Causality

Our interest in statistics is often to use the observed data to inform us about a policy or treatment that we have not yet instituted or used. As a colon cancer patient I would like to know whether chemotherapy or radiotherapy is going to increase my probability of survival. As a state government we would like to know whether increasing mandatory education will make our students better off. If I am an NFL coach, I would like to know if passing instead of punting on 4th down will increase the likelihood of winning the game. In these cases we are interested in making an exogenous change to the environment and predicting the outcome.

In much of statistics, this idea has been captured by the term "causality". Often heard in the refrain, "correlation is not causation."

One assumption often made in statistics is that in order to make a claim about a causal relationship, we must observe an exogenous change. Judea Pearl more or less assumes this in his definition of causality. Some go even further and state that causal claims require randomized controlled trials. Economists generally do not place as much stock in randomized controlled trials as other applied statisticians (see for example, Sir Angus Deaton's discussion in the video below). Some seem to conflate causality with estimation of the average treatment effect.

Deaton, A.,2009, “Instruments of development: Randomization in the tropics, and the search for the elusive keys to economic development,” Proceedings of the British Academy, 2008 Lectures.

Deaton, Angus, and Nancy Cartwright, 2016, “Understanding and Misunderstanding Randomized Controlled Trials," UNPUBLISHED.

Imbens, G and D Rubin, 2015, Causal Inference for Statistics, Social and Biomedical Sciences: An introduction.

Pearl, J., 2009, Causality: Models, reasoning and inference.

Rubin, D, 1974, "Estimating Causal Effects of Treatments in Randomized and Nonrandomized Studies" in Journal of Educational Psychology.