Research Design (pubar-research-design)
PAR accepts many methodologies but is demanding about each. The design must credibly connect the
argument (pubar-theory-building) to evidence drawn from public organizations, bureaucrats, citizens,
or jurisdictions. 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 pre-registration (PAR offers pre-registration badges)
- Justifying why your design adjudicates the rival account from
pubar-literature-positioning
PAR design-fit gate
PAR is a generalist flagship, so the design must support both an academic claim and a usable public-
management implication. Start with this gate before polishing methods language.
| Claim type |
Design burden |
Practice-relevance check |
| Reform or mandate effect |
Assignment/timing logic, counterfactual trend, spillover check, and clustering at assignment level |
The finding changes how agencies time, target, or evaluate reforms |
| Managerial behavior |
Sample frame tied to real public managers or frontline staff, realistic decision task, and measured behavioral outcome |
The recommendation is feasible inside public organizations |
| Citizen response / public trust |
Treatment realism, representativeness limits, manipulation checks, and ethical framing |
The takeaway does not overgeneralize from survey preference to administrative behavior |
| Case/process account |
Case-selection logic, process-tracing tests, chronology, and rival-account evidence |
The lesson transfers to a defined class of agencies, programs, or jurisdictions |
| Mixed-method mechanism |
Quantitative association/effect plus qualitative implementation or mechanism evidence |
The qualitative strand explains what managers can act on, not just why results are interesting |
Quantitative / causal inference (public-management settings)
-
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 common in PA: DiD/event study around a reform or mandate (use modern
staggered-adoption estimators — Callaway–Sant'Anna, Sun–Abraham, BJS — not naive TWFE); IV
(first-stage strength, exclusion, weak-IV-robust inference); RD around eligibility/funding
thresholds; matching/weighting with balance + sensitivity.
-
Inference: cluster at the level of treatment assignment (often agency, district, or
jurisdiction); wild-cluster bootstrap when clusters are few (a recurring PA problem with state- or
agency-level treatments).
-
Sensitivity: how strong must an unobserved confounder be to overturn the result (Oster / E-value)?
Experiments on bureaucrats and citizens
- Preregister the design and primary analyses; report power/MDE; pre-specify subgroups.
-
Bureaucrat/managerial experiments: realism of the decision task, sample frame (which public
managers), and generalization to real administrative behavior.
-
Citizen survey/conjoint experiments: treatment realism, attention/manipulation checks,
attrition, and ethics/IRB and consent.
Qualitative / case-based & mixed methods
-
Case selection justified by design logic (typical, deviant, most/least-likely, paired
comparison) — not convenience. Say what the case is a case of (a reform, a governance form).
-
Process tracing with explicit tests (hoop, smoking-gun, straw-in-the-wind); state what evidence
would have disconfirmed the argument.
-
Mixed methods: say what the qualitative strand adds that the quantitative cannot (mechanism,
context, implementation), and where the two corroborate or diverge.
The adjudication test (PAR-specific)
For the single strongest rival explanation, write one sentence: "If the rival were true rather
than my argument, the agencies/managers/citizens would look like ___; instead they look like ___." If
you cannot, the design does not yet identify the contribution — and the practitioner takeaway is unsafe.
Practice-safe inference rules
-
Separate evidence from recommendation. A credible association may justify a diagnostic warning;
a causal design may justify a stronger managerial recommendation; neither automatically justifies a
universal policy prescription.
-
Name the implementation margin. If the intervention is staffing, training, targeting, rule
design, citizen communication, or interagency coordination, say which margin the design actually tests.
-
Check administrative feasibility. A design can be internally valid but still imply an action no
manager can implement. Flag cost, authority, data availability, and equity constraints.
-
Bound external validity. Identify the agency type, policy domain, country/state/local context,
and population to which the evidence should and should not travel.
-
Route transparency early. If the result relies on confidential administrative data, plan the
restricted-data path with
pubar-transparency-and-data before claims harden.
Reviewer stress tests
- Would the result survive if the strongest agency-level selection story were true?
- Is the treatment/exposure measured before the outcome and at the right organizational level?
- Are standard errors clustered at the assignment or sampling level, not merely the observation level?
- For qualitative work, what observation would have disconfirmed the mechanism?
- For mixed methods, do both strands answer the same claim, or are they two parallel papers?
- Can the Evidence for Practice box be written without making a claim the design cannot support?
Execution bridge (StatsPAI / Stata MCP)
Estimate and audit the design, don't only describe it. Full map:
execution-with-mcp. PAR is public administration — survey/observational and some experimental work; identification + clustered/multilevel inference, magnitude for practice.
-
detect_design → recommend → fit with as_handle=true → audit_result.
-
Observational causal claims: staggered DiD (
callaway_santanna / sun_abraham +
bacon_decomposition + honest_did_from_result); IV (effective_f_test +
anderson_rubin_ci); RDD (rdrobust + mccrary_test).
-
Experiments: randomization-based inference +
romano_wolf for many-outcome control.
-
Sensitivity:
oster_delta / sensemakr for observational claims.
Report the magnitude in interpretable units; route the full battery to the appendix. A
run end-to-end (synthetic data, real returns) is in the JF execution walkthrough.
Anti-patterns
- Naive TWFE on a staggered reform rollout; clustering below the assignment level
- "Causal" language (and a managerial recommendation) on a design that only supports association
- Convenience case selection dressed up as theory-driven
- Bureaucrat/citizen experiments over-generalized to real administrative behavior with no caveat
- A design that cannot distinguish your argument from the leading alternative
Output format
【Mode】quant-causal / experiment / qualitative / mixed
【Estimand or claim】what is being identified/shown
【Design-fit gate】academic claim + practice relevance supported? [Y/N]
【Key assumption(s)】and how each is defended
【Rival ruled out】the adjudication sentence
【Robustness/sensitivity】planned checks (clustering, few-cluster, Oster/E-value)
【Practice-safe inference】recommendation strength + implementation margin + external-validity boundary
【Transparency handoff】public / restricted / qualitative-controlled-access path
【Next】pubar-data-analysis
Supplementary resources
