A PRELIMINARY STUDY OF THE ROBUSTNESS OF THREE DIFFERENT CONFIDENCE INTERVALS FOR CRONBACH'S ALPHA

摘要

Three different methods of constructing confidence intervals for Cronbach's alpha instrumentreliability measure are compared for their performance. Different conditions which both meetand violate the underlying assumption of normality of the instrument's item distributions arestudied by means of a limited simulation study. The underlying information loss whenconverting from the assumed continuous to discrete Likert type scales is also investigated. Areason why the presumably inappropriate Fisher's z transform appears to perform best isdiscussed.