Disclosure and Hiring under Multidimensional Applicant Quality: A Laboratory Experiment

We contrast an environment with positive correlation (where strong test performance aligns with employer value) and one with negative correlation (where the alignment is partial).

2026-05-16

2026-12-31

The primary outcomes are:
(i) block-level test uptake for types B and C, and
(ii) block-level acceptance rates following Opt-Out.

We have two primary hypotheses, corresponding to two distinct behavioral margins: Applicants’ disclosure decisions and Employers’ hiring responses following Opt-Out. Both primary tests use blocks as the unit of observation. Unless otherwise stated, the main analysis window is the last 20 periods of the experiment, that is, games 21–40.

H1a. The primary outcome is block-level test uptake for types B and C in the main analysis window. Because each block contains one Applicant of each type in every period, each block contributes exactly one observation of type B and one observation of type C in each of periods 21–40. For each block, we calculate U_BC as the share of B- and C-type observations in games 21–40 for which the Applicant chooses to take the test. This is the total number of test choices by B and C types in games 21–40, divided by 2 × 20 = 40. We compare the distribution of this block-level measure across treatments using a two-sided Wilcoxon rank-sum test, treating blocks as independent treatment observations.

H1b. The primary outcome is the block-level acceptance rate following Opt-Out in the main analysis window. For each block, we calculate E_Opt-Out as the number of cases in games 21–40 where the Employer hires after observing Opt-Out, divided by the total number of Opt-Out observations in that block during games 21–40. We compare the distribution of this block-level measure across treatments using a two-sided Wilcoxon rank-sum test, again treating blocks as independent treatment observations. Blocks with no Opt-Out observations in the main analysis window are excluded from this comparison and reported separately.

All-game robustness analysis. As a pre-specified robustness check, we repeat the primary and secondary analyses using all 40 games.

Secondary outcomes include overall test uptake across all Applicant types, Employer hiring decisions by signal category (High, Low, Opt-Out), and type-conditional test uptake rates for all four types $(A,B,C,D)$.

We contrast an environment with positive correlation (where strong test performance aligns with employer value) and one with negative correlation (where the alignment is partial).

Not available

Participants are recruited from the same poole of students: Participants do not know anything about the experiment before the session starts.

Subjects are pre-assigned to blocks of 8 (4 Applicants, 4 Employers). Random rematching occurs only within a block each game. Roles are fixed at session start. Applicant type order is a random permutation with 10-game blocks.

Yes

We plan to have 8 blocks per treatment for a total of 16 cluster.

Our planned sample of 8 blocks per treatment arm corresponds to 64 participants per arm and 128 participants in total.

We plan to have 8 blocks per treatment arm.

For H1a, our primary Applicant-side test compares block-level mean test uptake for types B and C across treatments, using the main analysis window (games 21–40), with a two-sided Wilcoxon rank-sum test at the 5 percent significance level. For H1b, power depends on how often Opt-Out occurs, especially in the Positive treatment where Opt-Out may be rare. We therefore do not provide a separate formal power calculation for H1b, but we will report the number of blocks with Opt-Out observations in each treatment. The standard deviation used in the power analysis is based on a pilot conducted at BI Norwegian Business School on March 18, 2026. The pilot included both a Positive and a Negative treatment, and we use the variance estimated from these pilot treatments as the basis for the calculations. Based on the pilot data from the final 20 rounds of the experiment, we estimate a pooled standard deviation of about 15.8 for our primary outcome variable, U_BC. With our planned sample size of 8 independent matching blocks per treatment arm, and assuming a standard deviation of about 15.8 and a minimal detectable difference of 25, this yields statistical power of approximately 87 percent, using the Pitman asymptotic relative efficiency adjustment for the Wilcoxon rank-sum test.