Background Sample selection can substantially affect the number of factors generated using exploratory factor analysis (EFA). The 12-item World Health Organization (WHO) Disability Assessment Schedule 2.0 (WHODAS-2.0) measures disability across six life domains (cognition, mobility, self-care, getting along, life activities, and participation in society). We investigated the influence of the sampling strategy on EFA results for the WHODAS-2.0.
Methods Data from adults aged 50+ from the six countries in Wave 1 WHO SAGE study were used to repeatedly select 1000 samples (n = 750) using two strategies: (1) simple random sampling reproducing the WHODAS-2.0 summary score country distribution (i.e., positively skewed distributions (PSD) with many zero scores indicating low prevalence of disability); and (2) stratified random sampling including people with varying degrees of disability (approximately symmetric distributions, ASD). Number of factors was determined using principal components analysis, polychoric correlations, parallel analysis and mean eigenvalue criterion.
Results PSD samples typically produced one-factor solutions, except for the two countries with low percentages of zero scores. ASD samples generally produced two-factor solutions (factor 1: getting along domain items, factor 2: all the rest) or three-factor solutions (factor 1: getting along, factor 2: self-care, factor 3: mobility, life activities, and participation in society).
Conclusions Samples with high prevalence of no disability (i.e., zero scores) produce heavily censored data (i.e., floor effects), limiting data heterogeneity and reducing the numbers of factors retained in EFA. Samples of convenience or collected for other purposes (e.g., general population surveys) would usually be inadequate for validating measures using EFA.