I have a meta-analysis dataset which I would like to analyze as a mixed
model, where the y-variable is a measure of effect size, the random effect
is the study from which the effect size was extracted, and the fixed
effect is a categorical explanatory variable. The complication is that we
often have multiple estimates of effect size from a single study (e.g. the
experiment was repeated in different years, or under different
conditions). Being a meta-analysis, I need to weight the effect sizes by
the inverse of the effect SE. Thus my dataset includes: study, effect
size, SE, explanatory variables.
The problem is that I have failed to find a mixed model package which
allows me to both weight the y-variable and include study as a random
effect.
Specifically, the linear mixed model packages (e.g. nlme) allow for random
effects but not weighting of the y (the "weight" command of these
functions means something quite different!). The rmeta package allows for
both weighting and random effects, but is only for binary data. The mima
function of Wolfgang Viechtbauer allows both weighting and random effects
but assumes that each row of data is from a separate study (not true in my
dataset).
Any help would be appreciated, as I seem to have hit a dead end.
Diane Srivastava
--
D.S. Srivastava
Associate Professor
Dept. of Zoology
University of British Columbia
6270 University Blvd.
Vancouver B.C. V6T 1Z4
Canada srivast at zoology.ubc.ca
www.zoology.ubc.ca/~srivast/