The Impact of Functional Form Complexity on Model Overfitting for Nonlinear Mixed-Effects Models.

Area of Science:

• Biostatistics
• Statistical Modeling
• Growth Curve Analysis

Background:

• Nonlinear mixed-effects models (NLMEMs) are used for curvilinear growth patterns.
• Model selection criteria often fail to account for intrinsic nonlinearity, risking overfitting.
• Overfitting hinders generalizability and reproducibility in NLMEMs.

Purpose of the Study:

• Evaluate eight model selection criteria for NLMEMs.
• Assess sensitivity to overfitting related to functional form complexity.
• Identify criteria that effectively capture overfitting in intrinsically nonlinear models.

Main Methods:

• Conducted a Monte Carlo simulation study.
• Compared performance of eight distinct model selection criteria.
• Analyzed the impact of residual variance and sample size.

Main Results:

• Information criteria and stochastic information complexity criterion demonstrated superior performance.
• These criteria recovered the true model more frequently than average or conditional concordance correlation.
• Residual variance and sample size significantly impact NLMEMs model selection outcomes.

Conclusions:

• Certain information criteria and stochastic information complexity are recommended for NLMEMs.
• Researchers should consider functional form complexity in model selection.
• Findings offer guidance for applying NLMEMs and improving research reproducibility.