Donald R. Davis, Eric Mengus, Tomasz K. Michalski23 April 2020
In recent years, there has been a great deal of concern about modern labour markets causing both inequality between people as well as inequality between places. The by-now voluminous literature on labour market polarisation has focused mainly on the disappearance of middle-class jobs as an explanation. Prominent papers in this literature include Acemoglu (1999), Autor, Katz and Kearney (2006) and Autor and Dorn (2013) for the US and Goos et al. (2014) for Europe. At the same time, there has been a growing literature on how cities that are skilled and typically larger have become even more successful while cities that started with lower skill levels and are typically smaller have fallen further behind. Most notable in launching this literature is Enrico Moretti who in his book New Geography of Jobs termed the contrasting fate of these cities “the great divergence.” His work has been followed by important contributions such as Diamond (2016) and David Autor (2019) started gathering evidence on how labour market polarisation is experienced differently in areas that differ in density in his Ely lecture. So far however, no connection has been made between the two literatures. In Davis et al. (2020), we fill this gap by joining labour market polarisation and the great divergence into a common theoretical framework. We further provide evidence supporting this framework based on French cities.
A model of labour market polarisation and the great divergence
Following Autor and Dorn (2013), there are three core tasks in our model and a capital or offshorable good that is a close substitute for the middle-skill task. As the price of capital or offshoring declines, pressure is placed on the middle-paid jobs. We insert this framework into a heterogeneous labour model that builds on Davis and Dingel (2020). Our key results include both labour market polarisation in all cities as well as a great divergence across cities.
Furthermore, our theory predicts that large cities will experience a sharper decline in middle-paid jobs, that these job losses will be concentrated in an upper tier of the middle-paid jobs, and that they will be primarily replaced by high-paid jobs. In contrast, smaller cities experience a more modest decline of middle-paid jobs, these job losses will be relatively concentrated in the lower tier of middle-paid jobs, and the lost middle-paid jobs will be replaced to a greater extent by low-paid jobs.
Evidence on French cities
We examine these theoretical predictions for a set of all 117 French cities with a population above 50,000 over the period 1994-2015 using French administrative data.
First, we find that labour market polarisation occurs both in the aggregate and at individual city level where it is ubiquitous as shown in Figure 1.
More precisely, all of the 117 largest cities in France experienced a decline in employment of middle-paid jobs over the period 1994-2015. Labour market polarisation is found in 115 cities out of 117 studied, and the two exceptions both have less than 60,000 inhabitants.
Figure 1 Labour market polarisation within cities and the great divergence
Note: This figure shows percentage point changes in employment shares of middle-paid against high-paid jobs for individual cities for the period 1994-2015. Each red square, blue dot or green check symbolizes, respectively, a large (above >0.5m inhabitants), medium-sized (0.1-0.5m) or small (0.05-0.1m) city. Names of cities with more than 0.5m inhabitants are shown. N=117; 11 cities > 0.5m, 44 cities between 0.1-0.5m and 62 cities between 0.05-0.1m inhabitants in 2015.
Second, labour market polarisation happened in a diverging way between large and small cities. In large cities, the loss of middle-paid jobs is sharper (by a factor of roughly two relative to the smallest cities), but these middle-paid jobs are primarily replaced by high-paid jobs. In small cities, the loss of middle-paid jobs is still strong but more moderate, and the middle-paid jobs are primarily replaced by low-paid jobs. This pattern is consistent with the great divergence and can be observed in Figure 1, where cities above the dotted line witnessed a stronger increase in the share of high-paid occupations. Eight out of eleven cities with more than 0.5 million inhabitants are in this category while three (Nice, Toulon and Douai-Lens) experienced a stronger growth of low-paid occupations instead. It is clear from Figure 1 that a great majority of middle-sized and small cities experienced a stronger growth in low-paid than in high-paid jobs.
This stronger reallocation towards high-paid jobs in large cities can be better observed in Figure 2, where we gather cities into bins depending on their size but split middle-paid jobs into two subgroups depending on their initial average wages. We observe that in largest cities, roughly two high-paid jobs are created for each low-paid job, while the reverse happens in small cities. Moreover, in large cities the strongest margin of adjustment is between higher-wage middle-paid jobs and high-paid occupations.
Figure 2 Labour market polarisation and the great divergence across three different city size groups, 1994-2015 and four employment groups
Note: This figure shows percentage point changes in employment shares of high-, low- and different types of middle-paid jobs with hours worked summed by job types and 3 city sizes: large (above >0.5m inhabitants), medium-sized (0.1-0.5m) and small (0.05-0.1m) in the period 1994-2015. We divide middle-paid jobs into two groups depending on the average wages in 1994 for the 2-digit jobs categorized accordingly to the French occupation classification called PCS (Nomenclature des professions et categories socio-profesionnelles). This leads to a group above the median (CS codes 48, 46, 47, 43 and 62 in decreasing wage order) and a group below the median (CS 54, 65, 63, 64, 67).
Third, we show that the initial exposure to middle-paid jobs is not a good predictor of the loss of middle-paid jobs, since the large cities started the period with the lowest exposure to these jobs yet also experienced the largest percentage point decline.
What do we learn?
Overall, our findings show a diverging trend in labour market outcomes for small and large cities in France. While in large cities, a lot of middle-paid jobs were replaced with high-paid jobs since 1994, in smaller cities fewer middle-paid jobs were replaces but relatively more with low-paid ones. Importantly, we can directly tie two prominent labour market phenomena of the last 40 years. Technological shocks such as automation or price pressure from globalization (offshoring) drive not only the destruction of a wide range of middle-paid jobs, but also the divergence of large, skill-abundant cities from small ones. Our conclusions for France may also apply more generally as the coexistence of labour market polarisation and diverging fates of cities depending on their size appears to be a robust fact across many advanced countries.
These developments certainly have important consequences for other economic phenomena such as the provision of local public goods and other political and social outcomes. France for example, experienced increasing political polarisation and the rise of extreme political parties over the period of interest. Similarly, spatial inequalities became an important topic in policy discussions in the country after the yellow vest protests that erupted in the fall of 2018. Our paper shows that it may not be easy to counter these developments as they are driven by profound economic forces.
Acemoglu, D (1999), “Changes in Unemployment and Wage Inequality: An Alternative Theory and Some Evidence”, American Economic Review 89(5): 1259–1278.