Productivity Signals and Disability-Related Hiring Discrimination: Evidence from a Field Experiment

We extend the well-established correspondence study approach used in the literature whereby hypothetical applications are sent to real-world job vacancies, and a gap in employer callback rates between otherwise equivalent disabled and non-disabled applicants is used as a measure of hiring discrimination. Our focus is, however, on applying a credentials approach to explore the source of discrimination. More specifically, we test the effectiveness of productivity signals that we hypothesise reduce the disability-related gap in callbacks in the presence of statistical discrimination where employers use the group characteristic, in this case disability, to infer lower individual productivity in the presence of incomplete information. Our productivity signals are considered along two dimensions. As part of the application, we first randomly include the offer of a positive employer reference, and second, an enhanced set of education and technical skills, and explore the impact on the disability gap in callbacks. In short, we assess whether disabled individuals can influence the probability of callback disproportionately by enhancing their productivity signals, in line with these being more important for disadvantaged groups where productivity is more uncertain. Our context is the UK, where disabled people, who represent nearly a quarter of the working-age population, are protected from discrimination under the 2010 Equality Act.

2022-10-01

2023-07-31

Employers' responses to the applications is the primary outcome of the intervention. Responses are recorded as positive where an invitation to interview is received, negative where the applicant is rejected or no response is received within six weeks and ambiguous where there is, for example, a response with a request for more information. Written requests for further information and invitations to attend interviews are declined using a pre-specified template within 24 hours to minimise the impact on employers.

As is standard in the literature, the outcome variable in the main analysis is the aggregate callback rate, which combines positive and ambiguous responses.

Only positive responses received from the employers.

The experiment requires several careful design choices, including the choice of disability and occupations, the design of credible CVs and cover letters, nature of disability signal in the job application, which productivity signals to employ and how to introduce these, how to identify vacancies and submit applications, how to measure callback, and what employer and job characteristics to measure and explore.

-We focus on wheelchair users, since disabled and non-disabled candidates with this type of mobility impairment are equally productive in many settings. Furthermore, this disability is easily understandable meaning the accessibility requirements are more likely to be known and already addressed. We selected the occupations of accountants and financial accounts assistants, because such impairment is assessed to have no direct impact on worker productivity in these occupations given the low physical requirements (Ameri et al., 2018; Bellemare et al., 2023).

-The application materials – fictitious CVs and cover letters – were designed to mirror industry standards and were constructed based on examples stored by real job seekers on the online hiring platform who were searching for employment in the same occupations. Applications varied across two key dimensions, the presence of (i) disability and (ii) signals of enhanced productivity. Otherwise, applications were kept similar. Applicants were intended to be credible for the role, with appropriate (but not outstanding) qualifications and work experience tailored to the occupation and city. For example, those applying for an accountant had obtained a degree level qualification and were a certified chartered accountant. Applicants were designed to be white British males with everyday sounding names, at an early career stage, employed, and with no gaps in the employment history. ,

-Following a well-established approach in the literature (e.g., Bellemare et al., 2018, 2023; Bjornshagen and Ugrenunov, 2021), we disclose wheelchair use in the cover letter among a randomly selected half of the applications: “If invited to attend an interview in person, I would like your reassurance that your building will be accessible to my needs since I use a wheelchair.” Discussions with our third sector partners confirmed this was a natural choice since it would need to be discussed at the next step in the hiring process.

-Our productivity signals take two forms, which we refer to as references and enhanced skills. For the reference, a statement was included within the cover letter of a randomly selected half of disabled and non-disabled applications: “I have letters of reference available from previous line managers, who can vouch for my past performance and productivity in my previous roles, my transferable skills, and the added value I create for the companies I work in.” For the enhanced skills, we focus on four occupation-specific skills designed to provide a positive productivity signal. For certified accountants, these skills included having a first class undergraduate degree to signal a higher academic achievement, having knowledge of relevant accounting software to signal technical fluency, having knowledge of coding data in structured query language (SQL) to signal coding skills, and being certified in advanced data analytics for accountants to signal additional upskilling. For financial accounts assistants, these skills included having an undergraduate degree to signal higher qualifications, having advanced knowledge of computer and accounting software to signal computer literacy, and having certified knowledge of financial reporting to signal upskilling. In summary, we have applicant profiles that differ in the presence of disability and productivity signals, that is, we have disabled candidates with and without productivity signals, and non-disabled candidates with or without productivity signals.

-Finding vacancies and submitting applications. Our search was constrained to occupations within 25 miles of 5 large UK cities (London, Birmingham, Manchester, Edinburgh, Cardiff) and covered the period from October 2022 to July 2023. We sent fictitious job applications to 4,004 job vacancies posted on a large online hiring platform. We adopted an unmatched approach so that either a disabled or a non-disabled application profile was submitted to each suitable vacancy.

-Aligned to evidence of the importance of corporate culture for disability-related gaps in employee outcomes (Schur et al., 2009) we manually recorded several employer and job characteristics based on the text of the job advertisement. Specifically, we identify whether the advert referred to equalities/equal opportunities/welcoming diversity of applicants which we refer to as pro-equality and whether there was a specific reference to disability equality, such as mentioning Disability Confident, the government accreditation scheme, mentioning being disability inclusive, or referring to reasonable accommodations being available for interview. We focus on an aggregate measure of employer equality defined as the presence of any one of these four characteristics but explore the role of the separate components by way of sensitivity analysis in Section 3.3. Given the potential for different forms of taste-based discrimination, particularly from co-workers and customers we further recorded job requirements relating to (i) teamwork and (ii) contact with clients/customers. Given the recent debate about the potential for working from home to support disability-related labour market equality (Hoque and Bacon, 2022) we also recorded the potential for remote work being mentioned in the vacancy.

When sending the CVs, we adopted an unmatched approach so that either a disabled or a non-disabled application profile was submitted to each suitable vacancy. Drawing from a pool of 16 disabled applicant profiles per city (differing based on the productivity signals) and 16 non-disabled applicant profiles per city, we randomized the order of profiles and used one disabled applicant profile and one non-disabled applicant profile each day, with the profiles varying across cities. The 16 types are formed as follows: half (8) included the offer of a reference. Within each half a further half (4) had no enhanced skills. Three of the four occupation specific skills were randomly included within the remaining half (4). For each vacancy posted on the same day, one application was sent alternating between the disabled and the non-disabled candidate profile. In all cities on odd calendar days, the profile of a non-disabled applicant was used for the first suitable vacancy, followed by the profile of a disabled applicant for the second suitable vacancy. For the third suitable vacancy, the profile of the non-disabled applicant was used again. On even calendar days, the order was reversed.

CVs.

No

0

4500

Wihtout productivity signals and without disability: 1125
With productivity signals and without disability: 1125
Without productivity signals and with disability: 1125
With productivity signals and with disability: 1125