Analysis of repeated measurements using nonparametric smoothers and randomization tests

Area of Science:

• Statistics
• Biostatistics
• Data Analysis

Background:

• Traditional mixed-model analysis of variance (ANOVA) for repeated measures has limitations.
• It relies on specific error distribution assumptions and ignores the temporal order of data.
• These limitations can reduce analytical power and introduce potential inaccuracies.

Purpose of the Study:

• To present an alternative statistical procedure for analyzing longitudinal data.
• The method aims to overcome the disadvantages of mixed-model ANOVA.
• It seeks to preserve the familiar framework of mixed-model ANOVA while enhancing its capabilities.

Main Methods:

• Nonparametric smoothing is used to estimate group mean profiles from observed data.
• Randomization tests are employed to assess group main effects, time main effects, and group-by-time interaction effects.
• F-test approximations are derived from existing results (Zerbe, 1979) for group and interaction effects, and a novel approximate F-test is proposed for time effects.

Main Results:

• A simulation study confirmed the good performance of the approximate F-tests.
• Nonparametric smoothing was shown to increase the statistical power for detecting time main effects and group-by-time interactions.
• The procedure was successfully applied to analyze hormone level data in cows.

Conclusions:

• The proposed procedure offers an improved statistical approach for analyzing repeated measures over time.
• It effectively utilizes the temporal ordering of data and relaxes restrictive distributional assumptions.
• The method enhances statistical power and provides a robust alternative to traditional mixed-model ANOVA.