Causality and Experiments

"These problems are, and will probably ever remain, among the inscrutablesecrets of nature. They belong to a class of questions radically inaccessible tothe human intelligence." —The Times of London, September 1849, on how cholerais contracted and spread

Does the death penalty have a deterrent effect? Is chocolate good for you? Whatcauses breast cancer?

All of these questions attempt to assign a cause to an effect. A carefulexamination of data can help shed light on questions like these. In this sectionyou will learn some of the fundamental concepts involved in establishingcausality.

Observation is a key to good science. An observational study is one in whichscientists make conclusions based on data that they have observed but had nohand in generating. In data science, many such studies involve observations on agroup of individuals, a factor of interest called a treatment, and anoutcome measured on each individual.

It is easiest to think of the individuals as people. In a study of whetherchocolate is good for the health, the individuals would indeed be people, thetreatment would be eating chocolate, and the outcome might be a measure of bloodpressure. But individuals in observational studies need not be people. In astudy of whether the death penalty has a deterrent effect, the individuals couldbe the 50 states of the union. A state law allowing the death penalty would bethe treatment, and an outcome could be the state’s murder rate.

The fundamental question is whether the treatment has an effect on the outcome.Any relation between the treatment and the outcome is called an association.If the treatment causes the outcome to occur, then the association is causal.Causality is at the heart of all three questions posed at the start of thissection. For example, one of the questions was whether chocolate directly causesimprovements in health, not just whether there there is a relation betweenchocolate and health.

The establishment of causality often takes place in two stages. First, anassociation is observed. Next, a more careful analysis leads to a decision aboutcausality.

This page was created by The Jupyter Book Community