中文摘要：

心理与教育测量的应用领域发生了较大变化，被测群体的知识和能力等特质在一定程度上不再服从偏度为0的分布。利用GH分布性质，模拟生成一定偏度的偏态分布数据，探讨数据的不同偏度对概化理论方差分量置信区间估计的影响。结果表明：（1）偏态分布数据的偏度对Traditional方法、Jackknife方法和MCMC方法估计方差分量置信区间有显著的影响；（2）校正的Bootstrap的PC和BCa方法估计偏态分布数据方差分量置信区间，要优于未校正的Bootstrap的PC和BCa方法；（3）使用“分而治之”策略，Bootstrap的PC和BCa方法能够找到一种（或几种）策略较好地估计偏态分布数据的方差分量置信区间。

英文摘要：

Because the application field of psychological and educational measurement makes a great change, some character such as knowledge and ability don＇ t obey this distribution whose skewness is zero again. Using the nature of Generalized Hyperbolic distribu- tion, we simulated some skewed distribution data to explore that how the skewness of distribution data had an effect on estimating confi- dence interval of variance component for Generalizability Theory. The results showed as follows ： （ 1 ） Skewmess of distribution data had a great effect on estimating confidence interval of variance component for traditional method ,jackknife method and MCMC method. （2）As for estimating confidence interval of variance component for skewed distribution data in Generalizability Theory, adjusted bootstrap PC or BCa method was more reliable than unadjusted bootsrtrap PC or BCa method. （3）There was at least one bootstrap PC and BCa method that could make a good estimation of confidence interval of variance component for skewed distribution data. But the ＇ divide - and - conquer＇ strategy needed be used.