Goodness of Fit Tests for Linear Mixed Models

Area of Science:

• Statistics
• Biostatistics
• Longitudinal Data Analysis

Background:

• Linear mixed models (LMMs) are essential for analyzing clustered or correlated data.
• Assessing the model fit, particularly the fixed effects, is critical for reliable statistical inference.
• Existing methods lack confirmatory goodness-of-fit tests for the mean structure of LMMs against general alternatives.

Purpose of the Study:

• To propose a novel class of goodness-of-fit tests specifically designed for the mean structure of LMMs.
• To provide a confirmatory statistical tool for evaluating the adequacy of fixed effects in LMMs.
• To address the unmet need for assessing LMMs against broad, unspecified deviations.

Main Methods:

• Developed a test statistic based on the quadratic form of differences between observed and expected values within covariate space partitions.
• Demonstrated the asymptotic chi-squared distribution of the test statistic under maximum likelihood, least squares, and method of moments estimation.
• Evaluated the test's power analytically and through simulations under local alternatives.

Main Results:

• The proposed goodness-of-fit test statistic asymptotically follows a chi-squared distribution.
• The test demonstrates power under local alternatives, indicating its ability to detect model misspecification.
• The methodology was successfully illustrated using real-world data from Chernobyl exposure studies.

Conclusions:

• The introduced goodness-of-fit tests offer a valuable confirmatory approach for the fixed effects in LMMs.
• These tests enhance the validity of inference in regression analyses involving complex data structures.
• The proposed methods are applicable to various estimation techniques and provide a practical tool for applied researchers.