Domain Prediction of Complex Indicators – Model-based Methods and Robustness
15 December 2015
Speaker: Dr Nikos Tzavidis, Associate Professor in Social Statistics, Department of Social Statistics & Demography, University of Southampton
Small Area (Domain) prediction of complex indicators for example, deprivation and inequality indicators typically relies on micro-simulation/model-based methods that use regression models with domain-specific random effects. When standard (Gaussian) assumptions for the model error terms are met, Empirical Best Prediction (EBP) for domains is possible and should be preferred.
In this talk we presented current research on alternative methodologies when the model assumptions are possibly violated. To start with, we discussed the use of ‘optimal’ transformations for trying to ensure the validity of the assumptions required for EBP. We then outlined alternative, possibly robust model-based methodologies. These methods are based on the use of a random effects model for the quantiles of the target distribution. By using such a model one can estimate the quantile function of the target distribution which in turn can be used for micro-simulating samples to be used in domain prediction. The link between maximum likelihood estimation and the use of the Asymmetric Laplace Distribution as a working assumption will be described. The proposed method can be used both with continuous and discrete (count) outcomes. The talk will briefly outlined future work on the use of this method with discrete outcomes in particular on how to impose smoothing for estimating the quantiles of discrete outcomes. Estimation of the Mean Squared Error of the domain parameters was discussed and open areas for research were described. Some results by using simulation studies and real data were also presented.