ABSTRACT
Connections With Other Chapters
In chapter 2, the most influential philosophical theories of measurement were discussed. Although these theories are important to psychometrics, they do not directly guide psychometric practice. Instead, this role is played by statistical modeling approaches. This chapter discusses the most important of these models: classical test theory, modern test theory, and generalizability theory. The relation between the construct of interest and the observed test score is conceptualized differently in each of these models. We link these differences to validity theory. In addition, the relation of validity to important psychometric concepts, such as reliability and measurement invariance, is analyzed.
The previous chapter discussed the properties that observations should have in order to be considered measurements, according to different conceptions of what measurement is. Although such theories are important, in psychometric practice they play a background role. In the foreground, psychometric practice is guided by a more statistical approach to test scores. That is, psychometrics generally starts off from the interpretation of the observations as data, for which an adequate statistical model should be found. This approach gave birth to the major psychometric models currently used: the classical test theory model, the modern test theory model, and the generalizability model. Each of these approaches could be described as data-oriented, in the sense that their main use lies in the analysis of empirical data (rather than, say, in characterizing tests theoretically or in proving properties of measurement structures). As such, they are heavily informed by practical concerns.