● Phase 2 · Vignette Experiment
Field-Generated Context and Algorithmic Override
A controlled experimental study of how administrative reviewers integrate informal field observations with rule-based algorithmic recommendations in welfare eligibility verification.
Research Question
How do reviewers weight informal context?
When reviewing a case with a formal eligibility record, how do administrative decision-makers incorporate informal signals produced through prior field-level verification, and how does this influence their decision to follow, override, or defer an algorithmic recommendation?
Experimental Structure
A 2 × 2 within-subject design
The vignette set identifies three interacting components, balanced to break the confound between algorithmic recommendation direction and signal direction.
Each operator evaluates 12 cases drawn from the pool, with 6 control + 6 treatment enforced and order randomised. Cells are balanced 4/4/4/4 across the recommendation × signal direction factorial.
Grounding & Sources
Signals paraphrased from field observation and rule documentation
Treatment-arm signals are paraphrased from one of two sources: a corpus of 260 verifier-observed cases coded by the researcher during PMAY field work in Maharashtra (January 2026), and disqualification clauses documented in the 2018 SRA Government Resolution.
Applicant profiles are synthetic composites identified by case number; no names, addresses, or survey numbers corresponding to identifiable individuals are used.
- Cutoff-date proof & voter roll evidence
- Family property holdings and household composition
- Tenancy and continuous occupancy
- Documentation authenticity
- Alternate-property declaration integrity
- Sympathy framing (corpus-attested only)
Operator Position
Final-officer review layer
Operators evaluate cases at the Sub-Divisional Officer / Sakshama Pradhikari position — the layer responsible for final certification of the SRA Annexure-II eligibility document. Cases shown have completed prior junior-officer field verification.
Interpretive Focus
Rule vs. context
The design isolates how evaluators respond to information not captured by structured eligibility rules — testing whether overrides reflect contextual information absent from the record, rather than simple disagreement with the algorithm.
Data Capture
What the instrument records
| Layer | Variables |
|---|---|
| Operator | Initials, role, experience, locale |
| Design | arm, algo_recommendation, signal_direction |
| Decision (primary) | decision, override flag |
| Timing (primary) | response_time, time_to_first_action |
| Reasoning (exploratory) | brief written reasoning |
| Post-task | 3-item Likert salience survey |
| Independent review | blind second-review subset |
Override is defined as disagreement with the algorithmic recommendation — not as rejection of the applicant. Reasoning text is collected but treated as exploratory in the analysis plan.
Status
Field deployment
Instrument finalised. Field deployment May 2026 in Mumbai. Pre-registration is timestamped and publicly available. Findings will be reported regardless of direction, including null and heterogeneous results.