To test the fit of alternative models, the meta-analytically pooled correlation matrix was used to estimate path models using LISREL 8 (Jöreskog, & Sörbom, 1993). As each entry of a correlation coefficient was based on a different set of studies and thus on different sample sizes, the sample size for each entry varied substantially. As expected, this analysis revealed that the number of studies decreased with the number of rounds of data collection. The harmonic mean of the sample size (Viswesvaran & Ones, 1995) was computed to test the fit of alternative models to the data.
Floyd. R.G., (2011)
倒是可以針對 X-M-Y vs. (X-M, M-Y)兩類的資料做一個seneitive analysis (如果不一樣那就有尷尬了~~~)