Reliability

Stability Coefficients: The Test–Retest Method

The test–retest method, a measure of stability, is used to determine the consistency of the examinees’ scores on a test over time. The test–retest coefficient is obtained by correlating the scores of identical tests administered to the same examinees twice under similar testing conditions. Carry-over effects and the interval of time between the two test administrations can influence the test–retest coefficient, so this method is most appropriate for tests measuring traits that are not susceptible to carry-over effects and that are stable across time intervals. In practice, the longer the time interval between administrations, the lower the estimated reliability.

Equivalence Coefficients: The Alternate Forms Method

The alternate forms method, a measure of equivalence, is used to examine the consistency of two sets of scores on two parallel forms of a test. The alternate form coefficient is obtained by correlating the scores of parallel (or equivalent) forms of a test to the same examinees under similar conditions in close succession. That is, one form is administered to a group of examinees followed (at a well-chosen close time point) by the administration of an alternate form. The quality or similarity of the parallel forms can influence the alternate form coefficient. In practice, if the forms are not parallel, the alternate form method produces low estimates of reliability.

Internal Consistency Coefficients: Split-Half, KR-20, and Coefficient α Methods

Both measures of stability and equivalence require two administrations of (or parallel forms of) a test, but the administration of two tests can be impractical or unnecessary in reality. Internal consistency coefficients, which require a single test administration, are used to assess the consistency of the examinees’ responses to the items within a test. There are two broad classes of methods for estimating internal consistency coefficients. The first class is generally denoted as split-half procedures. The second class of methods requires an analysis of the variance–covariance structure of the item responses. With respect to the split-half methods, a test is administered to a group of examinees, then the test is split into two parallel halves, and the two sets of scores from the two split halves are correlated. This half-test reliability estimate is then used to calculate the full test reliability using the Spearman-Brown prophecy formula, which is written as

...