Differences in ability and sorting

Dans le document Essays on location choice: agglomeration, amenities and housing (Page 79-85)

Getting a first job: quality of the labor matching in French cities

2.4 Empirical strategy

2.5.4 Differences in ability and sorting

People with different unobservable characteristics may sort differently across space. Ac-cording to Wheeler (2006), ability sorting is likely to be of minor importance for young individuals as workers gradually learn about their competencies with work experience.

However, this problem may still affect estimation results upwards if unobserved workers’

ability are correlated with city characteristics (Behrens et al., 2014; Combes et al., 2008, 2010; Mion and Naticchioni, 2009, for example). There are three ways to correct for the self-selection of workers. The first approach is to include observable workers’ char-acteristics that mirror differences in ability such as grants or tests scores (Baum-Snow and Pavan, 2012; Wheeler, 2006). The second is to compare patterns for movers and non-movers as workers with different abilities tend to sort unevenly across space (Combes et al., 2012; Di Addario and Patacchini, 2008; Glaeser et al., 2001). The third strategy is to include individual fixed effects to control for all unobservable worker characteristics (Combes et al., 2010; D’Costa and Overman, 2013; Matano and Naticchioni,2012; Mion and Naticchioni, 2009). Since there is only one observation per worker, this last approach is unfortunately not feasible18.

I first control for the influence of ability by including a set of individual characteris-tics that mirror workers’ unobserved abilities. I control for the role of social capital and networks in column (1) of Table 2.7 by including dummies for workers’ parents’ occu-pation. These variables aim at capturing the influence of access to informal sources of information in the job-market search and differences in social capital that may affect a worker’s ability to find a job19. I also include in column (2) a set of dummy variables for

18It is not also feasible to use information on subsequent job changes to get several observations per worker as the low number of repeated observations per worker is likely to introduce an incidental parameters problem which yield inconsistent estimates of other parameters (Wooldridge,2010)

19See Ioannides and Datcher Loury (2004) for a review of the literature on job-search and social interactions.

having graduated with laude at the end of high-school. I therefore include a dummy for graduating with cum laude, magna cum laude or summa cum laude. As the information can only be obtained for college graduates, the number of observations falls drastically in these estimations. However, including these proxies for unobserved ability does not affect the coefficients on agglomeration that remain positive and significant. Coefficients on added dummies provide mixed results. It seems indeed that graduating with laude does not significantly influence the quality of job match. Similarly, parents’ occupation does not play a role in explaining their offspring’s job match.

Table 2.7: Spatial Sorting: ability

In column (3), I control for the influence of the place where school leavers grew up (at the age of 10). This variable is designed to capture differences in access to amenities such as culture and good schools during childhood. Individuals that grew up in Paris or in other big cities may have faced a better access to museums, libraries, theaters or book stores than workers who have been raised in smaller cities. Similarly, school quality may vary across cities affecting workers’ knowledge and abilities. More generally, Bosquet and

Overman (2015) have shown the importance of the place of birth in explaining spatial sorting and a worker’s current wage. Conditioning out the influence of the place of birth, employment density still plays a positive role on the expected quality of match. The coefficient associated with agglomeration remains indeed stable across estimations.

Next, I follow Wheeler (2006) and Andini et al. (2013) and take into account workers’

mobility as a signal for sorting. I consider several patterns of geographic mobility and evaluate the effect of density for different types of workers.

In Table 2.8, I define workers as movers when they change d´epartement after gradu-ating. In terms of mobility, I find that around 45% of interviewees changed location after initial training while 55% stayed in the same area. In column (1), I restrict the sample to non-movers. As we expect more talented people to move ex ante to big cities, restricting the sample to non-movers aims at conditioning out the influence of high ability workers.

