When conducting a metaanalysis, look for moderating
variables. These are third factors that may influence the relationship
of interest. Hunter et al. (1982) present a ChiSquare test for systematic variation,
which is useful in determining whether
there is a moderator variable present.
Where K is the number of studies in the analysis.
If the Chisquare is not statistically significant, then no moderator variable
is present. Statistically this is a very powerful test, given a large enough
N, it will reject the null hypothesis even if there is only trivial variation
among studies. Alternatively Hunter et al (1982) give a rule of thumb, in
which S^{2}_{r} and S^{2}_{er} are
compared. If the error variance accounts for less than 75% of the uncorrected
variance, then a moderator variable may be present.
