Ukrainian refugees and labor market discrimination

To measure the differential treatment of refugees, we conduct a matched pair email correspondence test, in which each employer receives an email indicating the applicant is Czech and an email indicating the applicant is non-Czech, that is, either a refugee (i.e., from either Ukrainian or Russian linguistic group) or a permanent resident (i.e., with either Ukrainian or Russian citizenship). These two emails are sent in a randomized order (e.g. sometimes the first email is from the Czech applicant, other times it is from the non-Czech applicant), a few days apart. Every answer from employers is going to be followed by a courteous email, where the fictitious applicant thanks for the response and adds that she has found another job already. This is done to minimize the employer's inconvenience.

Given the protraction of the conflict and the civil unrest caused by sanctions consequences (e.g. increase in the gas price), we are prolonging the experiment, while expecting that this new scenario could worsen the attitudes toward refugees.

2026-03-26

2026-09-26

We record employers’ responses to our email inquiries.

The primary outcome is a binary variable for receiving a positive response to the query. We create various versions of this binary outcome.

We measure the time to respond, in days.
We measure the courteousness of the responses.

We signal the refugee status using some version of the following phrase: “I just came from Ukraine”; for refugees, we add one closing sentence similar to “Apologies for possible grammar mistakes, I am translating this message with Google translate, from Ukrainian (or Russian) to Czech.” We signal the permanent resident status with a phrase such as “I am Ukrainian (or Russian) and I am a permanent resident;” for them, there is no closing sentence citing Google translate. To signal the applicant does not have any child we insert a version of the following sentence in the message “My schedule is flexible since I do not have any kids.”

In the first wave of data collection, we assumed away that language proficiency did not make any difference in employers’ responses, given the language proximity and irrelevance in job for which we applied (see below). In this additional data collection wave, we additionally randomize language proficiency for some refugees. Language proficiency is signaled with a sentence similar to “I arrived to the Czech Republic from Ukraine about half a year ago, and I have been learning Czech in a course for Ukrainian / Russian speakers, so I can communicate and work in Czech.”

The main signal for whether the applicant is Czech, Ukrainian, Ukrainian-Russian/Russian is represented by combinations of names and surnames. Czech applicants are randomly assigned a combination of typical names and surnames, while non-Czech applicants are assigned a combination of typical names and surnames based on being either Ukrainian or Ukrainian-Russian/Russian. An employer has a 100 percent probability to receive a message from a Czech applicant. The assignment of the non-Czech identity and status is randomly assigned with the following probabilities: (a) a 12.5 percent probability of receiving a message from both a Ukrainian refugee, and a Ukrainian-Russian refugee, (b) a 7.5 percent probability of receiving a message from both a Ukrainian permanent resident, and a Russian permanent resident, and (c) a 30 percent probability of receiving a message from both a Ukrainian refugee, and a Ukrainian-Russian refugee, who has learnt Czech. Please note that these probabilities are approximated; the final distribution may not reflect this distribution due to random sampling.

We use the same combinations of name and surname for control and treatment groups as in the first wave. However, the reader should note that the list of name and surname in "Docs & Materials" project 9126 was changed immediately before the beginning of the first wave in response to inputs our Ukrainian RA and students provided us. Thus, names list for this (and the previous) data collection is therefore:
For Czech applicants: Katerina Novakova, Lenka Horakova, Lucie Dvorakova.
For Ukrainian refugees from the Ukrainian linguistic group and Ukrainian permanent residents: Olha Shevchenko, Zhanna Marchenko, Anzheia Kharchenko.
For Ukrainian refugees from the Russian linguistic group and Russian permanent residents: Evgeniya Sergeeva, Galina Goncharova, Evgenia Guseva.

Rather than applying for advertised job vacancies, we “fish for discrimination,” that is, we send short written messages to represent unsolicited job applications for ISCO 8 and 9 jobs; these are jobs that refugees usually apply to. This is a usual practice in Czechia, for basic jobs, based on opinions from Czech human resources experts. These are jobs for which previous experience is not required, a CV is not necessary, only basic education is required, there is no contact with customers, and the employee is not expected to manage coworkers. Based on these characteristics and based on the proximity of Czech to both Ukrainian and Russian languages, one could safely assume that employers’ perceived productivity differences between different fictitious candidates are economically risible. We are aware of two field experiments to study the role of refugees' language on their chances to find a (low-skilled) job. One study does not find evidence of an effect of language proficiency on chances of receiving a positive answer from employers (Ek et al., 2021), while another study (Carlsson et al., 2025) finds such an effect for any type of job, without differentiating based on language proximity. However, neither study focuses on treatment groups whose language belongs to the same group of the control group, which would plausibly reduce the role of language proficiency. However, with this additional data collection wave, we will be able to formally investigate the role of language proficiency.

To collect our sample of auditable employers, we proceed in two steps. First, we define the jobs we are going to apply for. Based on statistics from the Ministry of Labor, we select those ISCO 8 and 9 jobs in which female third-country nationals (i.e., neither Czech nor EU citizens) are more frequently employed (i.e., jobs that together represent 75% of the third-country nationals in ISCO 8 and 9). The selection of these jobs strengthens the message's realism and further reduces the probability that employers perceive the lack of knowledge of the Czech language as being a productivity-reducing characteristic. Second, from the national statistical office registry of Czech firms, we select a random sample of firms that typically employ workers in the jobs defined in the first step. Firms' probability of being selected is weighted by the regional population out of the national population. This procedure allows us to make sure that we contact employers from every region and proportionally to the population of that region.

Not available

Randomization is done in the office by a computer.

Firm

No

We expect to contact 4,800 employers

Each employer receives two emails, so we have 4,800 x 2 = 9,600

Each employer receives two emails, one from a Czech applicant (control group) and one from either of the four treatment arms: a Ukrainian refugee from the Ukrainian linguistic group, a Ukrainian refugee from the Russian linguistic group, a Ukrainian permanent resident, and a Russian permanent resident.
4,800 Czech fictitious applicants
600 Ukrainian refugee fictitious applicants
600 Ukrainian refugee from the Russian linguistic group applicants
1,440 Ukrainian refugee fictitious applicants
1,440 Ukrainian refugee from the Russian linguistic group applicants
360 Ukrainian permanent resident fictitious applicants
360 Russian permanent resident fictitious applicants