Evaluating Comparability of Activity Measures Across Public Use Data Sets

Background and Purpose: The benefits older adults experience as a result of engagement in various types of activities has been the subject of much attention in recent years. However, “activity” is defined in disparate ways across the literature, making comparison of results quite difficult. The aim of this analysis was to explore how activity is measured in commonly used public data sets. Specific research questions were: a) How many activity-related variables are in the sample of data sets and what types of measurements are represented (e.g. binary, categorical, interval, ordinal)?; and b) can discrete activities can be grouped into conceptual domains to facilitate parsimonious empirical analysis of activity engagement among older adults?

Methods: Publicly available data sets were identified for possible inclusion in the sample. Criteria for selection were: self-report survey, nationally representative population, inclusion of older adults (65+), and survey content covering at least one area of activity recognized in the literature (e.g. physical activity, volunteering). Additionally, each data set should contain global measures of health and wellness (e.g. life satisfaction, well being) suitable for modeling outcomes of activity participation. Five data sets were identified for inclusion in the final data set: American Changing Lives Study (ACL), Health and Retirement Study (HRS), Midlife in the United States (MIDUS), National Health Interview Survey (NHIS), Panel Study of Income Dynamics (PSID). From these, the most recent two waves were analyzed. Content analysis was used to identify activity measures within these data sets. Measures were categorized according to rules that had been developed during pilot work, and descriptive statistics were generated to facilitate comparison.

Results: The number of activity measures per data set ranged from 41 (HRS 2007) to 165 (MIDUS 2004/2006). Measurement type varied between different data sets and within waves of the same data set. Ten activity types or domains were identified across data sets. The three data sets with the most domains were MIDUS (n=10), HRS (n=9) and ACL (n=9). Employment/paid work and physical activity were the only domains present in all five data sets. Predominant measurement type varied across data sets and between waves of the same data set. Binary, ordinal and interval were the most common measurement types. For example, HRS 2008 has mostly ordinal (65%) and HRS 2007 Activity supplement has mostly interval (68%) activity variables. In MIDUS 1994/95, the greatest proportion of variables is binary (44%), while in 2004-2006, 59% are ordinal.

Conclusions and Implications: There is substantial variation in the number and type of activity measures in the five public use data sets we evaluated. Even when content is consistent across data sets, method of measuring often is not – where one survey uses an ordinal measure, another uses a continuous or dichotomous measure. Despite these differences, some areas of consistency were identified. We believe that to advance research on activities and older adults, a more consistent means of empirically measuring activity must be developed. Future work will focus on evaluating the potential for empirically evaluating and consolidating activity measures across these data sets.