CMStatistics 2021: Start Registration
View Submission - CMStatistics
B1193
Title: Record linkage, measurement error and unit level small area estimation: A Bayesian approach Authors:  Brunero Liseo - Sapienza Universita di Roma (Italy) [presenting]
Serena Arima - University of Salento (Italy)
Gauri Datta - University of Georgia and US Census Bureau (United States)
Abstract: Small area estimation is a statistical technique involving the estimation of parameters for small sub-populations, generally used when the sub-population of interest is included in a larger survey. Suppose we have $m$ small areas and the goal is to predict the mean of the variable of interest in each area. Using a set of covariates, we consider a well-known unit-level model. However, as widely discussed in the literature, covariates might be affected by measurement error and ignoring such an error may lead to misleading conclusions in terms of both parameters estimation and the small area mean predictions. We consider the very common situation in which covariates come from a different data file with respect to the data file of the response variable. As a consequence, covariates are affected by two sources of error: measurement error and linking error. We propose a multivariate nested error regression model that accounts for both sources of error. We conduct a noninformative Bayesian inference by assigning an improper prior that we prove to lead to a proper posterior distribution. The model performance is investigated using different simulation scenarios and with a real data application.