Question

Using the Rubin Causal Model explain why it is difficult to identify a treatment effect for a single individual?

Answer 1

                        The Rubin causal model is a statistical analysis approach of cause and effect based on the potential outcomes. For example, it may be used to analyse the current income status of an individual and assume another situation for the individual to predict the possible potential output considering the different scenario. Since it is impossible to see multiple potential outcomes at the same time, it causes a dilemma of ‘fundamental problem of causal interference’. Due to this dilemma, unit level causal effects cannot be determined at once. But population level causal effects may be determined by randomized experiments. Thus, the average causal effect or the average treatment effect can only be obtained by computing the difference in means between the treated and the controlled samples. Also due to ethical or practical concerns, randomized experiments are not possible in many circumstances. Here, non-randomized selection and analysis may be conducted for causal inference such as propensity score matching. This method is used to correct the assignment mechanism by finding control units similar to treatment units.

                        The causal effects can be found only when there is a possible outcome for the same that is well defined in the case. Once the potential outcomes do not have a possibility of outcome, then it may be difficult to find the treatment effect for the same. It is in such cases that the finding of treatment effect becomes difficult. In the Rubin model, the treatment effect is defined in terms of two possible outcomes. Here, each unit would have one outcome that would result if the units were exposed to treatment and another outcome when the units are exposed to control. The treatment effect is the difference between these two outcomes. However, this is unobservable as the individual can only be exposed to treatment or control but not both at the same time. Thus, always there is a difficulty in exact calculation of the effects and thus an average of the effects is always considered.