In many applications of statistics, a large proportion of the questions of interest are fundamentally questions of causality rather than simply questions of description or association. For example, a medical researcher may wish to find out whether a new drug is effective against a disease. An economist may be interested in uncovering the effects of a job-training program on an individual's employment prospects, or the effects of a new tax or regulation on economic activity. A sociologist may be concerned about the effects of divorce on children's subsequent education. In this text we discuss statistical methods for studying such questions.
The book arose out of a conversation we had in 1992 while we were both on the faculty at Harvard University. We found that although we were both interested in questions of causality, we had difficulty communicating our ideas because, coming from different disciplines, we were used to different terminology and conventions. However, the excitement about the ideas in these different areas motivated us to capitalize on these difficulties, which led to a long collaboration, including research projects, graduate and undergraduate teaching, and thesis advising. The book is a reflection of this collaboration.
The book is based directly on many semester and quarter-length courses we, initially jointly, and later separately, taught for a number of years, starting in 1995 at Harvard University, followed by the University of California at Los Angeles, the University of California at Berkeley, and Stanford University, to audiences of graduate and undergraduate students from statistics, economics, business, and other disciplines using applied statistics. In addition we have taught shorter versions of such courses in Barcelona, Beijing, Berlin, Bern, Geneva, Maastricht, Mexico City, Miami, Montevideo, Santiago, Stockholm, Uppsala, Wuppertal, Zurich, and at the World Bank as well as other associations and agencies.
There are a number of key features of the approach taken in this book. First of all, the perspective we take is that all causal questions are tied to specific interventions or treatments. Second, causal questions are viewed as comparisons of potential out-comes, with each potential outcome corresponding to a level of the treatment.