Score Linking

Scores are the end products of assessment processes. Scores are used for admissions, placement, diagnosis, and other purposes. Scores from different assessments are often compared. The urge to make comparisons can lead someone who would never consider using height as a measure of weight to fall prey to the temptations of presuming that scores from any one educational assessment can be used as substitutes for scores from another assessment, forgetting that assessments are designed for different purposes. This entry discusses the reasons that score linking is performed, how it is performed, and considerations when linking scores from different assessments.

Even when the purposes of two different assessments are similar, linking scores from these assessments can be a challenge. For example, the SAT and ACT are both used for college admissions. Scores on the tests need to be linked before they are compared. A symmetric score link is a transformation between the scores from one test to those from another such that the path from scores on test Y to test X is the inverse of the path from test X to Test Y. In one table, an ACT score of 28 corresponds to an SAT score of 640; in the inverse table, a 640 corresponds to a 28.

Features of testing situations affect the type of score linking or scale aligning that can be achieved. These features include test content, target test-taker groups, and conditions of measurement. Different types of linking that vary with respect to these dimensions are score equating, linking scores from tests in transition, concordance, and vertical scaling. Inferences that can be made from each of these types of linking scores are constrained by the features of the testing situation.

Score equating is the highest form of score linking. Its goal is to produce interchangeable scores. Large-scale testing programs develop different editions of the same test from a common blueprint. Equating adjusts for differences in the difficulty of different test editions to produce interchangeable [Page 1478]scores. Large representative samples of examinees, sound data collection practices, and appropriate methods are needed to produce equated scores on these test editions. These conditions benefit all types of score linking.

The interchangeability of scores associated with equating is not achieved, however, simply because proper numerical operations have been performed. Only tests that measure the same construct can be equated. A math test can be linked to a reading test, but it cannot be equated to a reading test. Likewise, even though both are measures of size, height cannot be equated to weight. A short test that produces erratic scores cannot be equated to a long test that produces very stable scores. In order to be equated, tests need to produce equally reliable scores. The relationships between scores on two equated tests need to be the same across different subgroups, such as males and females.

Another scenario occurs when there is an interest in linking scores across related but distinct tests. The term concordance is used to describe this type of linkage. Typically, the tests measure similar constructs, are administered to similar kinds of examinees, and are used for the same purpose but differ in test specifications. In some cases, one test is a redesign of the other, as with an old and new version of the SAT.

...