Polarisation du marché du travail : Y a-t-il davantage d’emplois peu qualifiés ? – A Response

Ariell Reshef1 and Farid Toubal2

A recent Note d’analyse of France Stratégie has argued that employment in France has not polarized over the last 25 years3. While the effort to compare different data sources and methodologies – as in the Note d’analyse – is absolutely commended, one must recognize that some data sources and methodologies are better than others. Using careful analysis, and relying on the best data sources available, we conclude that job polarization in France is real, however complex this reality may be.

Job polarization is the gradual process of separating the labor force into two “poles”: highly paid workers and low wage workers, with few workers in between. It is commonly described as the concurrence of three tendencies: (1) an increase of the employment share of low wage workers, together with (2) a decline in the employment share of middle wage workers and (3) an increase of the employment share of high wage workers. This would manifest in a U-shape evolution of employment shares, ranked by wages from low, to medium, to high. There is no argument at all about the second and third tendencies in France over the last 25 years. The debate is about the first. Different data sources, samples, levels of aggregation and definitions of what constitutes low versus medium wage workers may change our perception of whether the left branch of the U is real, important, or not. In particular, the Note d’analyse seems to argue that the employment share of low wage workers (the bottom two deciles) has not changed much – suggesting a J-shape pattern. Graphique 5 below which is reproduced directly from the Note d’analyse does imply a pattern of polarization: overall, employment of middle wage workers decreases, while both high and low wage workers gain employment shares. The dashed line in Graphique 5 delineates this general tendency.

Notes to Graphique 5. The figure is copied from France Strategie, Note d’analyse, déc. 2020 n° 98. Data source : Insee-DGI, enquête Revenus fiscaux et sociaux rétropolés de 1996 à 2017. The X axis ranks each occupation (classifications from INSEE) according to its average wage. The size of each circle is proportional to the employment share of the corresponding occupation in 2017. The Y axis reports the percent point change in the employment share from 1996 to 2017 of the corresponding occupation. The dashed-line curve represents the tendency for polarization.

Graphique 5 is constructed using survey data. Surveys tend to under-represent low wage workers and thus provide relatively poor quality indicators for these workers, which is the crux of the controversy. We argue that even if the employment share of low-wage workers has been stable, as the Note d’analyse seems to argue, then this would still imply overall polarization. All this notwithstanding, reliable analysis that is based on the same underlying data source clearly demonstrates a polarization pattern, shown in Figure 14.


Notes for Figure 1. The figure is constructed using the methodology of Autor and Dorn (2013) using the French Enquête Emploi (INSEE)5. This is the survey that underlies the data used by the Note d’analyse in Graphique 5. The X axis allocates each occupation to a wage percentile (there are 100 percentiles). The Y axis reports 100 times the percent point change in the employment share of each occupational wage percentile from 1994 to 2008. We thank Sebastien Bock for creating this figure for us.

Our own work, which is based on quasi-exhaustive French administrative data that do not rely on sampling (DADS, déclarations annuelles de données sociales) show a clear U-shape pattern, as shown in Figure 26. We discuss below why we think that these data are a superior source of information.

Notes to Figure 2. This figure is constructed using les DADS (déclarations annuelles de données sociales). The X axis ranks each broad occupation group (classifications from INSEE) according to its average wage. The width of each column is the employment share of the corresponding occupational group in 1994. The Y axis reports the percent point change in the employment share from 1994 to 2013 of the corresponding occupational group.

In what follows we wish to clarify the issues, and argue that Job Polarization indeed took place in France over the last 20-25 years.

First and foremost, it is important to understand that, in fact, a J-shape evolution also implies job polarization over time. Why? Because even if the share of low wage workers does not change (say, the bottom quarter), it increases relative to that of middle wage workers (which declines). If this process continues, then society bifurcates into a low wage class and a high wage class, with little in between (this is precisely the, so-called, “hollowing out of the middle class”). A U-shape type polarization process may deliver us there more quickly, but the tendency is the same. A better scenario would have been a net diminishing share of low wage workers (as was observed in the 1980s and early 1990s), thus leading to “occupational upgrading”, as some sociologists call this phenomenon7. But this is not what we observe, whether the U or the J pattern is correct. Hence the social and political concerns that job polarization undermines the foundation of democratic institutions.

