Dr. Payam Mokhtarian
Research Fellow in Biometrics
Centre for Bioinformatics and Biometrics (CBB)
National Institute for Applied Statistics Research Australia (NIASRA)
School of Mathematics and Applied Statistics
University of Wollongong
Title:
On Outlier-Robust Prediction of the Empirical Distribution Function
Abstract:
Outliers are a well-known problem when fitting models with survey data. Estimates of the model
parameters and also prediction of population quantities using the fitted model become unstable in
presence of outliers in data. The main robust-projective approaches that have been developed so far
for this problem have focused on modifying the parameter estimating equations to make them less
sensitive to sample outliers. A problem with the robust-projective approach is that it assumes that
all non-sampled units follow the working model, or, in what essentially amounts to the same thing,
that any deviations from this model are noise and so cancel out "on average". Therefore, the
proposed outlier robust estimators for population mean form perspective of a fold-nested error
model can be substantially biased when outliers are drawn from a distribution that has a different
mean from that of the rest of the survey data. This naturally leads one to consider an outlier robust
bias correction for these estimators. In this presentation, a robust-predictive approach based on
robust random effect block bootstrap (RREB) technique for fitting a fold-nested error model is
presented. Also, we develop a robust-projective approach to predict the empirical distribution
function of the population mean for clustered data under a fold-nested error model. Monte Carlo
simulation results are presented to provide some evidence for our claim that the proposed RREB
method is robust to the influence of outliers. This also leads to more reliable population mean
empirical distribution function predicts under a fold-nested error model for clustered data than
comparable outlier robust approaches that have been proposed in the literature.