thedixitjain Multi-Country Research Design (jim-methods)
When to trigger
- Countries, samples, or data sources are being chosen for a JIM-bound study
- Scales developed in one language are about to be fielded in others
- A cross-cultural experiment or an export panel is on the table
- A reviewer asks why these countries, or whether samples are comparable
Country selection is a theoretical act
Countries are your levels of the theoretical variable — pick them the way an experimentalist picks conditions:
- Theory-driven contrast. Choose countries that sit far apart on the focal dimension (e.g., high vs. low uncertainty avoidance; strong vs. weak contract enforcement) while as similar as possible on rivals. Two well-chosen countries beat six convenient ones.
- Confound audit. For every focal dimension, list the country characteristics that co-vary with it (income, language family, region, market maturity) and state how the design or the models separate them.
- Many-country designs (10+). Move from contrast logic to variable logic: measure the country dimension continuously, plan multilevel estimation, and check that the country sample spans the dimension's range rather than clustering at one pole.
- Justify the count either way: with 2–4 countries, country-level "effects" are illustrations, not tests; say so honestly and lean on theory-driven contrast.
Equivalence before comparison — the JIM discipline
Cross-national comparison is meaningless unless the instrument travels. Build equivalence into the design, in this order:
- Construct equivalence. Does the construct exist and mean the same thing in every country? Qualitative pre-work (interviews, pilot focus groups) is cheap insurance; an emic construct forced into an etic scale fails later at the latent level.
- Translation equivalence. Committee translation plus back-translation by independent bilinguals; reconcile discrepancies formally; pretest each language version. Document the protocol — JIM reviewers ask.
- Sampling equivalence. Match samples across countries on the frame (students vs. panel vs. probability), demographics, and recruitment channel. A U.S. Prolific sample against a Chinese student sample confounds country with everything else.
- Measurement invariance plan. Pre-commit the MGCFA sequence — configural → metric → scalar — and the decision rules (ΔCFI ≤ .01 alongside χ² difference), before fielding. Plan for partial invariance fallbacks and, with many groups, the alignment method. Steenkamp and Baumgartner (1998, JCR) remains the reference protocol; execution lives in
jim-data-analysis. - Response-style protection. Acquiescence and extreme-response styles differ systematically across cultures. Design against them (balanced keying, some anchoring vignettes or forced-choice items where feasible) and plan statistical controls.
Design lanes
Multi-country survey (the JIM staple)
Everything above, plus: common-method-variance protection (temporal separation, marker variable) in each country; a priori power in the smallest country sample; informant-quality screens for firm-level surveys (export managers who actually make the decision).
Cross-cultural experiment
Location is not a manipulation. Either (a) manipulate the cultural mechanism directly (e.g., prime self-construal) and show country moderates as theorized, or (b) measure the individual-level cultural orientation and treat country as the macro layer. Stimuli must be equivalence-checked (brands, prices, and scenarios pretested for familiarity/realism per country); randomize within country; power the interaction.
Export / entry-mode secondary data
Firm-level export panels, customs data, subsidiary databases, or matched country statistics (World Bank, WTO, Euromonitor-type sources). The gate is identification: exporting and entry-mode choices are endogenous strategy decisions. Name the strategy — firm fixed effects with within-firm variation, DiD around policy shocks (tariff changes, FTA entry), IV, or selection models (export-market entry is selected) — and defend its key assumption. Cluster inference at the country or firm level to match the variation.
Meta-analysis of cross-national effects
Code country context (dimension scores, development indicators) for every primary study a priori; model them as moderators; report search protocol and inter-coder reliability; publication-bias diagnostics are expected.
Execution bridge (StatsPAI / Stata MCP)
For quasi-experimental and panel lanes, run the design checks rather than merely listing them. Full map: execution-with-mcp. Typical JIM chain: detect_design → recommend → fit with as_handle=true → audit_result for the owed diagnostics — staggered-policy DiD via callaway_santanna plus honest_did_from_result, IV via effective_f_test, few-country clustering via wild_cluster_bootstrap. Invariance and multilevel execution details are in jim-data-analysis.
Checklist
- Country choice justified on the focal dimension; confound audit written
- Construct, translation, and sampling equivalence documented (protocols, not assertions)
- Invariance sequence and decision rules pre-committed; partial-invariance fallback planned
- Response-style and CMV protections designed in per country
- Experiment: mechanism manipulated/measured, stimuli equivalence-pretested, interaction powered
- Secondary data: identification strategy named; endogeneity of the international choice addressed
- Smallest-country sample passes the power analysis
Anti-patterns
- Countries chosen by coauthor passports, with the cultural rationale reverse-engineered
- One-shot single translation with no back-translation record
- Comparing latent means without any invariance testing planned — a desk-reject trigger at JIM
- Treating data-collection location as a cultural manipulation
- Export-performance regressions that ignore self-selection into exporting
- Pooling countries into one sample and calling the study cross-national
Output format
【Design lane】multi-country survey / cross-cultural experiment / secondary panel / meta
【Countries】list + focal-dimension contrast + confound audit result
【Equivalence】construct / translation / sampling: protocol status for each
【Invariance plan】configural→metric→scalar, ΔCFI rule, partial fallback: committed?
【Identification (if secondary)】strategy + key assumption
【Power】smallest-country sample vs. target effect: pass/fix
【Next skill】jim-data-analysis
Source: brycewang-stanford/Awesome-Journal-Skills → Journal-of-International-Marketing-Skills/skills/jim-methods/SKILL.md