Column (2) focuses instead on movers exclusively. Estimates show that the coefficient on agglomeration is positive and significant only for non-movers. This suggests that the better job match in large cities is not explained by the sorting of talented people after graduation. In column (3), I reestimate Equation 2.1 on the full sample and add a dummy variable for mobility (which equals 1 if the worker changed location after graduating and 0 otherwise). The coefficient of interest survives this sensitivity test even if the significance level is lowered. Finally, I add in column (4) an interaction term between the indicator of mobility and the density of the labor market in which the individual works. We can observe in this estimation that the coefficient associated with density remains positive and similar in terms of magnitude. Besides, the coefficient on the interaction term is nil. This implies that individuals who move after graduating do not experience a disproportionate match in dense cities.

In Table 2.9, I replicate the previous strategy using a more restrictive measure of mobility. In panel A, workers are considered as non-movers when they are currently working in the region in which they were in high school, and asmoversotherwise. Similarly in panel B, workers are referred to as movers when they changed region from childhood (at the age of 10) to adulthood. Hence, non-movers represent in both cases workers that are

Table 2.9: Spatial Sorting: mobility Panel A Mobility since high school

(1) (2) (3) (4)

Non-movers Movers All All

Density 0.016*** 0.005 0.008* 0.015***

(0.006) (0.005) (0.005) (0.004)

Movers 0.005 0.055***

(0.008) (0.017)

Density × Movers -0.010***

(0.003) Observations 10,468 10,489 20,957 20,958

Panel B Mobility since childhood

(1) (2) (3) (4)

Non-movers Movers All All

Density 0.021*** -0.002 0.009** 0.015***

(0.004) (0.004) (0.004) (0.004)

Movers 0.003 0.053***

(0.007) (0.017)

Density × Movers -0.011***

(0.004) Observations 19,527 13,397 32,924 32,996

strongly tied to a local labor market. The results of Table 2.9 provide similar conclusions:

columns (1) depict a bigger effect of urban density for non-movers. The analysis of the interaction terms shows that the influence of density is lower for movers – but remains statistically positive. I report in Table 2.10 the results of a replication of Table 2.9 using employment areas instead of departements as geographical units of analysis. The results are qualitatively and quantitatively similar.

Table 2.10: Spatial Sorting: mobility

Schools and universities are located in cities with a minimum density. As young individuals are relatively mobile, it might be the case that individuals who do not find a good match in these large cities after graduating return to a smaller city afterwards.

This might explain the low quality of skill match observed in small cities. One step further is taken by using information on workers’ history in terms of mobility. I then distinguish 5 types of workers. Type 0 (the benchmark population) represents workers who never changed location (at the level of departements): they found their first job in the same region in which they grew up and studied. Type 1 represents workers who moved for their studies but ultimately returned to their initial region afterwards. Type 2 represents workers who are the most mobile: they changed location after childhood for studying, and moved to a third location after graduating. Finally, types 3 and 4 respectively represent workers who changed location for their study and found a job in this location after graduation, and workers who studied in the region in which they grew up but changed location after graduating.

The results are provided in Table 2.11. All regressions include the full set of control variables as well as fixed effects. The analysis of the interaction terms in column (1) shows that non-movers (type 0) are indeed workers who experienced the strongest effect of density in their index of skill matching. Interestingly, the interaction term for the most mobile workers (type 2) is negative and highly significant: the effect of density on skill match is lower for movers than for non-movers.

Finally, column (2) replicates this analysis using the region in which the individual was located in high school instead of childhood. By contrast, column (3) replicates column (1)

using the place of residence in 2004 instead of the work place. The results are qualitatively similar. We can conclude with confidence that the main empirical results are not driven by the sorting of talented workers in large cities.

Table 2.11: Spatial Sorting: mobility

Density × Type 1 -0.002 -0.005 -0.006*

(0.005) (0.005) (0.003)

Density × Type 2 -0.012*** -0.012*** -0.008***

(0.004) (0.004) (0.003)

Density × Type 3 -0.013** -0.010 -0.006

(0.006) (0.006) (0.005)

Density × Type 4 -0.009 -0.009 -0.008

(0.006) (0.008) (0.005)

If each box represent a given location, types are represented by:

Type 0 (benchmark population): A B C Type 1: A C ↔ B

Type 2: A → B → C Type 3: A → B C Type 4: A B → C

Dans le document Essays on location choice: agglomeration, amenities and housing (Page 79-85)