Nov 1, 2016
S. Stanley Young, Alan Karr
“Women who eat cereal while pregnant will have a boy!”
We all know this kind of science reporting is silly, right? The authors of this article argue that findings from observational studies are unlikely to be true. Full stop. Things we find out from looking at populations are rarely replicated in controlled studies.
They say we should be looking closer at the research process, like Deming with Japanese car manufacturers.
Their suggestion of an eight step research strategy which includes open data, written protocols before testing and journals accepting before final results, look really interesting.
Susan Athey suggested using machine learning and split data sets for economics here.