Natural Experiments

This year's Nobel Prize in Economics went to several scientists for their role in developing a research design called "natural experiments". 

Natural experiments are extremely useful in situations where controlled experiments are unethical or impractical, such as in epidemiology (determining the health impact of exposure to a disease) or economic research (estimating return on schooling in US adults). 

Good. But what are natural experiments? 

There is big math-y logic behind part of the description, and I won't pretend to explain that. 

But the research-design part is straightforward. 

A natural experiment is an empirical study where subjects are exposed to test and control conditions in a way that is determined by nature.

That is, by factors outside the control of the investigators, but still in a process that resembles random assignment.

Natural experiments are possible when there is a clearly-defined exposure involving a well-defined subpopulation (and absence of exposure in a similar subpopulation) where changes in outcome can be attributed to the exposure.

The difference between natural experiments and other types of observational studies is that "experiments" make causal inference possible. An arrow can be drawn from cause to effect.

In my media research methods class in graduate school (not THAT long ago) we learned that observational approaches - interviewing, focus groups and ethnographies - were typically good for exploratory and descriptive research but not for establishing causality.

Time to update my models.

More here ...  https://en.wikipedia.org/wiki/Natural_experiment