Premium
Parametric modelling of M ‐quantile regression coefficient functions with application to small area estimation
Author(s) -
Frumento Paolo,
Salvati Nicola
Publication year - 2020
Publication title -
journal of the royal statistical society: series a (statistics in society)
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 1.103
H-Index - 84
eISSN - 1467-985X
pISSN - 0964-1998
DOI - 10.1111/rssa.12495
Subject(s) - small area estimation , quantile , quantile regression , outlier , estimator , estimation , parametric statistics , econometrics , statistics , inference , mathematics , population , linear regression , computer science , economics , artificial intelligence , demography , management , sociology
Summary Small area estimation methods can be used to obtain reliable estimates of a parameter of interest within an unplanned domain or subgroup of the population for which only a limited sample size is available. A standard approach to small area estimation is to use a linear mixed model in which the heterogeneity between areas is accounted for by area level effects. An alternative solution, which has gained popularity in recent years, is to use M ‐quantile regression models. This approach requires much weaker assumptions than the standard linear mixed model and enables computing outlier robust estimators of the area means. We introduce a new framework for M ‐quantile regression, in which the model coefficients, β ( τ ), are described by (flexible) parametric functions of τ . We illustrate the advantages of this approach and its application to small area estimation. Using the European Union Survey on Income and Living Conditions data, we estimate the average equivalized household income in three Italian regions. The paper is accompanied by an R package Mqrcm that implements the necessary procedures for estimation, inference and prediction.