Second, it is equally important to acknowledge that low-wage employment is almost surely underestimated, especially in surveys, on which the Note d’analyse relies upon. Immigrants, workers in the informal sector, and other low wage workers are less likely to be sampled and less likely to respond to surveyors even if they are sampled. Granted, statistical agencies try to address this bias (by allocating each respondent a “sampling weight”), but the methods used to do so are imperfect (since the sampling weights rely on estimates themselves). In contrast, quasi-exhaustive French administrative data (les DADS, déclarations annuelles de données sociales) that do not rely on sampling show a clear U-shape pattern, as demonstrated in our work (e.g., Figure 2), even though the DADS does not cover at all the informal sector. Nevertheless, using standard methodology, one obtains a very similar pattern when using the INSEE’s Enquête Emploi, the leading labor market survey for France; see Figure 1. In addition, not taking into account unemployment, which has increased significantly since 2008, particularly for low-wage workers, masks the real extent of the share of low wage workers in the labor force (which is not the same as their share in employment, the difference being the unemployed). This dimension is absent both in the analysis of France Stratégie and in ours.

Third, defining employment shares by “qualification”, i.e., diplomas, as the Note d’analyse does in part of its analysis, is misleading. For example, professional drivers (including trucks and other transport equipment) are typically without tertiary university degrees, enjoy middle-range wages, and saw their employment shares increase in recent times. At the same time, skilled industrial workers (ouvriers qualifiés de type industriel) and administrative workers, who are similarly positioned in terms wages – but not in terms of formal education – saw their employment share decline. The view through the lens of formal qualifications becomes even more complicated because ever more young people obtain a tertiary degree. If this sounds confusing, then it serves to make our point: formal qualifications are not the correct yardstick for measuring polarization. Indeed, one of the troubling aspects of the reductions of middle wage employment is that this comprises of many workers that previously invested in their education and obtained tertiary degrees.

Fourth, one must consider the role of specific sectors. As shown in our work, polarization, including the increase in the employment share of low wage workers, is most markedly observed in the non-farm, non-manufacturing private services sector, which accounts today for more than 85% of total employment. Agricultural and manufacturing workers (both with high qualifications and low) have been affected quite differently by technological change and globalization over the last 25 years, compared to workers in the services sector. In fact, ignoring clearly-defined agricultural occupations in Graphique 5 accentuates the rise of low-wage employment in France. And while the share of middle wage health system workers has increased, the forces that drive this tendency (mostly demographic needs) are quite different from the technological drivers of polarization (information and communication technology), which are a quasi-consensus among economists.

The Note d’analyse argues that it is not at all clear whether France has polarized more than other, similar European economies. We have no disagreement with this point, as international comparisons of this sort are inherently complicated due to different data sources, statistical methods and labor market institutions across countries. The OECD has tried to “harmonize” such datasets, a process that involves making many important calls of judgement. In fact, the OECD itself has placed France differently in terms of the degree of polarization compared to other economies across different publications8. However, this does not mean that France has not polarized at all.

We show in “La Polarization de l’Emploi en France », that private sector employment in France has experienced salient job polarization since the mid-1990s and that the 2008 crisis accelerated this trend.

  1. CNRS, Université Paris 1 Panthéon-Sorbonne, Paris School of Economics and CEPII, ariell.reshef@psemail.eu
  2. University of Paris-Dauphine – PSL, CEPII, CESIfo and CEPR, farid.toubal@dauphine.psl.eu
  3. Jolly, C. and Dherbécourt, C. « Polarisation du marché du travail : Y a-t-il davantage d’emplois peu qualifiés ? », France Strategie, Note d’analyse,déc. 2020 n° 98.
  4. Figure 1 is adapted from the PhD thesis of Sebastien Bock, a recent graduate of PSE under the supervision of Jean-Olivier Hairault, Francois Langot and Ariell Reshef. We thank Bock for producing for us this figure.
  5. David, H. and Dorn, D., 2013. The growth of low-skill service jobs and the polarization of the US labor market. American Economic Review, 103(5), pp.1553-97.
  6. Reshef A. and Toubal, F., 2019, La Polarisation de l’emploi en France, ce qui s’est aggravé depuis la crise de 2008, CEPREMAP. And Harrigan, J., Reshef, A. and Toubal, F. (forthcoming) “The March of the Techies: Job Polarization Within and Between Firms”, Research Policy.
  7. For example, Oesch, D., 2013. Occupational change in Europe: how technology and education transform the job structure. Oxford University Press.
  8. For example, in the 2019 OECD publication Under Pressure: The Squeezed Middle Class, https://read.oecd-ilibrary.org/social-issues-migration-health/under-pressure-the-squeezed-middle-class_689afed1-en#page1 , France exhibits an average degree of job polarization from 1995 to 2015, compared to other OECD members. In contrast, another 2019 OECD publication, Perspectives de l’emploi de l’OCDE 2019, http://www.oecd.org/employment-outlook/2019/index-fr.html , identifies France as one of the countries that polarized the most.