Factors Affecting Reliability Coefficients of Health Attitude Scales

This study determined the minimum number of health attitude items and minimum sample size required to achieve maximum scale reliability coefficients, using different methods of estimating reliability. A 54-item alcohol attitude scale was administered to 700 participants. The scale produced .96 and .91 reliability coefficients, using the Cronbach Alpha (CA) and the Split-half (S-B) methods, respectively. A computer program randomly selected groups of participants and items from the pool of participants and items using different increments. A matrix of coefficients of reliability for both methods was calculated for different groups of items and sample size. To replicate the study, a 30-item cancer attitude scale was administered to more than 1,000 representative participants and produced reliability coefficients of .94 (using CA) and .82 (using S-B). The same computer and statistical procedures were repeated for the second data set. Results from both analyses consistently demonstrated that sample size has an insignificant effect on the coefficient values of reliability. Reliability increased as the number of items reached 18. Adding more items only negligibly increased the coefficients. Overall, the CA method consistently produced higher coefficient values of reliability compared to the S-B method.