本研究探討以r值作為效果量時，加權值計算方式對後設分析結果的影響。加權值的計算與樣本數有關，所以除了比較不同加權方式下的平均效果量差異外，也要分析樣本數及效果量的關聯性。研究者以「學業成敗歸因型態對學術成就及能力知覺的影響」為題，搜尋中、英文期刊論文。結果發現：第一，效果量與樣本數之間無顯著相關；第二，若未考慮樣本依賴問題，會得到偏差的平均效果量；第三，在解決樣本依賴問題上，不同方法的比較結果依平均效果量計算公式而不同。與未加權的平均效果量相較，加權方式採Hunter與Schmidt方法，可能得到較低的效果量，而Hedges與Vevea方法之下所得結果無明顯差異。根據上述結果，本文建議以r為效果量時，應顧及樣本依賴性並採用隨機效果模式。

This study examined the impact of effect size (r) weight on meta-analytic outcomes. In the issue of “the effect of academic attribution on academic performance and ability perception”, the author conducted searches in the electronic database and deleted inappropriate studies. The correlation coefficient between sample size and absolute value of effect size was computed. The mean of unweighted effect sizes was computed along with those by means of Hunter and Schmidt (1990) and Hedges and Vevea (1998). It was found that there was no significant correlation between effect size and sample size. Data dependency contributed to biased effect size; however, it was not stated whether the biased effect size was overestimated. For avoiding data dependency, the researchers usually averaged the effect sizes from the same samples and then computed the mean effect sizes for all studies. An alternative method was that the outcome of the weights divided by the number of effect size from the same sample was used as the new weight (O’Mara, Marsh, Craven, & Debus, 2006). It was found that these two ways of avoiding data dependency contributed to the similar averaged effect sizes in the method of Hedges and Vevea, rather than in the one of Hunter and Schmidt. Because there was no significant relationship between effect size and sample size, the unweighted effect size was reasonably valid. The methods developed by Hedges and Vevea resulted in the outcomes similar to the unweighted effect size, which was different from those estimated
by the method of Hunter and Schmidt. The findings implied seemingly that the methods which could avoid data dependency and which were developed by Hedges and Vevea were suggested.