Organizes robustness checks by threat for World Development manuscripts. Guides quantitative sensitivity analysis (specification, sample, inference, attrition) and qualitative trustworthiness (triangulation, negative-case analysis).
How this skill is triggered — by the user, by Claude, or both
Slash command
/world-development-skills:worlddev-robustnessThe summary Claude sees in its skill listing — used to decide when to auto-load this skill
- The headline result moves under plausible alternative specifications
The single biggest WD robustness failure is a wall of appendix tables with no logic. A WD referee — often from a different discipline than the author — wants to see that you identified the threats to your specific claim and addressed each one. Structure the robustness work as a short list of named threats, each with the check that retires it and a one-line verdict. For each threat: what would break the claim, what test isolates it, what the test shows.
| Threat | Check |
|---|---|
| Specification dependence | Add/drop controls in a disciplined sequence (Oster-style δ/bounds); specification curve if the literature is unsettled |
| Sample / outlier dependence | Drop influential units, regions, or waves; leave-one-out; trim |
| Measurement error (acute in development data) | Alternative measures; validation against an independent source; bounds |
| Inference fragility | Cluster at the right level; few-cluster wild bootstrap; spatial (Conley) SEs; randomization inference for RCTs |
| Selection / attrition | Lee bounds; selection models; characterize who exits |
| Multiple hypotheses | Romano–Wolf or sharpened q-values across the family of outcomes |
| Mechanism vs. confound | Show the proposed mechanism's footprint; rule out the leading alternative explicitly |
Run the checks the threat justifies — not the full menu. A paper that reports forty robustness tables but never addresses the obvious confound has gold-plated the wrong corner.
Robustness for qualitative WD work is trustworthiness, and it is judged, not waived:
Show the strands converge or that divergence is informative. When quant and qual disagree, that tension is data — explain it rather than hiding the weaker strand.
【Journal】World Development (WD)
【Skill】worlddev-robustness
【Verdict】robust / fragile / mixed
【Threats addressed】[threat → check → verdict] for each
【Leading alternative】how it is ruled out
【Qual trustworthiness】triangulation / negative cases / audit trail (if applicable)
【Policy-relevant heterogeneity】shown / hidden
【Source status】verified URL / 待核实 / not asserted
【Next skill】worlddev-tables-figures
npx claudepluginhub brycewang-stanford/awesome-journal-skills --plugin world-development-skillsOrganizes robustness checks for World Bank Economic Review manuscripts by identification threat and developing-country data-quality risks, helping structure a logical appendix rather than a mechanical checklist.
Stress-tests the inferential logic of a World Development manuscript against the journal's pluralist standard, covering quantitative causal designs (DID, IV, RDD, RCT) and qualitative inference (process tracing, comparative case logic).
Organizes robustness and sensitivity checks for JLEO manuscripts by targeting institutional threats (selection, endogeneity, measurement, alternative mechanisms) with specific defusal tests.