Studies that draw on individual and environmental variables to explain differences in self-rated health status have increased gradually in Brazil, but are still limited in number. Due to time and cost issues, many studies use a complex sample design involving features (stratification, clustering, and different sample weights) that, when ignored, can influence odds ratios and standard errors in the statistical models. Using the National Household Sample Survey (PNAD 2008), this paper assesses the impact on these measurements when some or all of these features are not taken into account in fitting ordinal logistic models to establish associations between adults' self-rated health and various individual and environmental factors. According to this study, failure to take these three features into account simultaneously led to changes in the magnitude of the odds ratio between better self-rated health and most of the factors, besides important underestimation of standard errors.
Logistic Models; Morbidity; Sampling Studies
