Measurement Variance

Abstract

AbstractThis chapter covers the concept of measurement variance and construct validity. For a test to be valid, measurements must vary and they must vary in specific ways. First, scores for test items that measure the same construct should be highly related. Second, scores for test items that measure different constructions should be weakly related. When this pattern is found, the items are said to have construct validity. Next, the same patterns of relationships among test items should be found across subgroups. If the relationships among test items is different across subgroups, then the test is said to have measurement variance or measurement inequivalence. This is a form of test bias. While graphs can be used to illustrate measurement variance, ultimately the process is statistical. Confirmatory factor analysis is the statistical technique that is commonly used to explore construct validity and measure variance.