Abstract:
|
Physical activity (or lack of) for Americans has become such an important issue that in 2008, for the first time, the Department of Health and Human Services issued the 2008 Physical Activity Guidelines for Americans. They state that adults should participate in a minimum of 150 minutes of moderate to vigorous activity (MVPA). There is interest in knowing what proportion of the population adheres to these guidelines, as well as whether demographic disparities exist. In addition, there is a non-negligible portion of the population that never participates in MVPA, creating a spike at 0 for an otherwise continuous variable. To address these questions, we propose a Bayesian two-part model that accounts for measurement error. We use data from the Physical Activity Monitoring Study, which used armbands and 24-hour recalls. We model the probability of participating in MVPA as well as the amount of MVPA as a function of demographics. We include random effects to allow correlation in between regressions. We construct calibration equations for the biased 24-hour recalls. Our work provides a Bayesian implementation of the traditional two-part regression calibration model.
|