Self-report physical activity (PA) questionnaires have been widely used by major national surveillance systems to assess and monitor the PA status of the U.S. population. The comparisons among subpopulations using these data indicate evident PA disparities. However, it is unknown if the items used in these questionnaires are biased against a subgroup in the population. It is also not clear to what degree the disparities in PA are affected by the biased items if there are any.
Purpose
By applying differential item functioning (DIF) analysis, the purposes of this study were: (a) to determine if there are any DIF items in the National Health and Nutrition Examination Survey (NHANES) PA questionnaire; and (b) to examine the impact of DIF items, if there are any, on current PA disparity conclusions.
Methods
A sub-set of adult data from the “NHANES 2003-2004” data sets with an unweighted sample of 3,083, consisting of 1,511 males and 1,572 females, was used for data analyses. DIF analyses were conducted using three DIF procedures: Mantel-Haenszel, SIBTEST and ANOVA. Eight general PA items related to respondents' occupational, transportation, domestic, leisure time, and strength related activities and sedentary activities at home were analyzed for DIF. A “Jackknife” procedure was used to determine the impact of DIF items on the current PA disparities conclusions.
Analysis/Results
The results showed there are many DIF items among general PA questions used in the NHNAES PA questionnaire. A majority of general PA DIF items favored advantaged social groups (e.g., non-Hispanic White, male, high education or income group, and adults age 20 to 59 years old) while occupational and transportation-related PA items favored disadvantaged groups (e.g., Hispanics, low education or income group). Prevalence estimations with and without a DIF item being included in the calculation have led to different proportions of the U.S. population meeting PA recommendations.
Conclusions
Many DIF items have been identified in the NHNAES PA questionnaire. The reported PA participation among subpopulations were likely affected by these DIF items and the estimation of the population meeting PA recommendations using these items is likely inaccurate. Thus PA researchers need to learn DIF and make DIF analysis a standard practice in PA survey construction. There is also a need for the development of DIF software that can specifically handle PA data.