This article aims to estimate the determinants of attrition/permanence of people in the panel of the Brazilian Monthly Employment Survey (PME/IBGE) and also to test the bias emerged from ignorability of attrition selection in regression analyses. Results point out a strong correlation between attrition rate and characteristics closely related to geographic mobility. In addition, both survey design aspects and socioeconomic events are significantly related to the probability of attrition. In selection correction models, we find that all selection mechanisms are endogenous, even though they are not directly correlated with the equation of interest. Finally, we show that ignoring the endogeneity of sample selection, such as attrition, may imply to a biased analysis. However, controlling for only some selection mechanism, not controlling for all, may provide estimators even more biased than no controlling.
Brazilian Monthly Employment Survey; panel attrition; attrition bias; bias selection corrections
