Skip to content
Prev 17871 / 20628 Next

Testing assumption multilevel analysis

Most of the assumptions of mixed/multivel/hierarchical are inherited
from the standard (generalized) linear modeling assumptions, and at
least the graphical diagnostics are done in the same way (residual vs
fitted plot, scale-location plot, Q-Q plots of residuals, influence
plots).  The main additional assumptions have to do with the group-level
effects, which are typically assumed to be normally distributed.
Typically people use "caterpillar plots" (plots of the BLUPS/conditional
modes for each group, sorted, with error bars based on the conditional
variances) to evaluate these distributional assumptions.

  There are some worked examples of mixed models (in ecology) at
https://bbolker.github.io/mixedmodels-misc/ecostats_chap.html

  Giving more context (as Phillip Alday suggests) would be a good idea.
 In particular, different research fields may emphasize different
criteria more or less.
On 2019-08-26 8:14 a.m., Phillip Alday wrote: