ABSTRACT
Connections With Other Chapters
Chapter 6 distinguished formative from reflective measurement models as formal statistical models. Chapter 6 also discussed the idea of network measurement models. Chapter 7 discussed alternative non-causal and causal interpretations of reflective models in which constructs cause item responses (or their probability distributions). That chapter developed three advantages of a causal interpretation over a non-causal interpretation: reason to believe that changes in the construct will produce changes in the item responses, reason to believe that the patterns of statistical association will persist, and reason to believe that the measurement model will remain robust to certain forms of intervention on the population of interest. Chapter 7 also presented three broad approaches to causation: regularity theories, counterfactual theories, and causal process theories. As in chapter 7, these terms here refer to reductive theories of causation. Regularity theories reduce causation to patterns of succession in actual populations. Counterfactual theories reduce causation to counterfactual dependencies that hold across both actual and possible counterfactual populations. As noted in chapter 7, some well-known methodological approaches to causal inference make use of the concept of counterfactuals but do not provide a reductive account of causation. Process theories reduce causation to actual processes linking causes to effects. As elsewhere, the term ‘measurement model’ is adopted from the literature to apply broadly to all forms of assessment (chapter 1) and not just measures with quantitative structure (chapter 2). As discussed in chapter 1, the term ‘indices’ refers to observed variables in a formative model, in contrast to indicators in a reflective model.
The first half of the present chapter considers formative measurement models. Similar advantages to those found in reflective models also apply to formative models. However, formative models face two key challenges: causal misspecification and a seeming proliferation of causal composite variables. Different causal interpretations handle these challenges differently. A latter section briefly discusses a causal interpretation appropriate for network models. The last portion of the chapter focuses on causal interpretations appropriate to individual differences. Taken together, these topics canvass a number of open questions with respect to causation and assessment. Nonetheless, the chapter by no means exhausts the problems waiting to be solved. A box at the end summarizes some key open questions from the chapter.