A multilevel approach to examining time-specific effects in accelerometer-assessed physical activity

Area of Science:

• Physical activity research
• Biostatistics
• Behavioral science

Background:

• Traditional analysis of accelerometer data often aggregates physical activity into single variables, limiting research scope.
• This overlooks detailed, time-specific data crucial for understanding intervention impacts.

Purpose of the Study:

• To propose and evaluate a multilevel modeling approach for analyzing time-specific physical activity intervention effects.
• To compare the power of this novel approach against traditional methods using simulations.

Main Methods:

• Simulations were conducted based on the Active by Choice Today trial data.
• Six conditions tested type 1 error rates and power for time-specific vs. traditional effects.
• The multilevel model was applied to assess intervention effects at specific time periods.

Main Results:

• The multilevel approach maintained appropriate type 1 error rates.
• It demonstrated a 31-38 percentage point increase in power to detect time-specific intervention effects.
• No loss of power was observed when only traditional effects were present.

Conclusions:

• Multilevel modeling offers advantages for analyzing time-specific physical activity intervention effects.
• This method can reveal intervention impacts missed by traditional analyses.
• Further research is recommended to integrate this analytic tool for accelerometer data.