Uncertainty and Gendered Application Behavior: Evidence from a Field Experiment

Building on the naturally uncertain hiring setting, we randomly vary the content and timing of hiring-related information disclosed to job seekers who have expressed interest in the firm's advertised positions. We measure the effects on application continuation, search effort, document submission, and salary negotiation. Full details of the information conditions will be released upon trial completion

2026-08-03

2027-01-31

Continuation in the hiring process versus withdrawal following information disclosure (application rate on the extensive margin).
Asking salary: lower bound, upper bound, and median of the stated expected wage range.
Wage surplus: the proportional difference between the average industry wage and the expected median salary, (W − w)/W.

Because randomization occurs after pool formation, the primary application-behavior outcome is continuation in the hiring process (agreeing to proceed, submitting documents) versus withdrawal after information disclosure, rather than the initial decision to apply. Outcome variables take the value zero for pool members who withdraw post-intervention. The estimates capture the total effect of uncertainty across margins. The negotiation outcomes are elicited by asking salaries; the median of the stated upper and lower bounds is the main negotiation measure. The surplus benchmarks it against the average industry wage for the occupation. These outcomes are also evaluated writ gender differences under uncertainty: the Female × Treatment interactions are the central parameters of interest.

Search effort includes the time spent on the application (standardized), the proportion of the application/questionnaire completed, survey completeness, and proactive HR contact (calls/messages).
Expressed interest but documents not returned; display of dissent; withdrawal at later hiring stages.
Entry into negotiation (binary).

Search effort captures the intensive margin of application behavior after information disclosure. Document submission is operationally critical to the firm, which cannot bid for a tender if the proposed candidate withholds documents. Heterogeneity in all outcomes is examined along three pre-specified dimensions: (i) elicited ambiguity attitude (averse vs. neutral/seeking); (ii) gender interacted with the occupational gender composition of the job title; and (iii) prior awareness of or experience with tender-based/conditional hiring, proxied by survey elicitation and the duration of the last two jobs.

We collaborate with a small (20–99 employee) engineering consultancy firm in eastern India that hires project staff conditionally. Job titles reflecting the firm's hiring demand, spanning female-dominated, male-dominated, and gender-neutral occupations across multiple regions of India, are advertised across online job boards for at least 10 days each, with standard job descriptions and no mention of conditional hiring. From the resulting pool of interested candidates for each job title, individuals are randomly assigned to information conditions related to uncertain hiring conditions. Comparing each against control identifies the total effect of uncertainty disclosure. Outcomes on continuation, search effort, document submission, and negotiation are drawn from portal administrative data, the firm's recruitment records, and an applicant survey that also elicits ambiguity and risk attitudes, prior experience with conditional hiring, and expected salaries.

Not available

Randomization is done in office by a computer

Individual job seeker (within job-title pools)
Within each job-title applicant pool, individuals are randomly assigned in equal thirds to the three information arms, stratified by job title and by gender within each pool. This guarantees treatment balance for the primary Female × Treatment parameter. The disclosed probabilities in the risk arm derive from the firm's historical tender bid success record and short-term-to-permanent conversion rate, which the firm documented prior to randomization.

No

Not applicable in the clustered-treatment sense (individual-level randomization). Stratification units: 6 job-title pools across regions.

Approximately 8,100 individual job seekers in total across six job titles (≈ 1,350 per title).

The overall sample size is approximately 2,700 per arm, resulting in 450 per arm within each job title.

Recruitment spans six job titles selected to vary in occupational gender composition (female-dominated, gender-neutral, male-dominated). A target total pool of approximately 8,100 candidates (≈ 1,350 per title) is randomized individually in equal thirds within each title (n ≈ 2,700 per arm overall; ≈ 450 per arm within each title). Job titles serve as randomization strata and enter the analysis as fixed effects. Assuming a control-group continuation rate of 50% (conservative variance), α = 0.05 and power = 0.80, the minimum detectable effect for pairwise arm comparisons on binary outcomes is 3.8 percentage points (3.3 pp for the pooled uncertainty contrast J1+J2 vs. J3), and 8.3–7.6 percentage points for the Female × Treatment interaction for expected female pool shares of 30–50%. For continuous outcomes (standardized asking salary), the corresponding MDEs are 0.08 SD for pairwise main effects and ≈ 0.15–0.17 SD for the gender interaction across the same range of female pool shares. Within occupational categories (two titles each, n ≈ 900 per arm), main-effect MDEs are ≈ 6.6 pp and Female × Treatment MDEs ≈ 13.5 pp in female-dominated and ≈ 16.5 pp in male-dominated categories. Category-level interaction tests are therefore powered only for large gaps and are designated secondary. The pooled uncertainty contrast (J1+J2 vs. J3) is the primary comparison while the ambiguity-vs-risk decomposition and gender interactions within occupational categories are secondary comparisons.