A Study on Ukraine War Veterans

This study investigates discriminatory hiring practices against Ukraine war veterans in Russia through a correspondence audit. Two types of fictitious résumés are submitted to job postings: one signaling voluntary service in the Ukraine conflict (treatment group, labeled "Served during Special Military Operation, 2022–2023") and one signaling routine mandatory conscription (control group, labeled "Completed mandatory military service, 2018–2019"). The intervention tests employer responses to these military service signals to quantify any hiring penalty associated with voluntary, financially incentivized participation in the conflict, distinct from general military service.

The correspondence audit targets entry- and mid-level job postings on platforms like HeadHunter (hh.ru) and SuperJob (superjob.ru) across sectors (retail, logistics, IT, administration, manufacturing) in cities including Moscow, St. Petersburg, and smaller locales. Each pair of résumés is identical in qualifications (e.g., education, 3–5 years’ experience, skills) but differs in the military service signal. Applicant details (e.g., common Russian male names, unique emails, virtual phone numbers) are randomized to ensure comparability. A callback is defined as any positive employer response (e.g., interview invitation, request for additional information) within 30 days, tracked via dedicated voicemail and email accounts. The Bayesian sequential approach analyzes data in batches of 20 paired applications, stopping when P(Δ>0∣Data)>0.975P(\Delta > 0 \mid \text{Data}) > 0.975P(\Delta > 0 \mid \text{Data}) > 0.975 or after 300 pairs, ensuring efficient evidence collection.

2024-09-01

2024-12-31

The key outcome variable is the callback rate for résumés submitted, defined as the proportion of applications receiving a positive employer response (e.g., interview invitation or request for additional information) within 30 days.

The callback rate is a binary outcome (1 for a callback, 0 otherwise) calculated for each résumé. The primary comparison is the difference in callback rates between the treatment group (résumés signaling Ukraine conflict service) and the control group (résumés signaling mandatory conscription), denoted as Δ=pC−pU\Delta = p_C - p_U\Delta = p_C - p_U, where pCp_Cp_C is the conscript callback rate and pUp_Up_U is the veteran callback rate. This difference quantifies the hiring penalty associated with voluntary Ukraine service. No constructed outcomes (e.g., indices) are used; the analysis focuses directly on raw callback rates and their difference, as is standard in correspondence audits (Bertrand and Mullainathan, 2004).

Exploratory analyses examine callback rate differentials by job type (white-collar vs. blue-collar) and geographic region (urban centers like Moscow and St. Petersburg vs. smaller cities).

Secondary outcomes are constructed as subgroup-specific callback rate differences (Δ\Delta\Delta
) for white-collar jobs (e.g., IT, administration) versus blue-collar jobs (e.g., logistics, manufacturing) and for urban versus smaller city job postings. These are calculated similarly to the primary outcome, using the same beta-binomial Bayesian framework, but stratified by job type and region. Due to the sample size (60 pairs), these analyses are exploratory and not powered for definitive inference, serving to identify potential heterogeneity in discrimination patterns for future research.

The study employs a two-arm matched-pair correspondence audit design with a Bayesian sequential approach. Fictitious résumés, identical except for military service signals, are submitted to job postings: the treatment group signals voluntary service in the Ukraine conflict ("Served during Special Military Operation, 2022–2023"), while the control group signals routine mandatory conscription ("Completed mandatory military service, 2018–2019"). The design compares callback rates to quantify the labor market penalty associated with voluntary, financially incentivized Ukraine service, isolating it from general military experience. Data are analyzed in batches, stopping when sufficient evidence of discrimination is detected.

The audit targets entry- and mid-level job postings on HeadHunter (hh.ru) and SuperJob (superjob.ru) across retail, logistics, IT, administration, and manufacturing sectors in Moscow, St. Petersburg, and smaller Russian cities. Each job posting receives a pair of résumés, randomized for applicant details (e.g., names, emails, phone numbers) but identical in qualifications (e.g., education, 3–5 years’ experience). The Bayesian sequential analysis processes data in 20-pair batches, modeling callback probabilities with Beta(1,1) priors and binomial likelihoods. The posterior difference in callback rates (Δ=pC−pU\Delta = p_C - p_U\Delta = p_C - p_U
) is estimated via Monte Carlo sampling (10,000 draws per batch). The study stops when P(Δ>0∣Data)>0.975P(\Delta > 0 \mid \text{Data}) > 0.975P(\Delta > 0 \mid \text{Data}) > 0.975 (indicating discrimination) or after 300 pairs. A secondary criterion, P(∣Δ∣<0.05∣Data)>0.90P(|\Delta| < 0.05 \mid \text{Data}) > 0.90P(|\Delta| < 0.05 \mid \text{Data}) > 0.90, tests for negligible effects. Robustness is assessed using Beta(2,2) priors and a linear probability model controlling for job type and region.

Randomization is performed in office by a computer, assigning applicant details (e.g., names, emails, phone numbers) and pairing treatment/control résumés to job postings.

The unit of randomization is the job posting, with each posting receiving one pair of résumés (one treatment, one control).

No

No
Explanation: The treatment (veteran vs. conscript résumé) is assigned at the individual résumé level within each pair, and pairs are independently assigned to job postings. There is no clustering by groups (e.g., firms or regions), as each job posting is treated as an independent unit.

300 job postings (each receiving one pair of résumés).

600 résumés (300 pairs, with each pair consisting of one treatment and one control résumé).

150 job postings for the treatment arm (veteran résumés) and 150 job postings for the control arm (conscript résumés), as each of the 300 job postings receives one pair of résumés (one treatment, one control).