IASS Webinar 36: Latent Variable Models for Finite Population Inference

Watch the recording

The use of latent variable models has become common practice in estimation from complex surveys, particularly to address small area estimation problems and non-sampling errors. In their basic definition, the assumption of Normality is made for the variable of interest and/or for the random effect(s) introduced to account for unobserved heterogeneity. In this talk, I will discuss the use alternative latent variable models, such as latent class models and latent trait models in estimation from complex surveys. Latent class models are useful tools to deal with (possibly non-ignorable) unit nonresponse to build better response homogeneity groups and using, for example, generalized calibration, when measurement error on the response variables is suspected. Latent trait models can be used to obtain a measure of the "attitude to respond" to a survey that can be used as a covariate in response propensity estimation. Finally, these models are very useful when the variable of interest is not directly observable, such as disability, social integration, educational poverty, and/or measured with error by means of a set of binary/categorical variables. I will present applications of these methods to Italian Household Surveys, such as the Survey on Households Income and Wealth and the survey on Health Conditions and Appeal to Medicare, and discuss their use to handle the process of integration of administrative and survey data for the production of official statistics.