The reliability and stability of verbal working memory measures.

The psychometric properties of several commonly used verbal working memory measures were assessed. One hundred thirty-nine individuals in five age groups (18-30, 50-59, 60-69, 70-79, and 80+ years) were tested twice (Time I and Time II) on seven working memory span measures (alphabet span, backward digit span, missing digit span, subtract 2 span, running item span, and sentence span for syntactically simple and complex sentences), with an interval of approximately 6 weeks between testing. There were significant effects of age on all but two of the tasks. All the measures had adequate internal consistency. Correlations between performances at Time I and Time II were significant for all the tasks, other than the missing digit span task. The magnitude of the correlations was similar across the age groups and ranged from .52 to .81. Classification of subjects into discrete memory span groups on the basis of a single measure was highly inconsistent across testing sessions and tasks. Classification into upper and lower quartiles was more stable than using a cutoff score for group membership or than classification into high-, medium-, and low-span groups. Correlational analyses showed that there was a moderate relationship between performances on many of the span tasks. Confirmatory factor analysis suggested that six of the seven tasks reflected a common factor. Both test-retest reliability and stability of classification improved when a composite measure reflecting performance on several tasks was used.