再測資料之信度估計式

An Alternative Estimate of Reliability for Test-Retest Data

本研究針對同類問項之再測資料，提出一無需使用驗證性因素分析之信度估計式，該式係再測相關係數之記憶效果校正式，不會受困於估計問題。蒙地卡羅模擬結果顯示，無論是常態或非常態資料，當驗證性因素分析無估計問題存在時，在大樣本下，所提信度估計式與利用驗證性因素分析之既有信度估計式表現相當，而在較小樣本下，該估計式之表現較優。該估計式於再測資料之信度評估上具參考價值。

In this research, an alternative estimate of reliability for test-retest data with congeneric items is proposed without confirmatory factor analysis (CFA). It is the sample test-retest correlation corrected for estimated memory effects, and involves no estimation problems. Monte Carlo simulation results have indicated that, no matter for normal or non-normal data, when there exist no estimation problems in CFA, the proposed estimate performs as well as the existing estimate resulting from CFA for large samples, and better for smaller samples. It is useful for assessing reliability of scale scores in test-retest designs.