October 3, 2017
In new research “Weighting for External Validity,” Economist Emily Oster explains that experiments are increasingly important within economics, and the advantage of an experiment is that it can produce a causal estimate of a treatment effect for the people in your experiment.
These results, however, are subject to concerns about external validity: let's say you want to expand your treatment to the whole population. How do you know that the other people in the population will react the same way your sample did? This is an especially hard problem when you think that people in your sample differ from the overall population in ways that you do not observe. Our paper provides a framework to think about when this is likely to lead to misleading estimates. We then develop a method to combine intuitive assumptions about the role of observed versus unobserved differences in your population and sample with information in the data to estimate bounds on the effect in the overall population.
To read more, click here.