Wage subsidies and hiring chances for the disabled: some causal evidence

In addition, hourly wages and training opportunities are also lower among disabled employees [3, 9, 13, 37, 38].

Within the classical competitive framework in which wages and employment are formed by a confrontation of labor supply and labor demand, wage subsidies will shift out the labor demand curve for the targeted workers. This is the case because employers take into account the total labor costs when making their hiring decision. In case labor supply is infinitely elastic, this subsidy will expand only the employment of the targeted individuals without any impact on wages. The lower the labor supply elasticity is, the smaller the effect of the subsidy on employment will be and the larger will be its effect on wages [7, 22]. Findings demonstrate that this effect can also be expected based on models with moderate wage rigidities, labor market imperfections, or structural unemployment [18, 39]. In general, wage subsidies are aimed at compensating the disadvantage position and (thereby) the potentially (perceived) lower productivity of the targeted workers. In that respect, wage subsidies for the disabled are quite comparable with subsidies targeted at other groups, as theoretically and empirically evaluated, for example, by Bell et al. [5], Burtless [8], Gerfin et al. [16], Jaenichen and Stephan [20] and Kangasharju [21].

Flanders is the northern, Dutch-speaking part of Belgium.

During the last two decades, academics have provided evidence for labor market discrimination against disabled persons in the United States, Greece, Ireland, the Netherlands, Scotland, Canada, and France [2, 13, 15, 17, 25, 26, 33].

As a consequence, after the employee assigned the FSS is hired, an additional part of the subsidy might relate to the seriousness of the disability, even for blind, deaf, and autistic individuals. However, because our experiment was focused on the hiring stage only, the size of the subsidy was equal for all (fictitious) candidates included in our study.

For instance, the job characteristics that for all three disabilities led to ignoring the vacancy were (1) high requirements concerning communication skills (e.g., a great deal of selling contacts with customer), (2) safety risks being more prominent given the disability (e.g., working with toxic fluids for blind people), and (3) high mobility requirements (e.g., regular visits during the day to customers outside the firm). In addition, for the autistic candidates, job postings in which flexibility was disclosed as a key quality were ignored.

The only method to our knowledge that could be used to obtain causal measures of unequal treatment at later stages was audit testing. Audit tests go one step further than correspondence tests by sending matched pairs of actors to job invitations. However, in the economics literature, this methodology currently is only seldom used because it has been criticized on various grounds [19, 34]. The main critique is that audit tests suffer from the problem that it might be impossible to find and successfully train real-life job applicants so that they really represent a perfect match. In addition, auditors might consciously or subconsciously be motivated to prove discrimination and might therefore adjust their behavior accordingly in an interview.

24.74 % = (76 + 114)/768.

For instance, to test whether positive call-back ratios differ by education level, we regressed positive call-back on a dummy “highly educated,” an interaction dummy “disability × highly educated,” and an interaction dummy “disability × moderately educated.”

The reader might be puzzled by the fact that the statistics presented in the fourth and sixth columns in panel A of Tables 2 and 3 are not the same as those presented in the ninth column of panels C and D of Table 1. This is because the former statistics were obtained by comparing the disabled candidates with or without a wage subsidy with all the nondisabled candidates (768 individuals), whereas the latter statistics were obtained by accounting only for the nondisabled candidates who applied for the same vacancies as the disabled candidates under concern (384 individuals).

As should be clear based on the "Data" section, we did not randomize over the particular disability disclosed due to our aim of selecting occupations for which the disabled candidates could be expected to function as productively as the nondisabled candidates. Thus, the low positive call-back rates for both the fictitious disabled and nondisabled candidates in the pairs comprising a blind candidate seem only to be a reflection of the lower positive call-back rates in the occupations (in general or for our profiles of graduates in particular) we selected when applying with these pairs (accountant, informatician, administrative clerk, and teleseller).

The fact that 1.30 is not significantly different from 1 is related to the low interview invitation rate among the disabled candidates (6.90 %).

The reader will notice that this perception is, in fact, a misperception because wage subsidy entitlement is not related to the severity of the disability, at least for blind, deaf, and autistic people (see the "Institutional Context" section).

Further analysis, which is referred to when a heteroskedastic probit model is used (see further), indicates that this finding is, at least empirically, driven by the blind candidates. For them, mobility might always be an issue, for both small and long distances, so that living close to the workplace is less rewarded.

To see this more clearly, assume that both the average observed and unobserved determinants of productivity were the same for the nondisabled candidates, disabled candidates mentioning entitlement to a wage subsidy, and disabled candidates not mentioning a wage subsidy but that the variance of unobservable job-relevant characteristics was the lowest for the nondisabled candidates. In addition, suppose that the employer considered the observed determinants of productivity, inferred from the CV and the motivation letter, as relatively low compared with the job requirement. In that case, it was rational for the employer to invite the disabled candidate because given that the variance of unobservable job-relevant characteristics was higher for disabled candidates, it was more likely that the sum of observed and unobserved productivity was higher for these workers. A correspondence test that detects discrimination against disabled candidates could then underestimate the extent of discrimination.

In addition, if we re-estimate the model (4) of Tables 4 to 7 after leaving out the observations for blind candidates, we find no statistically significant effect of the interaction between the candidate’s disability status and the distance between the working place and his residence on his probability of positive call-back.