Potential Outcomes

In his 1974 paper published in the Journal of Educational Psychology, Don Rubin, developed a model for evaluating "treatments". A treatment may be a new drug like the Kadcyla drug tested in the Emilia trial or it may be policy like subsidizing college attendance or it may be a new mathematics curriculum for elementary school. In Rubin's model each person or "unit" can have a potential outcome associated with each possible treatment.

Consider a patient participating in the Emilia trial. Prior to randomization, but after acceptance into the trial, the patient has two potential treatments. They may receive Kadcyla or they may receive the more standard chemotherapy. Associated with each potential treatment is a potential outcome. That potential outcome may be whether or not they are alive two years after beginning the trial. If we are considering a policy associated with increasing college attendance, we are interested in the potential treatment of attending college or not attending college. We can associated income in ten years with each of these potential treatments. A person who has the potential to attend college has a potential income in ten years of attending college and a potential income in ten years of not attending college.

What is the effect of Kadcyla? It is the difference in the potential outcomes. We can see that Kadcyla increases a person's survival if that person's potential outcome on Kadcyla is alive at 2 years, while that person's potential outcome on the standard therapy is that they passed away prior to reaching the 2 year mark.

What is the effect of attending college? it is the difference in potential income ten years later. Attending college reduces someone's ten year income if their potential income from attending college is lower than their potential income from not attending college.

The treatment effect of a particular drug or policy is measured by the difference in potential outcomes. Some say that a treatment has a "causal effect" if there is difference in potential outcomes.