Skip to content

is a mixed effect model appropiate?

1 message · Andrew Robinson

#
Having a statistical history more closely related to experimental design
than to modeling, I am more relaxed about the use of a random effect to
capture even small numbers of levels.

The disadvantage of imposing a fixed effect where the design calls for a
random effect is that the random variation is being reduced in the model,
possibly unreasonably.

It really depends on what the intent of the model is.

The Westfall paper is interesting - thanks for pointing it out! - but I do
not think that its conclusions hold in this instance.  I vote for random
effects.

Best wishes,

Andrew


On 11 August 2017 at 20:23, Henrik Singmann <singmann at psychologie.uzh.ch>
wrote: