The article is here. The figure above demonstrates the non-fraud side of the problem. Central to the problem is the concept of statistical power.  A lot of published science studies are statistically underpowered. Kudos to The Economist for publishing a serious and understandable piece about this problem.