Hello. I am hoping to conduct a meta analysis in which each candidate study reports correlation coefficients between subjective ratings of spoken fluency (on a numerical scale) and objective measures of the rated speech (e.g. number of pauses >.25ms per minute of speech). In most studies, each speech sample is rated by a number of raters, and the coefficient reported in the study is from an average of the correlations for all the raters of all the speakers. For example, a study might involve 5 speakers performing a task and being rated for fluency by 5 different raters, and correlations being calculated between raters’ scores and the measure in question. Then these correlations are averaged to give the coefficient for the study.

In terms of coding the original data for the purpose of calculating effect sizes, (how) do I take into account the number of raters, as well as the number of speakers? In the above hypothetical case, is the sample size five, or ten, or something else?

Thanks in advance.