More education accounts for the rising share of women in the UK top 1%
Richard V. Burkhauser, Nicolas Hérault, Stephen P. Jenkins, Roger Wilkins21 July 2020
Top incomes – the share of total income held by those at the very top of the income distribution – are the central focus of much recent income distribution analysis, most especially that based on the pioneering research of Thomas Piketty and collaborators (see e.g. Atkinson et al. 2011 for a review) and the on-going work centred around the World Inequality Database project. Drawing on income tax and related administrative register data sources, the top incomes research field has highlighted how, in many countries, the most significant changes in income distribution are occurring at the top. The increasing share of income among this group is driving the much-discussed increase in overall inequality.
However, in the ‘recent research on top incomes, there has been little discussion of gender’ (Atkinson et al. 2018: 225). Learning more about the gender divide at the top of the income distribution is important not only because it contributes to our knowledge about trends in vertical inequality but also because differences between the sexes are a prominent horizontal inequality and hence interesting in their own right.
We build on recent research documenting a rising share of women in the top 1% of the income distribution of several countries (e.g. Atkinson et al. 2018, Ravaska 2018, Brewer and Sámano-Robles 2019, Joyce et al. 2019, Bobilev et al. 2020, Boschini et al. 2020, Yavorsky et al. 2020), and, focusing on the UK, analyse the factors that account for this significant shift using regression-based decompositions.
Earlier research has not analysed trends in this manner, focusing instead on changes over time in the types of income (labour, capital) that are held by men and women. Our approach looks at more fundamental factors such as differences in education between men and women and how differentials have changed over time.
Our research is also distinctive because we use household survey data – we have evidence that the oft-cited issue of survey under-coverage of top incomes does not prejudice our analysis (Burkhauser et al. 2020). Hence, we can exploit the much greater information about personal characteristics that is available in survey data compared to most of the administrative record data sources that top-incomes researchers use. We use the ‘households below average income’ sub-files of the Family Resources Survey. ‘Income’ is gross taxable (pre-tax pre-transfer) income distributed among individuals aged 15 or older.
Figure 1 shows that the share of women in the top 1% in the ‘households below average income’ sub-files closely tracks the corresponding share in the Survey of Personal Incomes (the UK’s main administrative data source about incomes). This is true both in terms of levels and trends. The clear trend is of a rising female share of the top 1% between financial years 1995/96 and 2015/16. Nonetheless, in 2015/16, the female share of the top 1% was still only 19%.
Figure 1 Share of women in the top 1% gross income group in UK survey and tax return data (1995/96 to 2015/16)
Notes: Estimates are based on the population of individuals aged 15 or above. HBAI: ‘households below average income’ sub-files of the Family Resources Survey; SPI: Survey of Personal Incomes.
We use an Oaxaca-Blinder style decomposition approach extended to non-linear regressions to account for the rise in the probability of being in the top 1% for women and the corresponding fall for men. We are looking to see the extent to which changes over time in the probability of top-1% membership are due to differential trends in the distributions of men’s and women’s characteristics (their education, employment status, where they live, etc.) or due to differentials trends in how much men’s and women’s characteristics translate into higher probabilities of top-income-group membership (i.e. changes in the ‘returns’ to characteristics).
Figures 2 and 3 summarise our main findings. Figure 2 shows the estimated components of our regression-based decomposition identifying the components related to changes in characteristics and to changes in returns to those characteristics. The characteristics we used are listed in the note to Figure 3.1 We show estimates for women in panel (a) and for men in panel (b).
Figure 2 Change (ppt) in the probability of being in the top 1% between 1999 and 2015, decomposed into ‘changes in characteristics’ and ‘changes in returns’ components, by sex
Notes: Estimates from an Oaxaca-Blinder decomposition extended to a logit model by Fairlie (2005). Based on 3-year pooled samples: ‘1999’ refers to 1998/99–2000/01 and ‘2015’ to 2014/15–2016/17. The decomposition formulae and regression estimates underlying the decomposition are reported by Burkhauser et al. (2020). The characteristics used are listed in the note to Figure 3.
