Statistical Methodology 8 (2011) 172–184
Contents lists available at ScienceDirect
Statistical Methodology
journal homepage: www.elsevier.com/locate/stamet
On model selection in the case of nested distributions — An
application to frailty models
P. Economou
Department of Engineering Science, University of Patras, Rion-Patras, Greece
article info
Article history:
Received 25 October 2009
Received in revised form
4 September 2010
Accepted 4 September 2010
Keywords:
Nested distributions
Test statistics
Diagnostic plot
Frailty
abstract
A number of criteria, test statistics and diagnostic plots have been
developed in order to test the adapted distribution assumption.
Usually, a simpler distribution is tested against a more complicated
one (by adding an extra parameter), which include the first
distribution as a special case (nested distributions). In this paper,
two new tests are developed in order to test nested distributions.
This work concentrates on the case where the extra parameter lies
on the boundary of the parameter space. The developed tests are
applied on testing the frailty term on frailty models. Simulation
results are presented for comparison with other test statistics and
the methods are illustrated on two real data sets.
© 2010 Elsevier B.V. All rights reserved.
1. Introduction
A number of tests have been developed in parametric data analysis in order to determinate if the
adapted distribution assumption is the proper one. An interesting case is when the tested distributions
are nested, in the sense that one is a special case of the other. There are many examples of such cases,
and the most common ones, but not the only ones, are formed by including an extra shift γ parameter
in the initial distribution.
Bai et al. [1] performed a simulation study using eight different criteria, namely the
Akaike’s Information Criterion (AIC), the Schwart’s Information Criterion (SIC), the Kullback–Leibler
Information Criterion (KL), a new Generalized Information Criterion (GIC) proposed by them, two
criteria related to the upper percentile error (UPE), the entire or Full Percentile Error (FPE), the
Chi-square test (CHI) and the Kolmogorov–Smirnov (KS) test, in order to discriminate among two-
parameter and three-parameter nested alternative distributions. In their study they have considered
three widely used three-parameter (shape–scale–location/shift) distributions, namely the Weibull,
the Gamma and the log-normal distributions, against their two-parameter versions in which the
location parameter is set equal to zero.
E-mail address: peconom@upatras.gr.
1572-3127/$ – see front matter © 2010 Elsevier B.V. All rights reserved.
doi:10.1016/j.stamet.2010.09.002