John Ioannidis has an interesting article in the L.A. Times titled “Economics isn’t a bogus science — we just don’t use it correctly.” Some excerpts are below:
Most published studies use limited data. By a conservative estimate, the average study has 18% power to detect a modest association if one exists. Due to this low power of prediction, researchers could easily miss a genuine association. Or, they could declare a spurious, non-existing association, having been led astray by small amounts of bias.
Not all the news is bad, however.
Thankfully, economists are increasingly turning to experimental methods, which have the best reproducibility record. According to one evaluation, two-thirds of experimental studies were fully reproducible when other researchers tried to repeat them.
Several economics journals, moreover, are now employing standards that are likely to enhance transparency and reproducibility. These journals require researchers to share all of their protocols, raw data, software and code.
Do read the whole article.