经典信度理论和概化理论的计量学透视

A Metrological Perspective of Classical Reliability Theory and Generalizability Theory

从计量学的角度分析了经典信度理论和概化理论的局限、错误和无力,并重点揭示了经典理论和概化理论几个方面的问题。经典理论由于采用了相关分析进路,既无力应对多不确定度源的测量情景,也无法较好地满足实践对条件不确定度的要求。概化理论虽然可以同时考虑多个不确定度源,但由于技术上采用了方差分析进路,即使使用得当,也不能够很好地解决条件不确定度问题。

This paper analyzes the limitations, mistakes and impotence of classical reliability theory and generalizability theory from the metrological perspective. The paper exposes some misuses or abuses, misconceptions, and methodological problems of both theories. Correlation approach makes it impossible for classical reliability theory to either deal with the multi-component uncertainty measurement situation, or to calculate the conditional uncertainty as is often demanded in practice. ANOVA approach enables generalizability theory to handle the multi-component uncertainty problem, when properly applied, but the conditional uncertainty problem remains.