Interesting Paper on Research Design
| Peter Klein |
Ed Leamer famously argued, back in 1983, that empirical economists should do more sensitivity analysis. A new NBER paper by Joshua Angrist and Jörn-Steffen Pischke says that econometric practice has indeed gotten much better, not because of sensitivity analysis, but because of a new focus on research design. “[T]he credibility revolution in empirical work can be traced to the rise of a design-based approach that emphasizes the identification of causal effects. Design-based studies typically feature either real or natural experiments and are distinguished by their prima facie credibility and by the attention investigators devote to making the case for a causal interpretation of the findings their designs generate.” They are clearly right that identification has become a Really Big Deal (choosing a dissertation topic in economics is sometimes referred to these days as “the search for a good instrument”). But natural experiments and instrumental variables have their own potential problems as well. Perhaps Verstehen can still play a role.