplines offer the statistical tools to identify the drivers of said gaps (such as the contributions of qualification levels versus outright discrimination) and the qualitative tools to understand more nuanced root causes (such as what explains women’s supposed reluctance to ask for pay rises).

Academics have also pioneered research on inequalities far beyond work and employment: racial inequalities in healthcare, for instance, and gender inequalities in access to credit. You might think that universities have a vested interest in promoting similar work regarding academic pay, particularly in an era of rising awareness of pay gaps and inequalities. Yet Elizabeth Scott’s work seems, if anything, even more pioneering today than it did half a century ago.

To put it simply, universities are loath to part with the granular payroll data that would enable their own faculty to research pay gaps. Part of this reticence is understandable and stems from the universal sensitivity of payroll information. But the more significant factor is the way modern universities are managed. With small exceptions (such as in education departments), faculty research expertise is directed firmly outwards, while potentially sensitive internal questions are handled by consultants and managers. Handled – but not researched.

In my own university, for instance, a non-academic “people data” team handles queries on payroll data. The team’s existence is part of a thrust to broaden data analytics across the university, and it has made several types of data more accessible, such as summary statistics on pay and student achievement by ethnicity and gender. However, access to the underlying granular data is proscribed. This prevents actual research, and leaves pay gaps to be measured and reported on more in terms of legal compliance, marketing and other outward-facing facets of the institution.

Granular data are necessary to answer two complementary types of questions. The first is quantitative and relates to the combinations of age, gender, ethnicity and discipline that correlate with pay and promotion outcomes. These can only be modelled using multivariate regression techniques. Examples include identifying recurrent patterns of individual work histories that lead to hesitancy in applying for promotion, or recognising the profiles most likely to make it to interview, get the job and be offered a higher salary. Machine-learning techniques can go even further, incorporating dozens of variables in their analyses, such as the timing and length of career breaks.

The second type of question is qualitative and relates to why and how such combinations determine those outcomes. Understanding those factors also supports the crucial next step: determining what actions could improve outcomes and building realistic expectations about change. For instance, do senior mentors need to be purposively paired with certain profiles of junior faculty and asked to hold annual conversations with them about promotion? Answering these questions requires interviews, but, to be fruitful, the interviews must be informed by a thorough understanding of the statistical correlations.

Privacy concerns about payroll information are genuine, particularly around institutions’ legal obligations in this area, but there are ways to handle this. Researchers who wish to handle and analyse the data can be required to sign non-disclosure agreements, and those who want to conduct interviews on pay and employment can be asked to seek ethical clearance. External researchers can also be hired to ameliorate the complications of interviewing one’s own colleagues.

Second, a halfway solution would be to set up statistical software (such as R) to let researchers specify and estimate models without having access to the underlying data. Third, data simulation techniques can be used to create pseudo data that support anonymity but preserve the correlation structures of the original data and thereby enable statistical analysis.

It is true that debates on how payroll data are handled touch upon larger conversations about managerialism in the academy. But unlike some of the issues that managerialism throws up, understanding and tackling pay gaps is within our reach. The spotlight might be uncomfortable, but allowing academics who research these things to study their own institutions would yield substantial insights and greater equality all round.

Sunil Mitra Kumar is a senior lecturer in economics at the King’s India Institute and the department of international development, King’s College London.