Causal Inferences and Fundamental Causation (Paperback)

The goals of identifying the effects of causes and measuring whether an intervention has an effect are central both to common life pursuits and to the sciences. Too often, because of the availability of sophisticated software and the advent of inexpensive computing power, people are content with running statistical analyses on data and assuming that the results are, within some generally set parameters, causally significant. The unwillingness to think carefully about the assumptions and methodologies is, I believe, a mistake. Indeed, while the Nobel Prize winning economist James Heckman is writing specifically about statistics, I believe his claim that the "existing literature on "causal inference" ... confuses three distinct tasks that need to be carefully distinguished" is generally true of all approaches to causal inference in which statistical analysis plays a central role. The three tasks Heckman believes people confuse are defining the set of hypotheticals, identifying parameters from hypothetical population data, and identifying parameters from real data. Using this framework, the purpose of this dissertation is to provide some guidance on the first two tasks Heckman identifies. Chapters 1, 2 and 3 fall, broadly, within the first task. More specifically, the goals of chapters 1 and 2 are to characterize, fully and clearly, what a causal inference is, and to provide a method for identifying statistical associations that merit close (statistical) study to determine whether they are also causal relationships. Chapter 3 focuses on one kind of cause that many people believe is especially important, fundamental cause. Finally, chapter 4 of the dissertation provides some guidance on Heckman's second task. The goal of chapter 4 is to identify the sorts of parameters needed to address confounding when using non-experimental, observational data. Collectively, then, the four chapters provide a propaedeutic to the third of Heckman's tasks - identifying parameters from real data; a task to which many people immediately jump while ignoring the critical importance of the first two.