The Effect of Study Participant Compliance on the Estimates of Physical Activity: Implications for the Predictions of Health Markers Thesis uri icon



  • Thesis (Ph.D., Movement & Leisure Sciences) -- University of Idaho, 2015 | Purpose: To investigate the influence of accelerometer data simulations and adherence on estimates of sedentary behavior and physical activity, and the subsequent application of these data. Methods: A sample of 50 female (age=25.6±8.6 years) and 50 male (age=25.4±7.2 years) participants wore an accelerometer at least 22 hours/day for 7 consecutive days (raw data) to assess sedentary behavior, light physical activity (LPA), moderate-to-vigorous physical activity (MVPA), and average intensity of daily physical activity (DPA). A series of health markers were also measured. Reductions in accelerometer adherence were simulated by randomly removing 60 minute blocks of time during waking hours until each participant had 17-10 hours of data. Four different accelerometer data simulation techniques were tested by inserting “zeroes” or “dots” in the raw data to simulate accelerometer removals during sleep or waking hours. One-way ANOVA with Bonferroni post-hoc tests on mean estimates and absolute percent errors (APE) were used to analyze the differences between each accelerometer data simulation technique and adherence when compared to the raw data. The raw and simulated accelerometer data were then used in univariate regressions to predict each health marker that was measured, and the results were compared. Results: Participants averaged 23.3 hours/day of accelerometer wear, for a total of 687 days across all participants. Significant differences (p<0.05) were detected in the estimates of sedentary behavior, LPA, and DPA when compared to the raw data for some data simulation techniques. APE increased in a stepwise fashion as adherence decreased for LPA and MVPA; however, the data simulation techniques yielded different patterns of error as adherence decreased for sedentary behavior and DPA. Decision changes were found when the simulated data sets were used to predict each health marker, but only in about 7% of analyses. Conclusions: Generally, accelerometer data simulation techniques and low adherence may negatively influence predictions of sedentary behavior and physical activity when compared to the raw data. However, decision changes resulting from applying spurious accelerometer data in univariate regression models was minimal. These results indicate accelerometer data simulation techniques should be carefully applied and high accelerometer adherence should be a priority for researchers.

publication date

  • June 15, 2015

has major professor