Limited access Past event
Department of Economics Seminar with Professor Rachel Griffith, University of Manchester Gender differences in pay amongst high educated workers: evidence from academia
Long abstract: The gender pay gap has fallen in many countries over the past half a century. In the UK this has largely been due to increased educational attainment by women, who are now on average more educated than men, and increased minimum wages, which has increased the wages of less educated women. However, amongst high educated workers the gender pay gap has fallen more slowly and remains large, particularly in occupations where there are high returns to longer and less flexible hours of work. For example, in 2025 the gender pay gap in the UK was 15.5% at the 90th percentile of the earnings distribution, compared with less than 1.8% at the 10th percentile.
We study gender differences in the determinants of pay for highly educated males and females in UK academia. Institutional features of the academic setting make it a convenient place to study the gender pay gap among highly educated workers. Academic researchers have invested heavily in their human capital. There are a small number of employers (universities) who offer broadly similar benefits and non-wage amenities both to each other, and to academics working in different disciplines. Staff perform similar jobs, yet they do so in discipline-specific labour market that vary in along a number of dimensions that are relevant for understanding the gender pay gap. These include variation in publishing norms and standards, returns to outputs, outside options and the ability to generate surplus for the university. The gender pay gap varies significantly across disciplines, from a high of over 17% in Economics to less than 3% in Sociology.
We bring together new administrative data on pay and employment contracts, with data on individual publication histories and other outputs. Our data has several major strengths. First, it covers the universe of UK-based academics across all disciplines. Second, it is well measured. The salary data is administrative data collected by a government agency. In these data we are able to accurately identify research active academics from those who are teaching focussed. This is important, as women are more highly represented in teaching-focused jobs, which have different expectations and salary structures. These data also allow us to accurately identify each individual's discipline (using information from the REF submissions). This can be difficult in some data, and it allows like-for-like comparison of academics across universities.
We collect comprehensive data on the publication histories of the same population of individuals. This allows us to construct standard measures of individual output such as their H-index, and publication record in prestige journals (such as the top 5 in economics). We also have information on other measures of productivity and output, such as obtaining grant funding, engaging in high impact activities, working as a dean, and others.
We are interested in learning about the extent to which differences in productivity and experience explain pay differences between women and men, and how that differs across disciplines. We do this by estimating the elasticities of these characteristics with respect to salary. Differences in salaries are driven by differences between women and men in characteristics (for example age, H-index, publications), and differences in the returns to these characteristics by gender.
Consider a standard Mincer earnings equation, where salary Y depends on the individual's human capital and other characteristics X. We have individual-level data on Y and X separately for the population of research active academics. We also observe some common characteristics in both data (age, gender, discipline, institution). We cannot directly link these data at the individual level. We can directly estimate the gender gap in salary, and separately the gender gap in outputs. Estimating the elasticities is more challenging. Our main approach is to impute output measures into the salary data based on the common characteristics. We also calculate bounds on the elasticities using the approach proposed in D'Haultfoeuille, Gaillac and Maurel (2024, REStud). As well as considering the returns at the mean we consider how these vary over quantiles of the salary distribution.
Preliminary results suggest that differences in observable characteristics account for over two-thirds of the gender pay gap, with differences in age accounting for one-third (women retire or exit research younger), gender differences in publications around 20%, and differences in returns the remaining third. Economics stands out with gender differences in publications account for close to half of the gender pay gap.