ABSTRACT
Connections With Other Chapters
Chapter 5 considered behavior domains as an alternative to causal measurement models. Chapter 6 considered various causal measurement models. A primary distinction from chapter 6 involves the distinction between reflective measurement models in which latent variables cause their indicators and formative measurement models in which indices cause their latent variables. The present chapter focuses primarily on the former, leaving formative measurement models and various alternatives to chapter 8. The present chapter also develops the distinction between statistically unidimensional models and causally unidimensional models mentioned in chapter 6. Statistical models need not be causally unidimensional. As noted in chapter 1, we adopt from the literature the term ‘measurement model’ but use this to describe assessment in general whether or not it involves the types of quantitative structure required by a strict sense of measurement (reviewed in chapter 2). Chapter 10 further develops the idea of interpretation and develops a rough criterion for deciding what interpretation to validate. Chapter 12 further applies material from this chapter to the process of test validation.
The methodological literature in the behavioral sciences generally discusses causation without broaching the issue that theories of causation attempt to answer. What do causal assertions mean when they refer to causation? Yet the answer to this question is crucial to understanding what a causal claim entails, and thus how one can test it empirically. The present chapter will consider some major approaches to understanding causation from the causation literature and relate them to issues of assessment. Different understandings of the term ‘causation’ differ from one another in at least two respects. They differ in terms of what they assume. They also differ in terms of what they deliver with respect to what causal assertions assert and what they entail. The basic strategy of the chapter will unfold as follows. First, a purely statistical, non-causal interpretation of measurement models provides a baseline for comparison. Then an evaluation of what a causal interpretation might add to a non-causal interpretation provides some sense of what one might want from a causal interpretation. What one might want then provides a framework for contrasting various causal interpretations.