Sample Size Requirements GMM

I am MedXY medical AI and can only answer questions related to medicine and health. I may not be able to answer your question accurately. Thank you.

Your synthesis of the methodological literature is **substantially correct**. A minimum sample size of **N ≥ 200** is a widely cited heuristic for growth mixture models (GMM) and latent class growth analysis (LCGA), and the specific thresholds you cite (BIC-based selection, >80% accuracy for three classes) are consistent with the simulation-based evidence retrieved. ### Evidence Supporting N ≥ 200 The retrieved literature consistently supports this benchmark: | Source | Key Finding | Citation | |--------|-------------|----------| | Liu et al. (methodological study) | When BIC is the primary model selection criterion, **N ≥ 200** is recommended | [1] | | Kim (Monte Carlo simulation) | At N = 200, accuracy in identifying **three trajectory classes exceeds 80%** | [1] | | TRAJECT project protocol | Simulations suggest samples starting from **N = 200** (and even N = 50) may be sufficient for detecting meaningful subgroups | [2] | | Pregnancy role adaptation protocol | **N ≥ 200** is generally adequate to ensure parameter convergence and correctly identify the true number of latent classes, with **statistical power > 0.80** to detect the smallest latent class at ≥10% prevalence | [3] | ### Important Caveats (Methodological Nuance) While N ≥ 200 is a reasonable **minimum**, the retrieved evidence also highlights several factors that may require a larger sample: 1. **No universal standard**: There is no single agreed-upon minimum sample size for GMM/LCGM across all contexts [2]. The requirement depends on multiple factors including: - Number of time points - Class separation (entropy) - Prevalence of the smallest class - Model complexity (number of parameters) 2. **Higher N for complex models**: Some simulations require "very high sample sizes" to draw statistically significant conclusions [2]. 3. **Attrition adjustment**: The protocols consistently add a buffer for attrition (10–20%), meaning the **target recruitment** should exceed 200 to ensure ≥200 valid cases remain for analysis [1][3]. ### Conclusion **Yes, your conclusion is correct.** A minimum sample size of **N ≥ 200** is an evidence-based benchmark for growth mixture models when using BIC-based model selection and aiming to identify three trajectory classes with >80% accuracy. However, this should be considered a **lower bound**; the final sample size should be adjusted upward based on anticipated attrition, model complexity, and the expected prevalence of the smallest latent class. --- *This response is based on retrieved methodological literature and simulation studies. For specific study planning, consultation with a biostatistician is recommended to account for study-specific parameters (number of time points, expected class separation, and attrition rate).*