โ† All posts

How to Choose the Right Correlation-Corrected Statistic: Model Hierarchy for Repeated Measures (Best โ†’ Fallback)

Clinical Epidemiology ResearchUniqcret doctor knowledgesData Analytics or StatisticsMethodology and Research Design
On this page

๐Ÿ“Š Model Hierarchy for Repeated Measures (Best โ†’ Fallback)

RankModelDescription
๐Ÿฅ‡ 1Conditional Model โ€” Multi-level โ€” Random EffectSubject-specific; captures both random intercepts and slopes
๐Ÿฅˆ 2Conditional Model โ€” Multi-level โ€” Fixed EffectSubject/group-specific using dummy variables; no generalization
๐Ÿฅ‰ 3Marginal Model โ€” Single-level โ€” Model-Based VariancePopulation-averaged; requires correlation structure assumption
๐Ÿช™ 4Marginal Model โ€” Single-level โ€” Empirical (Robust SE)Uses sandwich SE; structure-free fallback


๐Ÿ” Downgrade Pathway (When Data is Limited)

LimitationDowngrade to...Reason
No group ID (e.g. id, cluster)โŒ Can't use Conditional โ†’ Use MarginalCan't define within-subject correlation
Few groups (<5)โŒ Random โ†’ โœ… Fixed EffectRandom effects may overfit
Insufficient data for random slopesโŒ Random slopes โ†’ โœ… Random intercept or FixedModel may not converge
Unknown correlation structureโŒ Model-based โ†’ โœ… Empirical (robust SE)Robust to mis-specification


โš”๏ธ 1. Random Effect vs Fixed Effect (Conditional Multilevel)

CriterionFixed EffectRandom Effect
Uses dummy variables?โœ… YesโŒ No (learns from distribution)
Can predict new/unseen groups?โŒ Noโœ… Yes
Requires all groups in model?โœ… YesโŒ No
Learns group-level variance?โŒ Noโœ… Yes (estimates variance components)
Allows random slopes?โŒ Noโœ… Yes (e.g., `
Suitable for few groups?โœ… YesโŒ Not stable with few clusters
Stata commandsxtreg, fe, reg outcome i.group`mixed outcome ...


โš”๏ธ 2. Fixed Effect vs Marginal Model-Based

CriterionFixed EffectMarginal (Model-Based)
Focus on individual/group effect?โœ… YesโŒ No (population-averaged)
Can predict individual response?โœ… Limited to included groupsโŒ Not designed for this
Command in Stataxtreg, fe, areg, reg i.groupxtgee ... corr(exchangeable)
Population-level interpretation?โŒ Not primary aimโœ… Yes
Needs large sample for stability?โŒ Less dependentโœ… More sensitive to cluster size
Suitable for few clusters?โœ… YesโŒ Risky, may produce unstable estimates


โš”๏ธ 3. Marginal Model-Based vs Empirical (Robust SE)

CriterionModel-Based VarianceEmpirical (Robust SE)
Requires correlation structure?โœ… Yes (e.g., AR1, exchangeable)โŒ No
Sensitive to mis-specification?โŒ Yesโœ… No (SE remains valid)
Uses sandwich estimator?โŒ Noโœ… Yes
Higher power when correct?โœ… YesโŒ Lower power if model structure unknown
Best for...Known structure, more clustersSmall samples or uncertain correlation
Stata commandsxtgee ... corr(ar1)xtgee ... vce(robust)


๐Ÿ”š Final Decision Strategy

Start with:

  1. Mixed Model (Random Intercept/Slope)โ†’ if not feasible due to convergence/small N
  2. Fixed Effects Modelโ†’ if still not feasible
  3. GEE with correlation structureโ†’ if structure is unclear or unreliable
  4. GEE with robust (Empirical) SEย โ† safest fallback

Let me know if you want side-by-side Stata syntax, visual simulations, or a clinical use case example.

0
Message for International and Thai ReadersUnderstanding My Medical Context in ThailandRead more โ†’Message for International and Thai ReadersUnderstanding My Broader Content Beyond MedicineRead more โ†’

Comments

No comments yet. Be the first to share your thoughts.

Sign in to comment