Quantile-like Models for Nominal Data with Application in Small Area Estimation Seminar

Event details

Quan​tile regression can be used as a robust alternative to random-effects models in small area estimation. In doing so, the group effects are captured by aggregating individual units using quantile indices within each group. However in cases with non-continuous responses such as the binary or categorical case, quantiles are not well-defined. In this talk expectiles are introduced which are simply L2 versions of the L1 quantiles. It is shown how expectiles can be easily generalised to the nominal case, making them applicable to small area estimation.