If you have all the raw data and it has been collected under comparable conditions (metaanalysis is mainly used to combine the results of more than one separate study) then just combine all the data and calculate the correlations you are interested in. Your sample size will be large and presumably you can have great confidence in the resulting correlation coefficients.
A difficulty may arise if the conditions under which the data were collected are different in the different studies you are trying to combine, and then you are back in standard metaanalysis territory - how to render disparate studies as comparable as possible. I presume then it would be better to calculate correlations for each particular study and to exclude from your analysis those studies that used very different procedures.
firstname.lastname@example.org wrote: > Dear groups, > > I some question concerning how to do meta-analysis of correlations when > raw data are available. > > I have obtained the raw data from a group of studies that measure > variables that I am interested in the correlations among. The studies > themselves did not report the correlations which is why I contacted > their authors to obtain the data. > > However, all the books on meta-analysis I have consulted do not explain > in detail how my situation is different from the typical situation in > which meta-analyses are conducted (no raw data available). So my > question is if anyone can point me to books or papers that highlight > the differences between these situations and the possible opportunities > for analysis that having the raw data opens. > > In addition, I am interested in references or advice on how to handle > multiple correlations from individual studies (Hunter & Schmidt say a > bit about this, but perhaps there is are other good sources out > there?). In particular I am unsure about how to handle correlations > from an individual study where the number of observations differs > between the variables that I am correlating: in my case task > performance and post-experiment rating. > > Thanks a lot, > Kasper Hornbæk