smj-methods

Research Design & Methods (smj-methods)

When to trigger

• Sample, time window, and unit of analysis are not yet pinned down
• Measures (especially of performance and of the strategic choice) are not validated
• You have not yet chosen an identification design for a causal claim
• You are deciding between quantitative panel work, a natural experiment, qualitative theory-building, or formal modeling

Design families SMJ publishes

The hallmark SMJ empirical paper uses panel data with fixed effects and then deploys one or more identification tools to address endogeneity. Choose the design from the threat, not from convenience.

SMJ's empirical standard is codified in its own editorial — Bettis, Gambardella, Helfat & Mitchell (2014), "Quantitative empirical analysis in strategic management," SMJ 35(7): 949–953 — which asks authors to acknowledge endogeneity and make a good-faith effort to address it, and disapproves of data snooping and p-hacking. SMJ also accepts that perfect causal inference is sometimes impossible; correlations that rule out alternative mechanisms can still be valuable if framed honestly.

Sample & unit of analysis

• State the population, sampling frame, time window, and the exact unit (firm-year, BU-year, alliance, deal). Mismatched units (theory at the firm level, data at the deal level) draw fire.
• Justify the panel length: enough pre-periods to test parallel trends if using DID; enough within-firm variation if using FE.
• Report and justify all sample-construction filters (survivorship, missing data, industry exclusions). SMJ reviewers probe sample-selection artifacts.

Measurement (the quiet rejection reason)

• Performance DV: pick the construct deliberately — accounting (ROA), market (Tobin's q, abnormal returns), or operational. Justify why it matches the theory; report sensitivity to alternatives.
• Strategic-choice IV: validate that the measure captures the construct (e.g., diversification, alliance scope) and not something correlated. Cite the source of any established measure.
• Controls: include theoretically motivated controls (firm size, age, slack, industry, prior performance) but avoid "bad controls" — outcomes of the treatment that absorb the effect.
• Pre-register, or at minimum pre-specify, the primary specification to avoid the appearance of specification search (SMJ disapproves of data snooping / p-hacking).

Identification design (decide before estimating)

State the identifying threat explicitly, then the design that defeats it:

• Selection into the strategic choice (firms that ally/acquire/diversify differ): matching, IV, selection model, or a shock that assigns the choice.
• Reverse causality (performance drives the choice): lagged structure plus a design that breaks simultaneity (shock/IV); lags alone are not enough.
• Unobserved heterogeneity (a firm trait drives both): firm fixed effects, plus discussion of time-varying confounds FE cannot absorb.
• Omitted environmental confounds: industry-year fixed effects, region controls.

Name the design here; smj-data-analysis executes and stress-tests it.

Execution bridge (StatsPAI / Stata MCP)

For the empirical / causal lane, estimate and audit rather than only specify. Full map: execution-with-mcp. SMJ is strategy — firm-level panels where strategic choices are endogenous; foreground IV / DiD identification and the endogeneity-of-strategy objection.

• detect_design → recommend → fit with as_handle=true → audit_result to enumerate the checks the design owes.
• Panel / 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 and romano_wolf for the many-outcome family-wise correction reviewers expect.

Match the toolchain to the reviewer pool, and report the effect size the venue wants. A run end-to-end (synthetic data, real returns) is in the JF execution walkthrough.

Checklist

• Population, frame, time window, and unit of analysis are explicit and matched to theory
• Every sample filter is reported and justified (survivorship / missingness addressed)
• Performance DV is justified; alternative operationalizations identified for robustness
• Strategic-choice measure is validated and sourced
• Controls are theory-driven; no post-treatment / "bad" controls
• The identifying threat is named and a design is chosen to address it
• For DID: pre-periods support a parallel-trends test; for IV: a candidate instrument with a relevance + exclusion story

Anti-patterns

• Choosing OLS-with-controls and hoping reviewers ignore endogeneity (they will not)
• Unit-of-analysis mismatch between theory and data
• Undisclosed sample filters that could drive the result
• A performance measure chosen because it "works," not because it fits the theory
• Controlling for a variable that is itself an outcome of the treatment
• Picking the design before identifying the threat it is supposed to solve

Output format

【Design family】panel FE | DID/natural experiment | IV | matching+DID | Heckman | qualitative | formal
【Sample】population / frame / window / unit
【Filters】[list + justification]
【DV (performance)】construct + measure + alternatives
【Focal X (strategic choice)】measure + source + validity note
【Identifying threat】selection | reverse causality | unobserved heterogeneity | omitted confound
【Chosen identification design】...
【Next step】smj-data-analysis

Templates & resources

• ../../resources/external_tools.md — strategy data sources and Stata/R/Python packages for FE, IV, DiD, matching, RDD
• ../../resources/official-source-map.md — SMJ endogeneity policy and the Bettis et al. (2014) methods editorial