pubar-data-analysis

Data Analysis (pubar-data-analysis)

PAR reviewers are methodologically capable public-management scholars, and the journal endorses the TOP transparency guidelines — so analyses should be reproducible and documented (see pubar-transparency-and-data). Because PAR articles carry Evidence for Practice, every estimate that drives a managerial takeaway must be analyzed honestly enough to bear that weight. This skill covers execution and reporting; design decisions live in pubar-research-design.

When to trigger

• Running main and supporting analyses; building the results section
• A reviewer asked for robustness, heterogeneity, or alternative specifications
• Reconciling preregistered vs. exploratory analyses
• Making the analysis reproducible before deposit

Analysis norms PAR expects

1. Report uncertainty honestly. Confidence/credible intervals, not just stars; the magnitude and substantive/managerial meaning of the estimate, not just its significance. A practitioner needs effect size, not a p-value.
2. Robustness that probes, not decorates. Show specifications that could break the result (alternative measures, samples, estimators, fixed effects), and say what you learn.
3. Heterogeneity with discipline. Pre-specify subgroups where possible (agency type, jurisdiction size, sector); correct for multiple comparisons; don't mine an interaction and theorize it post hoc.
4. Right inference. Cluster at the assignment/sampling level (agency, district); wild-cluster bootstrap when clusters are few — a common public-management data situation.
5. Preregistration discipline. Clearly separate registered from exploratory analyses; reconcile and justify deviations.
6. Measurement. Validate constructs (red tape, PSM, performance); report reliability; show results are not an artifact of a coding/scaling choice — measurement debates are central in PA.

Mixed-methods integration

• State explicitly where the qualitative evidence corroborates, refines, or contradicts the quantitative estimate; do not present them in parallel silos with no integration.

Reproducibility while you work (not at the end)

• One master script regenerates every table and figure from the (raw or constructed) data.
• Set and report seeds for bootstrap, randomization inference, simulation, any stochastic step.
• Pin software/package versions (renv.lock, requirements.txt, recorded ssc/net installs).
• Keep table/figure numbers matched to script outputs; document design/prep decisions in the supplementary document PAR recommends.

Anti-patterns

• Stars-only tables with no effect sizes or intervals (a practitioner can't act on stars)
• "Robustness" that only reruns near-identical specs to manufacture stability
• p-hacking / fishing for a significant interaction; HARKing exploratory results into hypotheses
• Clustering at the wrong level or ignoring few-cluster problems
• An Evidence-for-Practice point that the analysis does not actually support

Output format

【Main estimate】magnitude + interval + managerial meaning
【Identification check】(per research-design) result
【Robustness】specs that could break it → what held
【Heterogeneity】pre-specified? MHT-adjusted?
【Registered vs exploratory】clearly separated?
【Reproducible】master script + seeds + pinned versions? [Y/N]
【Next】pubar-tables-figures

What PAR reviewers probe, by analytic tradition

Analytic tradition	The check a PAR referee runs first	The fix that earns the benefit of the doubt
Survey / managerial experiment	Is inference randomization-based and pre-registered?	Randomization inference, pre-registered estimand, MDE reported
Observational causal (reform)	Is the "causal" word (and the policy advice) doing more than the design licenses?	State estimand + assumption; sensitivity to an unobserved confounder
Performance / administrative data	Are measures validated, and is gaming/selection ruled out?	Construct validation, reliability, selection checks
Mixed methods	Do quant and qual estimates actually corroborate?	Show where they agree, and own where they diverge

Worked micro-example (illustrative numbers)

A hypothetical PAR survey experiment tests whether a performance-feedback framing raises frontline managers' willingness to adopt a new reporting tool. The pre-registered ATE is +7.4 points (95% CI 3.0 to 11.8) on a 0–100 willingness scale, randomization-inference p = 0.006. An exploratory subgroup ("low-tenure managers") shows +13 points, but it was not pre-registered and after a Bonferroni adjustment across five exploratory subgroups its interval crosses zero. The disciplined write-up reports the +7.4 confirmatory effect with its interval and a managerial interpretation, flags the +13 figure as exploratory and not multiplicity-robust, and frames it as a hypothesis — so the Evidence-for-Practice point rests on the confirmatory estimate only. (All numbers illustrative.)

Calibration anchors (hedged)

• The bar is field-wide PA significance plus honest practice relevance; an effect only a specialist values, or a takeaway the data can't support, rarely clears PAR review.
• PAR practices methodological breadth — a rigorous mixed-methods or case analysis is not second-class to a regression. Match the inference standard to the design.
• TOP transparency expectations evolve; confirm the current data-policy wording on the journal's page (检索于 2026-06；以官网为准).

Supplementary resources

• ../../resources/external_tools.md — estimation, inference, and survey packages
• ../../resources/official-source-map.md — TOP transparency policy