[R] Meta-regression with lmer() ? If so, how ?

[Emmanuel Charpentier]

Dear Bernd, dear list,

Bernd Weiss a écrit :
[ ... ]

> Have a look at "MiMa" at Wolfgang Viechtbauer's page. Is that what 
> you are looking for? 
> 
> <http://www.wvbauer.com/downloads.html>

As far as I can tell, mima does what I mean to do, but there are some
limits :

- mima works on effects, and therefore has an "unusual" form in R models

- as far as I can tell, mima allows to asses the effect of variables
*nesting* studies, but not of variables *crossed* in each study ;
therefore, ypou cannot directly test the effect of such variables ;

- as far as I can tell, the variables of interest ("moderators", in mima
parlance) can be either two-level factors, booleans or numeric
variables, i. e variables having a single regression coeffiient : mima
builds an estimator for the regression coefficient of each variable and
its variance, and tests by a Z-test. This is not applicable to n-valued
factors (n>2) or ordered factors, which could be tested by
{variance|deviance} analysis.

Of those limitations, the least important is the first. The second may
be worked around (at least for cases I have in mind writing this), but
the last one is quite serious.

A cursory look at mima code lets me think that it may be used to
implement a form of deviance analysis ; in this case, the recoding of
n-valued factors is something standard in R, that can be easily retrofitted.

In short : mima will be quite helpful, but is not really what I had in
mind. to be more precise, what I want to do is :

mod1<-foo(Outcome~Treatment*(Presentation+Design),
          random=(1|Study %in% Design),
          sd=Sds, data=RawData, ...)

Testing for an interaction of Treatment and Presentation (i. e. Is the
efficacy of the treatment the same for all possible clinical
presentations ?) would be done with

mod2<- update(mod1,.~.-Treatment:Presentation) # build a restricted
#model, nested in the first

anova(mod1,mod2) # test it with some form of deviance analysis

Similarly, testing for possible bias of design would be done with

mod3<- update(mod1,.~.-Treatment:Design)
anova(mod1,mod3)

Currently, mima seems to allow for this latter comparison (by building
and testing a regression coefficient for Design), but not the former.

Sincerely,

					Emmanuel Charpentier