Wber Robustness

Robustness Strategy (wber-robustness)

When to trigger

• The headline result moves under reasonable alternative specifications
• A referee could question measurement quality (survey error, recall, attrition, undercoverage)
• Inference is shaky: few clusters, spatial correlation, multiple outcomes
• The robustness appendix is a long mechanical list with no logic
• You need to know which checks are load-bearing before submission

The WBER robustness philosophy

WBER referees are sophisticated about both econometric threats and the realities of developing-country data — surveys with recall and measurement error, administrative records with coverage gaps, sampling frames that miss the informal sector, attrition in panels. So robustness here has two axes: the standard identification-threat axis (does the estimate survive plausible violations of the design's key assumption?) and a data-quality axis (does the result survive how the data were actually constructed and measured?). Organize the section by threat, not by a checklist; each check should answer "if a skeptic believed X, would my conclusion change?"

Organize by threat, not by appendix

Data-quality robustness (the development-specific layer)

• Measurement: consumption, income, and yields in LDC surveys are noisy and often non-classically mismeasured (e.g., underreporting). Show the result survives alternative recall windows, deflators, or an independent data source.
• Coverage and frame: if the sampling frame misses the informal sector or remote areas, bound how much that could move the estimate.
• Currency/price comparability: when pooling across countries or years, show robustness to PPP conversion, deflator choice, and exchange-rate regime.
• Seasonality: agricultural and labor outcomes are seasonal; show timing of measurement does not drive the result.

Sequencing the robustness section

Order matters for how a WBER referee reads the section:

1. Lead with the design-violation sensitivity — the check that addresses the headline identifying assumption (honest-DiD, RD bandwidth, Oster δ). This is what the identification referee turns to first.
2. Then the data-quality checks — measurement, coverage, currency — the development-specific layer the policy referee scrutinizes.
3. Then influence and inference — leave-one-out, wild bootstrap, spatial-HAC, multiple testing.
4. Close with the specification curve — a single figure that says "the headline is modal, not cherry-picked."

State in the main text which one or two checks are load-bearing; relegate the mechanical remainder to the appendix (which still counts against the 40-page cap).

Execution bridge (StatsPAI / Stata MCP)

Run the battery, don't just enumerate it. Full map: execution-with-mcp. WBER is development economics — RCTs and observational designs in low/middle-income settings; randomization inference + DiD/IV, magnitude in policy units.

• Many outcomes / specifications: romano_wolf (step-down FWER) or benjamini_hochberg.
• OVB sensitivity: oster_delta / sensemakr.
• Inference: wild_cluster_bootstrap (few clusters), twoway_cluster / conley.
• Re-fit off one handle: audit_result(result_id) lists missing checks + the exact suggest_function for each.
• Exhibits: etable / did_summary_to_latex from the handle — no retyped numbers.

Decisive checks in the body, exhaustive battery in the appendix. JF execution walkthrough.