Reliability of AIDS knowledge scales: conceptual issues.

A common approach to evaluating the reliability of AIDS knowledge scales has relied on the demonstration of internal consistency through the use of Kuder-Richardson formula 20 (KR-20) or coefficient alpha. In this study, the argument was made that, in a knowledge-based scale with diverse content (exemplified by AIDS knowledge scales), internal consistency does not make sense conceptually and underestimates the scale's true reliability. It was therefore hypothesized that a split-half approach, in which items in each half were matched for content, would provide superior estimates of reliability when compared to either KR-20 or an odd-even split-half approach. Reliability analyses performed on a knowledge scale developed by the author supported this hypothesis.