Kieffer Kevin M, Cronin Christopher, Fister Matthew C
Department of Psychology MC2127, Saint Leo University, P.O. Box 6665, Saint Leo, Florida 33574, USA.
J Stud Alcohol. 2004 Sep;65(5):663-71. doi: 10.15288/jsa.2004.65.663.
Reliability Generalization, a relatively new meta-analytic score reliability technique, was used to examine reliability coefficients for Alcohol Expectancy Questionnaire (AEQ) subscales in published research. Specifically, the present study identified the typical reliability coefficients of scores on AEQ subscales across published studies and examined sources of measurement error across AEQ subscales and studies.
Based on literature reviews of PsyclNFO and PubMed, a total of 71 studies were analyzed and coded on 10 different dimensions believed to affect score reliability.
Only 37% of the studies surveyed reported reliability coefficients for the data in hand. Analysis of these studies revealed that the average score reliability across studies varied considerably in subscales and samples, with the Arousal/Interpersonal Power and Sexual Enhancement subscales evidencing the greatest variability. Gender homogeneity and racial homogeneity were found to be the two most important predictors of the magnitude of score reliability coefficients across subscales and studies.
Results of the study indicate that the AEQ tends to generate reliable scores, with some noted exceptions. Because the Arousal/Interpersonal Power and Sexual Enhancement subscales tend to generate unacceptably low reliability coefficients and because the AEQ holds promise in furthering prevention and treatment outcome research, it is suggested that future research with the AEQ explore instrument factor structure and item consistency. Because reliability estimates can influence substantive statistical analyses, it is recommended that authors using the AEQ report reliability information in their published work.
可靠性概括分析是一种相对较新的元分析分数可靠性技术，用于检验已发表研究中酒精预期问卷（AEQ）分量表的可靠性系数。具体而言，本研究确定了已发表研究中AEQ分量表分数的典型可靠性系数，并考察了AEQ分量表和研究中测量误差的来源。
基于对PsyclNFO和PubMed的文献综述，共分析了71项研究，并在10个被认为会影响分数可靠性的不同维度上进行编码。
在接受调查的研究中，只有37%报告了手头数据的可靠性系数。对这些研究的分析表明，各研究间的平均分数可靠性在分量表和样本中差异很大，其中唤起/人际权力和性增强分量表的变异性最大。性别同质性和种族同质性被发现是各分量表和研究中分数可靠性系数大小的两个最重要预测因素。
研究结果表明，AEQ倾向于产生可靠的分数，但有一些明显的例外。由于唤起/人际权力和性增强分量表往往产生低得令人无法接受的可靠性系数，且由于AEQ在推进预防和治疗结果研究方面具有前景，建议未来对AEQ的研究探索量表因素结构和项目一致性。由于可靠性估计会影响实质性统计分析，建议使用AEQ的作者在其发表的作品中报告可靠性信息。
