isr-data-analysis

Analysis, Identification & Proof (isr-data-analysis)

When to trigger

Data are collected, or the model is built, and it is time to estimate, derive, or evaluate
You are unsure whether your estimator matches the design, or whether a proof is complete
Reviewers will probe identification, measurement validity, or assumption sensitivity
A reviewer says "the analysis does not support the inference"

Empirical genre — identification and validity first

ISR empirical reviewers expect causal claims to rest on a credible identification strategy, not on a fitted regression:

Design / claim	Estimator / strategy
Manipulated IT design/policy	Experiment: randomization checks, manipulation/attention checks
Quasi-experiment, staggered adoption	DiD (modern estimators), event study, parallel-trends evidence
Endogenous IT investment/adoption (archival)	IV/2SLS, RDD, matching, panel FE with cluster-robust SE
Latent behavioral constructs	SEM/CFA (fit: CFI/TLI/RMSEA/SRMR), AVE, discriminant validity; PLS-SEM where appropriate
Nested data (users in teams/firms/platforms)	Multilevel / HLM; cluster SEs to the sampling/nesting
Counts, choices, durations (clicks, churn)	Poisson/NB, logit/probit, hazard models as the DV demands

Address common-method bias by design first (separate sources/waves), then statistically (marker variable or unmeasured latent method factor — a Harman single-factor test alone is weak). Report effect sizes and practical magnitude, not only p-values.

Analytical genre — proof discipline

For modeling papers, "analysis" means correct, complete derivations: state the equilibrium concept, prove existence/uniqueness where claimed, and present the comparative statics as the substantive results with their IS interpretation. Run robustness as extensions that relax key assumptions (alternative information structures, costs, timing) and show which results survive. Full proofs and lemmas belong in the electronic companion, with the main text carrying the intuition and the load-bearing steps.

Design-science genre — rigorous evaluation

Demonstrate the artifact's utility: benchmarks against credible baselines, controlled user studies, or field deployment, with metrics tied to the stated design objectives. A demo is not an evaluation.

Claim-to-evidence ledger

Before writing results, create a ledger that binds every contribution claim to an analysis:

Claim type	Minimum evidence	Reviewer stress test
Causal empirical claim	Design logic, identifying assumptions, pre-trends/placebos or randomization checks, effect magnitude	What unobserved selection or timing story would overturn the claim?
Construct/measurement claim	Item provenance, reliability, CFA/discriminant validity, CMB defense	Would a different construct name or common-method explanation fit the data as well?
Analytical claim	Proposition, proof sketch in main text, full derivation in companion, comparative statics	Which assumption drives the result, and does an extension relax it?
Design-science claim	Baseline comparison, objective-linked metrics, user/field evidence where relevant	Is the artifact useful beyond the demonstration case?

If a claim lacks a row, downgrade the language before submission. ISR reviewers are receptive to careful boundaries; they are much less receptive to causal, theoretical, or design-utility claims that outrun the evidence.

Reproducibility and the electronic companion

ISR's source-backed compliance rule is data provenance certification: authors certify rights to use data and publish results, and any legal or corporate permissions must be obtained before submission. Regardless, keep clean scripts/solver inputs that regenerate every exhibit, and use the electronic companion for proofs, full measurement items, and supplementary analyses given the 32-page text / 38-page total caps.

Execution bridge (StatsPAI / Stata MCP)

Run the battery, don't just enumerate it. Full map: execution-with-mcp. ISR is empirical IS with strong econometric and experimental work; identification (DiD / IV) for observational claims, randomization inference for experiments.

Many outcomes / specifications: romano_wolf (step-down FWER) or benjamini_hochberg — report the adjusted threshold.
OVB sensitivity: oster_delta / sensemakr.
Inference: wild_cluster_bootstrap (few clusters), twoway_cluster / conley; multilevel data → cluster at the right level.
Re-fit off one handle: audit_result(result_id) lists the missing checks and the exact suggest_function for each.
Exhibits: etable / did_summary_to_latex from the handle — no retyped numbers.

Keep the decisive checks in the body and the exhaustive battery in the appendix. See the executed chain in the JF execution walkthrough.

Checklist

Empirical: identification strategy executed; assumptions/threats discussed
Measurement validity (reliability, CFA fit, AVE/discriminant) reported where latent constructs used
CMB addressed beyond a single-factor test; effect sizes reported
Analytical: equilibrium/existence stated; comparative statics interpreted; extensions show robustness
DSR: evaluation demonstrates utility against baselines/objectives
Claim-to-evidence ledger completed; no claim outruns the analysis
Proofs/measurement detail routed to the electronic companion

Anti-patterns

Regression-as-causal with no identification.
Single-factor CMB test as the sole defense.
Algebra dump with no economic/IS interpretation of the comparative statics.
Demo-not-evaluation for a design-science artifact.
Results-first writing that lists tables without saying which inference each table licenses.

Output format

【Genre】empirical / analytical / design-science
【Identification or proof】[...]
【Validity / robustness】CFA fit, AVE, CMB / extensions / baselines
【Effect size or comparative statics】[...]
【Electronic companion】proofs/items/supplements routed
【Open issues for reviewers】[...]
【Next step】isr-contribution-framing