Construct Validity

Origins and Definitions

Construct validation generally refers to the collection and application of validity evidence intended to support the interpretation and use of test scores as measures of a particular construct. The term construct denotes a distinct dimension of individual variation, but use of this term typically carries the connotation that the construct does not allow for direct observation but rather depends on indirect means of measurement. As such, the term construct differs from the term variable with respect to this connotation. Moreover, the term construct is sometimes distinguished from the term latent variable because construct connotes a substantive interpretation typically embedded in a body of substantive theory. In contrast, the term latent variable refers to a dimension of variability included in a statistical model with or without a clear substantive or theoretical understanding of that dimension and thus can be used in a purely statistical sense. For example, the latent traits in item response theory analysis are often introduced as latent variables but not associated with a particular construct until validity evidence supports such an association.

The object of validation has evolved with validity theory. Initially, validation was construed in terms of the validity of a test. Lee Cronbach and others pointed out that validity depends on how a test is scored. For example, detailed content coding of essays might yield highly valid scores whereas general subjective judgments might not. As a result, validity theory shifted its focus from validating tests to validating test scores. In addition, it became clear that the same test scores could be used in more than one way and that the level of validity could vary across uses. For example, the same test scores might offer a highly valid measure of intelligence but only a moderately valid indicator of attention deficit/hyperactivity disorder. As a result, the emphasis of validity theory again shifted from test scores to test score interpretations. Yet a valid interpretation often falls short of justifying a particular use. For example, an employment test might validly measure propensity for job success, but another available test might do as good a job at the same cost but with less adverse impact. In such an instance, the validity of the test score interpretation for the first test would not justify its use for employment testing. Thus, Samuel Messick has urged that test scores are rarely interpreted in a vacuum as a purely academic exercise but are rather collected for some purpose and put to some use. However, in common parlance, one frequently expands the notion of test to refer to the entire procedure of collecting test data (testing), assigning numeric values based on the test data (scoring), making inferences about the level of a construct on the basis of those scores (interpreting), and applying those inferences to practical decisions (use). Thus the term test validity lives on as shorthand for the validity of test score interpretations and uses.

...