Finding Physical Activity Patterns Using Cluster Analysis

Abstract

Without fully understanding patterns of physical activity (PA) participation of a targeted population, population specific designed PA interventions may not work effectively. PURPOSE: Using cluster analysis, the purpose of this study is to identify PA patterns of urban women. METHOD: This study used data from a previous urban women PA study in which PA participation information was collected from 480 urban women using a questionnaire composed of 11 activity categories and 99 activities. PA participation was converted into energy expenditure using the Compendium of Physical Activities (Ainsworth et al., 2000). PA patterns were detected and the pattern characteristics were explained. Major activities that determined the PA patterns were investigated. Specifically, cluster analysis using dissimilarity distance between subjects was used to find PA participation clusters. Descriptive statistics were then used to provide an explanation of PA patterns by cluster. RESULTS: Among the 11 activity categories, the most energy was spent on working (29.5%), housework (28.4%), walking (10.8%) and moving (9.4%). Sports/exercise was only 6.1% of the total energy expenditure. 16 activities played key roles in differentiating the eight distinct PA patterns, identified by cluster analysis. Group 1 did the most housework activities and the least work related activities among all the groups, and had a high activity level in sports and walking. Group 2 was the most active group in sports/exercise, and was also active while working. Group 3 showed the lowest activity level in sports, walking up stairs and walking of all the groups, and the highest energy expenditure in activities while working and driving a car. Group 4 spent the most energy in caring for others of all the groups and had a high energy level while driving a car. Group 5 had rather high activity levels in housework and sports, and average levels in other activities compared to the other groups. Group 6 showed some unique activity patterns in arts and dancing in which they had higher activity levels than other groups. Groups 7 and 8 were different but did not show clearly different patterns. CONCLUSION: Cluster analysis proves to be a useful means to identifying PA patterns, which should help design more effective, population-specific, targeted PA interventions.