Hsu Louis M
School of Psychology, Fairleigh Dickinson University, Teaneck, NJ 07666, USA.
Psychol Methods. 2005 Dec;10(4):420-7. doi: 10.1037/1082-989X.10.4.420.
One version of r equivalent, calculated from Fisher's exact test p values and recommended for small samples, is considered "a more realistic . . . [and] a more accurate estimate of the population correlation than . . . the sample correlation, r sample" (R. Rosenthal & D. B. Rubin, 2003, p. 494). Small sample properties of r sample and of two effect size estimators (r equivalent* and r hybrid) that use r equivalent were examined: r sample is preferable to r equivalent* (defined as r equivalent used without restrictions) in terms of bias and mean squared error (MSE); r hybrid (defined as r equivalent only when r sample = 1.0) is generally preferable to r equivalent*, and preferable to r sample in terms of MSEs, except when population correlations are very large. Conditions favoring r sample over r equivalent* and r hybrid in meta-analyses are noted.
根据费舍尔精确检验p值计算得出的r等效值的一个版本，适用于小样本，被认为是“比……样本相关性r样本更现实……[且]对总体相关性更准确的估计”（R. 罗森塔尔和D. B. 鲁宾，2003年，第494页）。研究了r样本以及使用r等效值的两个效应量估计值（r等效值和r混合值）的小样本属性：在偏差和均方误差（MSE）方面，r样本比r等效值（定义为无限制使用的r等效值）更可取；r混合值（定义为仅当r样本 = 1.0时的r等效值）通常比r等效值更可取，并且在均方误差方面比r样本更可取，除非总体相关性非常大。文中指出了在元分析中有利于r样本优于r等效值和r混合值的条件。
