Related Experiment Video
Updated: Jun 9, 2025
05:59
Visualization of Intensity Levels to Reduce the Gap Between Self-Reported and Directly Measured Physical Activity
Published on: March 7, 2019
6.7K
Simulation-Based Evaluation of Methods for Handling Nonwear Time in Accelerometer Studies of Physical Activity
Kristopher I Kapphahn1, Jorge A Banda2, K Farish Haydel3
1Quantitative Sciences Unit, Stanford University, Stanford, CA, USA.
Journal for the Measurement of Physical Behaviour
|October 25, 2024
Summary
Multiple imputation (MI) methods improve accelerometer data analysis by reducing bias compared to discarding data or single imputation (SI). Proper application of MI, including using past acceleration data, enhances model performance for physical activity research.
Area of Science:
- Biomedical Engineering
- Epidemiology
- Data Science
Background:
- Accelerometer data are crucial for objective physical activity measurement in research.
- Missing data, termed nonwear periods, occur when participants remove accelerometers.
- Common methods for handling nonwear include data discarding and single imputation (SI).
Purpose of the Study:
- To evaluate discard-, single imputation (SI)-, and multiple imputation (MI)-based methods for handling accelerometer nonwear.
- To assess the accuracy and precision in characterizing the relationship between mean counts per minute and body mass index.
Main Methods:
- Simulated realistic accelerometer data with induced nonwear periods.
- Analyzed data using common discard and SI methods alongside various MI techniques.
- Compared bias, relative standard error, relative mean squared error, and coverage probabilities.
Main Results:
- MI approaches demonstrated superior performance with significantly lower bias compared to discard and SI methods.
- Bias for MI ranged from -0.001 to -0.028, versus -0.050 to -0.057 for discard and -0.061 to -0.081 for SI.
- Significant variation was observed among MI strategies, with coverage probabilities ranging from .04 to .96.
Conclusions:
- Multiple imputation (MI) methods offer significant benefits over discard and SI approaches for analyzing accelerometer data with nonwear periods.
- The specific application of MI is critical, with inclusion of prior acceleration measurements improving model performance.
