Dear Elizabeth,
There is no strictly 'correct' answer to your question, although it raises
complex issues that have been rehearsed several times in the mailbase. SPM99
'allows' you to model either fixed or random effects analyses between two
studies; the main difference being what conclusions you can draw from any
significant areas you find, which in turn depends on which question you are
attempting to ask of you data sets (e.g. are you hoping to generalize any
findings from your 'sample' population to the general or 'study' population).
You could do worse than search out some of e-mails on the spm archive, try
Thomas Edgar Nichols e-mail entitled 'Random effects analysis', posted on Mon
Apr 17 17:06:17 2000.
Cheers.
Alex.
> Dear SPMers
> I've recently become interested in comparing the results of 2 different
> PET studies.
>
> The studies used the same type of tasks (lexical decision and semantic
> categorization) but differed on a number of parameters including number
> of conditions, timing, number of subjects.
>
> I was under the impression that in order to compare these two different
> studies I must use a random effects analysis.
> However, I have noticed that some people conduct a fixed effects
> MetaAnalysis to compare 2 or more studies.
>
> Can anyone tell me which method (if either) is preferable? What are the
> benefits and weaknesses of these 2 methods?
>
> Many thanks
>
> ************************************************
> Elizabeth Dick
> Centre for Speech and Language
> Dept. of Experimental Psychology
> Cambridge University
> Downing Street, Cambridge, CB2 3EB
>
> Email: [log in to unmask]
> Tel: (01223) (7)66451
> ************************************************
|