Jpart Data Analysis

Data Analysis (jpart-data-analysis)

JPART reviewers are methodologically sophisticated public-management scholars, and the journal requires authors to release the data and software code underlying the paper as a condition of publication (see jpart-transparency-and-data). Analyze as if a referee will re-run the code — because the materials are public. This skill covers execution and reporting; design lives in jpart-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 the mandatory data/code deposit

Analysis norms JPART expects

1. Report uncertainty and magnitude. Confidence/credible intervals and the substantive size of the effect (e.g., a fraction of an SD of PSM), not stars alone.
2. Robustness that probes, not decorates. Show specifications that could break the result (alternative measures of red tape/PSM, samples, estimators, fixed effects), and say what you learned.
3. Confront the PA-specific threats. Common-method/common-source bias, social desirability, and self-selection into public service are the objections raised first — address them, don't ignore them.
4. Heterogeneity with discipline. Pre-specify subgroups where possible; correct for multiple comparisons; do not mine for a significant interaction and theorize it post hoc.
5. Right inference. Cluster at the assignment/agency level; randomization inference for experiments; small-cluster corrections (wild-cluster bootstrap) when agencies are few.
6. Preregistration discipline. Separate confirmatory from exploratory analyses; reconcile any deviation from the plan and justify it.

Measurement (a perennial JPART referee focus)

• Validate constructs (PSM, red tape, goal ambiguity); report reliability; show the result is not an artifact of a single scale or coding choice. Concept defined in jpart-theory-building must match the measure used here.

Reproducibility while you work (not at the end)

• One master script regenerates every table and figure from raw/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 — the materials are public and will be checked.

What JPART reviewers probe, by design

Worked micro-example (illustrative numbers)

A hypothetical JPART field experiment tests whether a goal-clarity intervention raises frontline performance among real caseworkers. The pre-registered ITT is +0.18 SD (95% CI 0.06 to 0.30), randomization-inference p = 0.006. An exploratory split by tenure shows +0.41 SD for new hires, but it was not pre-registered and the interaction p = 0.03 before correction; after a Bonferroni adjustment across five exploratory subgroups it crosses 0.20. The disciplined write-up reports the confirmatory +0.18 SD effect with its interval and substantive meaning, flags the +0.41 figure as exploratory and not multiplicity-robust, and frames it as a hypothesis for future work. (All numbers illustrative.)

Referee-pushback patterns and the JPART repair

• "This is common-method bias, not an effect." → Use a separate/objective outcome or a marker variable; report the sensitivity, don't wave it away with a single Harman test.
• "The robustness table only reruns near-identical specs." → Replace decorative checks with specs that could break the result (alternative PSM/red-tape measures, samples), and say what held.
• "This is selection into public service." → State the estimand and assumption; report how strong an unobserved confounder must be to overturn it.
• "I cannot tell confirmatory from exploratory." → Segregate them explicitly; the deposited code is public, so the split must survive a re-run.

Calibration anchors (hedged)

• The bar is a public-management theory payoff carried by credible numbers — an estimate with no mechanism rarely clears JPART review.
• JPART increasingly rewards experimental and causal designs, but a rigorous multilevel or mixed study is judged on its own terms.
• The data-and-code release is mandatory (where ethically possible) — write the analysis so the public package reproduces every printed number. Confirm exact wording on the live policy page.

Execution bridge (StatsPAI / Stata MCP)

Run the battery, don't just enumerate it. Full map: execution-with-mcp. JPART is public management — observational and experimental designs on public organizations; identification + clustered/multilevel inference.

• 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.

Output format

【Main estimate】magnitude + interval + substantive meaning
【PA threat handled】common-method / selection — how?
【Robustness】specs that could break it → what held
【Heterogeneity】pre-specified? MHT-adjusted?
【Confirmatory vs exploratory】clearly separated?
【Reproducible】master script + seeds + pinned versions? [Y/N]
【Next】jpart-tables-figures

Supplementary resources

• ../../resources/code/ — Stata + Python estimation/inference skeleton
• ../../resources/external_tools.md — estimation, inference, and experiment packages
• ../../resources/official-source-map.md — data-and-code release policy