Predicting Older Adults' Physical Activity Frequencies Using a Two-Part Model

The distribution of frequencies for physical activity (PA) participation is usually positively skewed with excess zeros, which challenges the statistical analyses. This study examined the patterns of explanatory variables in predicting the number of days a week that older adults participated in moderate- and vigorous-intensity PA (MPA & VPA) using a two-part model (Duan, Manning, Morris, & Newhouse, 1983). The study was exploratory therefore no specific hypotheses were formulated. Participants 65 years or older in the 2005 Behavioral Risk Factor Surveillance System study were included (N=64,745, Mean Age=74.43, SD=6.64). The distributions of MPA and VPA were a mix of skewed positive values (heavy tail) with excess zeros and sevens. Hence, this study applied an extended two part-model that combined a logistic regression on any participation (0 versus others) and an ordered regression on the number of days of participation (i.e., non-zero observations). Explanatory variables included age, gender, race, education level, household income, marital status, employment status, perceived health status (PHS), activity limitations, and BMI. Results from the logistic part showed that age (OR_MPA=.79, 95%CI=.78-.81; OR_VPA=.74, 95%CI=.73-.76, for every SD increase in age), PHS (OR_MPA=1.51; 95%CI=1.47-1.54; OR_VPA=1.47, 95%CI=1.44-1.50, for every SD increase in PHS), and activity limitations (OR_MPA=.57, 95%CI=.54-.59; OR_VPA=.59, 95%CI=.56-.62) were the strongest predictors among the explanatory variables on any participation. Males were more likely to participate in VPA (OR=1.65, 95%CI=1.58-1.71) with no gender differences observed for MPA. For the ordered regression part, employment status of self-employed versus employed (OR_MPA=1.45; 95%CI=1.33-1.59; OR_VPA=1.44, 95%CI=1.27-1.64) emerged as a strong predictor relative to the other predictors on the number of days of participation in MPA and VPA. Interestingly, no significant effect was found for activity limitations on the number of days of participation. In addition, older age also predicted slightly more days a week of VPA (OR=1.10, 95%CI=1.07-1.14). The results suggest that the underlying mechanisms between whether older adults participate in any physical activity and how many days of participation may be different, and the patterns of predictors may also differ between MPA and VPA. Health concerns and limitations seem to influence whether older adults participate in any activity, while the time factor (or competing activity), which is reflected from the employment status, becomes an important determinant on the number of days of participation. It also seems that activity limitations may be strong barriers for initiation, but once people overcome these barriers, activity limitations have no effect on participation levels.
Keyword(s): aging/older adult issues, measurement/evaluation, research