Employment polarisation – i.e. the increase in employment shares both at the bottom and at the top of the skill distribution, combined with a decline in the middle – has been extensively documented for the US economy in the last 30 years and has become a well-known stylised fact (Autor et al. 2006, Acemoglu and Autor 2011, Autor and Dorn 2013). Less well known in the literature are the characteristics of employment polarisation when distinguishing by gender. In a recent paper we show that women contribute to a large extent to the phenomenon of employment polarisation in the US, while the role of men is more contained (Cerina et al. 2018).
Changes in employment shares of men and women
The black line in Figure 1 reports the pattern of employment polarisation in the US, which displays the typical U-shaped pattern documented in Acemoglu and Autor (2011) and Autor and Dorn (2013).1 Occupations at the bottom and the top of the skill distribution experience an increase in their employment shares, while shares of occupations in the middle shrink. By retaining the same construction of percentiles, we decompose the aggregate change in the employment share of each percentile that is due to a change in female hours and in male hours. The red line in the same figure reports the behaviour of women, which displays a well-defined U-shape. Instead, the changes in the shares of men, while also displaying a U-shape, are more homogeneous along the whole distribution.
Figure 1 Changes in employment shares in the US between 1980 and 2008 by skill percentile, decomposition between men and women
In Figure 1, the vertical sum of the coloured lines gives the black line because we decompose, for each percentile, the contribution of each gender to the change in the employment share of that same percentile. Thus, Figure 1 embeds both the fact that changes in hours worked along the skill distribution are different for men and women, and the fact that, on average, women increase their hours in the market relative to men. For this reason, in Figure 2 we compute male and female changes in employment shares by percentiles within each gender, i.e. using gender-specific total hours worked instead of aggregate total hours to compute shares.
Figure 2 Changes in employment shares in the US between 1980 and 2008 by skill percentile normalised for gender specific total working hours
Thus, while the vertical sum of the coloured lines in Figure 2 would not give the black aggregate line appearing in Figure 1, this treatment of the data provides a measure of the heterogeneity of changes in employment shares for each gender, which is independent (in an accounting perspective) from the fact that, on average, female employment shares increase relative to male employment shares. Within gender the differences of changes in employment shares along the skill distribution are even more striking than those appearing in Figure 1.
Table 1 Change in employment shares by groups of occupation and gender
In Table 1 we also report the changes in employment shares by gender according to Census broad groups of occupations, classified by their average wage in 1980. As widely documented, occupations in the middle of the skill distribution shrink in terms of total hours worked, while those at the extreme of the distribution expand. However, Table 1 also documents the remarkable differences between men and women in the dynamics of the employment shares among occupational groups. Such differences can be only partly captured by a general level effect, implying that women increase their total employment share by 6.7% (with a corresponding decrease for men), as the changes in the employment shares are highly asymmetric along the skill distribution. In particular, women more than double their employment share in occupations at the upper tail of the distribution (from 8.2% to 17.3%), while the male share increases by less than 20% (from 15.8% to 18.7%). On the other extreme, the only other group of occupations where women increase their employment share is within the service sector, where their employment share grows by almost 30% (from 5.4% to almost 7%) compared to only 13% for males (from 4.9% to 5.5%). These service occupations are highly concentrated in sectors producing services that are highly substitutable to household production (especially childcare workers, gardeners, cleaners, home health aides). Some of these jobs (e.g. food service workers, security guards, janitors) also support the jobs of high-skilled workers and, therefore, are complementary to the highest paid occupations.
A gender theory of employment polarisation
In Cerina et al. (2018) we show that the differential patterns of employment shares of men and women can be accounted for by a model of skill-biased technological change (SBTC) in which educated women initially devote a high fraction of their time to home production. In fact, the Census data document that in 1980, educated women spent a high fraction of their working time at home (51%) relative to men (17%). By fostering an increase in the labour market hours of skilled women, SBTC accounts for most of the increase of employment shares at the top of the skill distribution. This increase indirectly generates additional demand for low skilled labour (and therefore an increase in the lower tail of the distribution) through two different channels. First, the reduction in home production generates the need for the household to replace home services with some substitutes provided in the market. Second, the increase of high skilled labour, by production complementarity, generates an additional demand of low skilled labour within the firm, needed to support the productivity of the former (as also argued by Eeckhout et al. 2014). As the changes in employment shares at the top and the bottom of the skill distribution are positive, the changes of employment shares in the middle turn out to be negative. Through these mechanisms, SBTC is then able to explain both the increase in the upper and lower tails of the skill distribution. Compared to women, men in 1980 allocated most of their time to the labour market, so the emergence of SBTC did not affect the home/market labour choice to a large extent.
The importance of the SBTC channel in explaining employment polarisation during the period 1980-2008 is also emphasised by two out-of-sample counterfactual exercises performed in our paper. First, we test the predictions of the model running backward from 1980 to 1960. Figure 3 reports employment polarisation in the data and in the model for the 1960-1980 period. The model can account for the absence of any polarisation pattern both at the aggregate and at the gender-specific level. Since the only relevant difference between the 1960-80 and the 1980-2008 periods is due to the absence of SBTC, this exercise confirms that the latter is a main driver of employment polarisation.
Figure 3 Employment polarisation during the period 1960-1980
Second, we test the ability of the model to replicate the patterns of employment polarisation by decade (1980-90, 1990-2000, 2000-2008). As initially documented in Acemoglu and Autor (2011), the right panel of Figure 4 shows that the polarisation graph displays a clockwise tilting over time, with the increase at the top of the skill distribution determined mainly in the 1980-2000 period, and the increase at the bottom being a feature of the 2000-2008 period. When we feed the model with decade-specific exogenous trends as measured in the data, this tilting behaviour is captured well (top panel of Figure 4). However, from counterfactual experiments for each decade, we are able to show that such tilting behaviour completely disappears once SBTC is removed. Again, this exercise reinforces the idea that SBTC is a first-order driver of employment polarisation.
Figure 4 Employment polarisation by decade, 1980-2008.
Our results suggest that employment polarisation in the US is largely generated by a differing gender-specific pattern of employment shares along the skill distribution. This implies that any policy intervention aimed at reducing the overall pattern of employment polarisation should carefully consider the various demographic groups that are contributing to this phenomenon.
Acemoglu, D, and D Autor (2011), “Skills, Tasks and Technologies: Implications for Employment and Earnings”, Chapter 12 of Handbook of Labor Economics, vol. 4, part B, Elsevier, pp 1043-1171.
Autor, D H, and D Dorn (2013), “The Growth of Low-Skill Service Jobs and the Polarization of the US Labor Market”, American Economic Review, 103 (5), 1553-97.
Autor, D H, L F Katz, and M S Kearney (2006), “The Polarization of the U.S. Labor Market”, American Economic Review, 96 (2), 189-194.
Cerina, F, A Moro, and M Rendall (2018), “The Role of Gender in Employment Polarization”, LSE, Centre for Macroeconomics DP series CFM-DP2017-04.
Eeckhout, J, R Pinheiro, and K Schmidheiny (2014), “Spatial sorting”, Journal of Political Economy, 122 (3), 554-620.
 We follow Acemoglu and Autor (2011) and Autor and Dorn (2013) in constructing employment polarisation graphs for the US. First, we sort the population of occupations by mean wage in 1980, which is interpreted as a proxy for skills, and then construct occupation percentiles by weighting each occupation by hours worked in 1980. Next, we plot, for each percentile, the change in the employment share in total working hours in the economy.