by Angela Guess
Jon Levin, a professor of economics at Stanford, recently wrote in Forbes, “Machine learning methods are really powerful for fitting predictive models and for doing classification on large-scale, high-dimensional data. These are the data we increasingly use in economics. So I think there’s no doubt many machine learning methods will get used more and more often. One area that’s going to get a lot of attention is combining machine learning with causal inference. A big fraction of empirical microeconomics is about finding ways to exploit natural experiments, whether by using instrumental variables, regression discontinuity, matching, difference-in-difference estimators, or other methods.”
Levin goes on, “Large-scale data has great advantages in terms of finding natural experiments (to take a trivial example, if you want to measure how a July 15 price change affected sales, it’s much more powerful to have daily sales data than monthly sales data). But for the most part economists trying to estimate causal models on large-scale data are using traditional methods like fixed effects linear regression. Having some easy to use alternatives would probably make a significant difference in empirical research.”
Photo credit: Flickr