Skip to content
Prev 11861 / 20628 Next

Why do extremely similar parametric bootstraps on lmer() objects yield (not only marginally) different results?

I ran some more tests and I can safely say that the culprit in Method 2
and Method 3 is the call to:

dattemp <- mod at frame

and subsequently using it as the data argument in the foreach() loop calls
to lmer().

If I use the original data frame in Method 2 and Method 3 and set
na.action = na.exclude when fitting the model the (significant) difference
to using refit() disappears. It seems that copying only the data frame of
an S4 object and using this data frame when fitting/refitting a new model
with lmer() causes some sort of "behaviour" (I refrain from calling it
trouble. I do not fully grasp what is happing there.).

Best,
Christian
Eberhard Karls Universit?t T?bingen
Mathematisch-Naturwissenschaftliche Fakult?t - Faculty of Science
Evolutionary Cognition - Cognitive Science
Schleichstra?e 4 ? 72076 T?bingen ? Germany
Telefon +49 7071 29-75643 ? Telefax +49 7071 29-4721
christian.brauner at uni-tuebingen.de
On Sat, May 03, 2014 at 09:38:30PM +0200, Christian Brauner wrote: