Causal Inference Background and Review Potential Outcomes and Average Treatment Effect In general, we can measure the causal effect of a binary treatment on an outcome by considering the potential outcomes and . Recall that these are potential outcomes: they represent thought experiments about what would happen if the treatment was or wasn't applied. In the real world, we only ever get to observe one of them for any individual, depending on whether that unit received the treatment or not. We defined the average treatment effect (ATE), represented by the Greek letter tau ( ), as: This represents the causal effect of a treatment on an outcome . We saw that in general, we were unable to estimate this without making assumptions. In the special case where our data come from a randomized experiment, then we saw that the Simple Difference in Observed group means (SDO) was an unbiased estimate of the ATE