Introduction
This web page redoes the analysis of Example 2 in the paperRuth G. Shaw, Charles J. Geyer, Stuart Wagenius, Helen H. Hangelbroek, and Julie R. Etterson (2008).
Unifying Life History Analyses for Inference of Fitness and Population Growth.
American Naturalist, 172, E35–E47.
following the course slides, deck 3.
Data
The data are in the dataset echin2
in
the R package aster
.
Set-Up and Model Fitting
Predictors In and Out
Conclusion: model with model matrix m2
is the best so far.
Indoors Effect
Conclusion: model with model matrix m2
is the still best so far.
Model m3
has no population effects on fitness.
Model m2
has population effects on fitness (acting both
outdoors and indoors).
Model m4
has population effects on fitness (acting both
outdoors and indoors) and also indoors-only population effects on fitness.
m3
is nested within
m2
is nested within
m4
.
m2
fits the data much better than m3
and
just as well as m4
. So among these m2
is the
most parsimonious model that fits.
Model m5
has only the
indoors-only population effects on fitness.
m3
is nested within
m5
is nested within
m4
.
m5
fits the data much worse than m4
and m3
fits worst than that. So among these m4
is the
only model that fits.
m2
is not nested within m5
and vice versa.
Hence they cannot be compared directly (using a likelihood ratio test).
Thus we compared them indirectly by comparing each to
m3
and
m4
.
Predictions