NO-RESPUESTA DE ITEMS EN ESTUDIOS DE MERCADO

Statistical procedures for missing data have improved significantly in the last years. This study put the missing data in context and makes a revision of the recent literature for the case that the missing data problem is ignorable. In an application, based in real data of a psychographic profile study, several different methods to treat missing data are implemented. The results of this exercise show that most of the traditional imputation procedures induce bias and do not consider the whole variability in the estimates. Multiple imputation, based on Bayesian models and data augmentation gives the best results in the application.