Testing for redlining in the labour market

ABSTRACT

When an employer refuses to recruit a job applicant due to the applicant's place of residence, we speak of redlining in the labour market. There are two explanations for this practice by the employer. The first is the excessive distance between the applicant's place of residence and the workplace, justified by a logic of spatial mismatch. The second is based on the characteristics of the neighbourhood in line with a signal logic. We propose to measure the effects of these two mechanisms using a correspondence test conducted in the Paris region of France for two occupations: waiters and cooks. It appears that distance plays a significant role and reinforces the effect of a disreputable neighbourhood. The most deprived neighbourhoods combine these two types of drawbacks.

KEYWORDS:

• distance from place of employment
• testing
• redlining

JEL:

• C81
• C93
• J15
• J71

ORCID

Yannick L’Horty http://orcid.org/0000-0001-6258-8565

Mathieu Bunel http://orcid.org/0000-0002-0038-4759

Notes

1 In the 1930s, the Home Owners’ Loan Corporation used a colour code to identify in red (i.e., redline) residential areas with an African-American majority and/or impoverished population.

2 Correspondence tests in hiring are carried out without the knowledge of the recruiter who is kept in the dark about the fact that s/he is being sent and the ethical aspect of this is rarely discussed. Following Riach and Rich (2004) about the markets: that great harm is done to the social fabric by discriminatory practices in such markets; that minimal inconvenience is imposed on the entrepreneurs in the experiment; and that the technique provides evidence with a degree of accuracy and transparency that is not available from any other procedure.

3 It is difficult to find out the number of résumés sent in response to each offer. According to APEC, for 2011, the average number of résumés per job offer was 41.

4 A French administrative division is hereafter referred to as a department.

5 The closest English equivalent is a borough.

6 As the name of the ZUS is not necessarily known to all employers in Île-de-France, we strengthened the signal by indicating on the application the term ‘Pavilion Building’ in the body of the address of residence, which evokes without ambiguity a dwelling in a set of uniform buildings, corresponding in France to a deprived neighbourhood.

7 The websites of the Pôle d’Emploi and L’Hôtellerie-Restauration that centralize most job offers in the food services sector were used between mid-October 2011 and the beginning of February 2012 to identify potential offers. In total, 498 job offers from different facilities were tested: 253 for cooks and 245 for waiters. This corresponds to 2988 applications (6 × 498) filed.

8 We prefer this time measurement given that the résumés mention that the applicants have a driver's licence and access to a vehicle.

9 In a correspondence test in Stockholm and Gothenburg, Sweden, Carlsson et al. (2018) found that commuting distance is negatively associated with call-back rate. Phillips (2016) observed also in Washington, DC, that commuting time had a negative and significant effect.

10 Given that these applicants live slightly closer to the jobs than their counterparts from non-disadvantaged neighbourhoods in the same department, the distance effect should work in their favour.

11 Various boundaries were tested, with no change in the results.

12 When using public transit commuting time, results are largely the same, but they are not as pertinent because applicants had mentioned in their résumés that they owned a vehicle. Furthermore, there are few (7%) cases where information on commuting time was available and where commuting time differences are limited.

13 To compute the estimated probabilities, we follow the Train (2003) approach by computing the sample average probabilities instead of probabilities at sample means for the whole sample. To obtain each line, we fixed the value of DLAU and DNU and then computed the probabilities by using the coefficient from model 3 for different travel times. To take into account the multilevel structure of our model, we introduced a random number generated from a normal distribution of standard deviation of 3.757 clustered for each job.

Additional information

Funding

This study received financial support from ONZUS.