B0266
Title: Multivariate small area estimation in case of non-continuous variables
Authors: Angelo Moretti - Utrecht University (Netherlands) [presenting]
Abstract: Policymakers require reliable information on the geographical distribution of social indicators for small areas. However, large-scale sample surveys are not designed to produce reliable direct estimates at a small area level due to the unplanned domain problem. Univariate mixed-effect models are widely adopted in small-area estimation to improve direct estimates by using auxiliary information. The literature has shown that multivariate small-area estimators, where correlation structures are taken into account, provide more efficient estimates than the traditional univariate approaches. However, there are still many gaps in the case of non-continuous response variables, such as binary, count or mixed-type response variables. The use of multivariate generalised mixed-effect regression models is relevant and promising. The focus is on the unit-level approach, where information is available at the unit level. Particular attention is paid to binary, count and mixed-type variables. The latter case is when, for example, one variable is binary, and the other variable is continuous. There are several factors that play a role in the efficiency of the multivariate estimators over their univariate setting (e.g., correlation structure and intra-class correlation coefficient). This is shown via simulation studies and applications.