Modelling count, bounded and skewed continuous outcomes in physical activity research: beyond linear regression...

Area of Science:

• Statistics
• Biostatistics
• Physical Activity Research

Background:

• Standard linear regression models (LMs) have assumptions rarely met in practice.
• Violated LM assumptions, especially with count, bounded, or skewed data, can invalidate research findings.
• Transforming data is a common but often insufficient method for addressing these violations.

Purpose of the Study:

• Introduce generalized linear models (GLMs) as a superior alternative to LMs for specific data types.
• Demonstrate the appropriate application of GLMs for analyzing physical activity data with non-normal outcomes.
• Highlight the limitations of traditional LMs in the context of physical activity research.

Main Methods:

• Utilized generalized linear models (GLMs) to appropriately model count, bounded, and skewed outcome variables.
• Applied GLMs to a dataset from a physical activity study involving older adults.
• Compared the analytical outcomes of GLMs versus standard LMs for non-normally distributed data.

Main Results:

• Fitting standard LMs to inappropriate data types significantly impacts analysis, inference, and conclusions.
• GLMs provide a more accurate modeling approach for count, bounded, and skewed outcomes compared to transformed LMs.
• The study illustrates substantial differences in results when using GLMs versus LMs for physical activity data.

Conclusions:

• Generalized linear models (GLMs) are recommended for handling count, bounded, and skewed outcomes in physical activity research.
• GLMs offer a more suitable statistical approach than data transformations and standard LMs.
• Physical activity researchers should incorporate GLMs into their statistical methods for improved accuracy.