# Recruiting study memo — two-page template

**Author:** [name]  
**One intervention:** [documented test override / reviewer capacity / randomized interview ordering / referral program / marginal trial invitations]  
**Economic question:** [one sentence naming the organizational choice and outcome]

## 1. Population, risk set, and decision

Define the initial applicant or vacancy cohort, inclusion rules and follow-up horizon. Name the treatment-assignment unit. Describe the rule before and after the intervention and the information available when each decision occurs. Distinguish the intervention from take-up, realized referrals or an evaluator's later yes/no choice.

## 2. Outcomes and event ledger

Choose one primary outcome observed for the baseline population. Fill a compact ledger; add only stages needed for your question.

| Event/outcome | Unit and risk set | Observation horizon | Data producer | Missingness or censoring rule |
|---|---|---|---|---|
| Baseline cohort | [all initially eligible applicants/vacancies] | [date range] | [source] | [rule] |
| Intervention/exposure | [assignment unit] | [time available] | [source] | [rule] |
| Primary outcome | [baseline denominator] | [common horizon] | [source] | [rule] |
| Conditional transition | [explicit selected denominator] | [horizon] | [source] | [rule] |
| Hire-only performance | [who is observed] | [horizon] | [source] | [missing for whom?] |

Explain why conditional transitions are descriptive or require further assumptions for a causal interpretation. State which counterfactual performance outcomes are unavailable for rejected applicants. Do not replace missing performance with zero.

## 3. Credible comparison and implementation

Describe one randomization or defensible comparison and its strongest threat. For scheduling, define the pool, ordering rule, independent candidate information and boundaries; prospective event-time requirements are a new data request, not a claim about an existing archive. For rollout, explain timing and comparison assumptions. For exploration, record known invitation probabilities and the outcome actually measured by the trial.

Name the independent assignment/dependence units and report how many the proposed study could provide. Address capacity displacement, common evaluators, shared vacancies or spillovers. State the inference method and what clustering cannot repair. Do not invoke weighting or cross-fitting without support and credible assignment.

## 4. Data access, selective labels, and conclusion

List the minimal records, linking keys, timing fields and access status: public, restricted, or requiring a partnership. Do not call public code original microdata. Identify which rejected-candidate outcomes remain unknown and any additional assumption required for extrapolation. If considering selection bounds, defend individual one-direction monotonicity; stable hiring rates alone are insufficient.

End with one limited economic conclusion the proposed comparison could support, and one conclusion it could not. Keep tenure separate from productivity and firm outcomes separate from worker welfare. Do not label historical eventual hiring as universal productivity or retrospective evaluation as a prospective productivity RCT.

**Reproducibility note:** [command, seed, files inspected, synthetic/empirical mode, and deviations from the supplied lab]