For women, the probability of being in the top 1% increased from 0.27% to 0.43% between 1999 and 2015. Of this 0.16 percentage point increase, virtually all of it – 0.15 percentage points – is accounted for by changes in women’s observable characteristics, which leaves a contribution of 0.01 (0.16 – 0.15) percentage points attributable to changes in estimated returns to characteristics.
For men, the probability of being in the top 1% declined by 0.111 percentage points from 1.96% to 1.85% between 1999 and 2015. Changes in observable characteristics contributed to increasing this probability by 0.353 percentage points. In other words, given the changes in their characteristics, men should have seen a large increase in their probability of being in the top 1%. Instead, there was a decline because these changes were more than offset by the reduced returns to their characteristics (0.638 percentage points). Clearly, the factors accounting for the trend in top income group membership for men differ from those for women.
One feature of the trend is common to both men and women, however. Figure 3 shows that, for both sexes, changes in the distribution of education account for by far the largest proportion of the increase in top-1%-group membership that is attributed to changes in observable characteristics as a whole. For women, it explains more than two-thirds of the increase (0.111/0.150). For men, it explains nearly all the increase (0.422/0.527).
For women, only the estimate of the education variables’ contribution to the total characteristics component differs significantly from zero (estimate around four times larger than standard error). For men as well, the education variables’ contribution is statistically significant (ratio of estimate to SE of at least 7.7) but so too is the contribution of the partner variables (ratio over 5).
Figure 3 Characteristics contributing to the ‘changes in characteristics’ component, by sex
Notes: Demographics includes age and age squared and a dummy for non-white; Family is family type (6 categories depending on whether the head is of pension age, number of adults [one or two], and whether children present); Education is age completed full-time education (and its square); Region is the region of residence (London, South East, rest of the UK); Employment includes employment status (employee, self-employed, NILF or unemployed), whether working part-time, job occupation (5 categories) and job industry (9 categories); Partner includes binary indicators for whether respondent’s partner belongs to top 10% or to top 1% income group.
Although the return to staying longer in education in terms of chances to get into the top 1% is much larger for men than for women in both 1999 and 2015, this return hardly changed over time for men whereas it increased for women more noticeably. We show (Burkhauser et al. 2020) that the return to an extra year of education increased from 0.057 to 0.072 percentage points if evaluated using 1999 sample characteristics. In contrast, we find that the return for men of an additional year of full-time education increased from 0.342 percentage points to 0.373 percentage points if evaluated using 1999 sample characteristics.
The large negative ‘change in returns’ component in the decomposition for men that is shown in Figure 2 reflects several factors. For instance, for men, the penalties – lower chances of top-1%-group membership – grew for non-white people, for individuals not in paid work, and for individuals living in the rest of the UK outside London. The same penalties exist for women and also increased but, in each case, the penalty for women is much smaller than the corresponding one for men in absolute magnitude, and so too is the contribution to the ‘changes in returns’ component.
In summary, the rising share of women in the top 1% of the UK income distribution is largely accounted for by women having increased the time they spend in full-time education by more than men did. A minor supporting role is played by an increase for women in the return (in terms of securing top-income-group membership) from having longer education that is larger than the increase for men.
Atkinson, A B, A Casarico and S Voitchovsky (2018), “Top incomes and the gender divide”, Journal of Economic Inequality 16(2): 225–56.
Atkinson, A B, T Piketty and E Saez (2011), “Top incomes in the long run of history”, Journal of Economic Literature 49: 3–71.
Bobilev, R, A Boschini and J Roine (2020), “Women at the top of the income distribution: What can we learn from LIS-data?”, Italian Economic Journal 6: 63–107.
Boschini, A, K Gunnarsson and J Roine (2020), “Women in top incomes: Evidence from Sweden 1974–2013”, Journal of Public Economics, online ahead of print.
Brewer, M, and C Sámano-Robles (2019), “Top incomes in the UK: Analysis of the 2015-16 Survey of Personal Incomes”, ISER Working Paper 2019-06, Colchester: ISER, University of Essex.