TWO-SAMPLE TESTS FOR MULTIVARIATE REPEATED MEASUREMENTS OF HISTOGRAM OBJECTS WITH APPLICATIONS TO WEARABLE DEVICE...

Area of Science:

• Biomedical research
• Statistics
• Data science

Background:

• Repeated observations are common in biomedical and longitudinal studies, often using wearable sensors.
• Analyzing complex biosignal data (e.g., probability densities, histograms) across groups is challenging.
• Traditional statistical methods struggle with non-Euclidean data structures from repeated measures.

Purpose of the Study:

• To develop novel nonparametric, graph-based two-sample tests for multivariate object data with repeated measures.
• To evaluate differences in daily biosignal distributions across disease groups and demographics.
• To overcome limitations of traditional methods for complex, non-Euclidean data.

Main Methods:

• Proposed novel nonparametric, graph-based two-sample tests for object data with repeated measures.
• Treated repeatedly measured data as multivariate object data, removing assumptions on observation errors.
• Developed test statistics to capture various alternatives and derived their asymptotic null distributions.

Main Results:

• The proposed tests demonstrated substantial power improvements over existing methods.
• Type I errors were effectively controlled under finite samples, confirmed by simulation studies.
• Tests provided insights into location, inter-, and intra-individual variability in physical activity for mood disorder studies.

Conclusions:

• Novel nonparametric, graph-based tests are effective for analyzing complex repeated measures data in biomedical research.
• These methods offer significant advantages in statistical power and error control compared to traditional approaches.
• The tests provide valuable insights into biosignal variability across different groups, aiding in understanding disease-related differences.