Use when defending the research design of an American Political Science Review (APSR) manuscript — causal identification for quantitative work, case selection and process tracing for qualitative work, experimental and survey-experimental design, or formal-empirical linkage. APSR judges each tradition on its own terms. Strengthens the design; it does not write code.
NewMaintained
Automatic
◆Research pack
Last verified 8/15/2026
What it changes
Passed 19 of 22 test cases with the skill, and 19 of 22 without it.
Without skill86%
With skill86%
Checks it passed that the model alone missed
Rejects manual selection of control states.
Requires reporting proportional selection ratios for unobserved variables.
2 fixed17 passed either way1 still failing2 broke
Pass-rate lift
0 pp
Tokens
+15%
Agent turns
0%
Response time
-24%
Model gemini-3.6-flashMeasured 8/15/2026v119 of 22 cases passed with skill
Tokens, turns and time describe resource use, not a dollar price.
See it in action
One of 2 test cases the skill fixed. Same model, same prompt, with and without the skill.
1 of 2
Prompt
I have an observational cross-sectional design examining civil society density and local public goods provision. A reviewer noted that unobserved institutional quality could drive both. How should I incorporate sensitivity analysis into my research design plan?
What was checked, without and with the skill
Checked
Without skill
With skill
Recommends Oster's delta for evaluating unobserved confounder strength.
Passed
Passed
Requires reporting proportional selection ratios for unobserved variables.
Missed
Passed
case-09 · gemini-3.6-flash · 8/15/2026 · answers are capped and redacted
What's inside
The instructions your model reads when the skill loads.
SKILL.md85 linessingle file
Research Design (apsr-research-design)
APSR accepts many methodologies but is demanding about each. The design must credibly connect the
argument (apsr-theory-building) to evidence. This skill is mode-aware: pick the section that matches
your work and defend it against the strongest alternative explanation.
When to trigger
Specifying identification, case selection, or experimental design
A reviewer questioned causal claims, case choice, external validity, or a confound
Preparing a pre-analysis plan or a Registered Report Stage 1 design
Justifying why your design adjudicates the rival account from apsr-literature-positioning
Quantitative / causal inference
Identification first. State the estimand and the assumptions that license a causal reading
(ignorability, parallel trends, exclusion, continuity). Defend them, don't assert them.
Designs: experiments (incl. survey/conjoint), DID/event study (use modern staggered-adoption
estimators, not naive TWFE), IV (first-stage strength, exclusion, weak-IV-robust inference), RDD
(density/manipulation tests, bandwidth robustness), matching/weighting with balance + sensitivity.
Inference: cluster at the level of treatment assignment; randomization inference for
experiments; multiple-comparison adjustment when testing many implications.
Sensitivity: how strong must an unobserved confounder be to overturn the result?
Qualitative / case-based
Case selection justified by design logic (typical, deviant, most/least-likely, paired
comparison) — not convenience. Say what the case is a case of.
Process tracing with explicit tests (hoop, smoking-gun, straw-in-the-wind); state what evidence
would have disconfirmed the argument.
Source transparency: archives, interviews, fieldnotes — plan how they will be documented and
cited (see apsr-transparency-and-data-policy).
Experiments (lab / survey / field)
Preregister the design and primary analyses; report power/MDE; pre-specify subgroups.
Address attention/manipulation checks, attrition, and ethics/IRB and consent.
For survey experiments: sampling frame, treatment realism, and generalization claims.
Formal-empirical linkage
Make the empirical test follow from the model's comparative statics, not a loose analogy.
Distinguish predictions that are unique to your model from those shared with rivals.
Use when defending the research design of a British Journal of Political Science (BJPS) manuscript — causal identification for quantitative work, case selection and process tracing for qualitative work, experimental and survey-experimental design, or formal-empirical linkage. BJPS judges each tradition on its own terms. Strengthens the design; it does not write code.