Model-Based Estimation of Small Area Dissimilarity Indexes: An Application to Sex Occupational Segregation in Spain

This paper introduces a new statistical methodology for estimating Duncan dissimilarity indexes of occupational segregation by sex in administrative areas and time periods. Given that direct estimators of the proportion of men (or women) in the group of employed people for each occupational sector are not accurate enough in the considered estimation domains, we fit to them a three-fold Fay–Herriot model with random effects at three hierarchical levels. Based on the fitted area-level model, empirical best predictors of the cited proportions and Duncan segregation indexes are derived. A parametric bootstrap algorithm is implemented to estimate the mean squared error. Some simulation studies are included to show how the proposed predictors have a good balance between bias and mean squared error. Data from the Spanish Labour Force Survey are used to illustrate the performance of the new statistical methodology and to give some light about the current state of sex occupational segregation by province in Spain. Research claims that there is a sex gap that persists despite advances in the inclusion of women in the labour market in recent years and that is related to the unequal sharing of family responsabilities and the stigmas still present in modern societies.