World Bank Group

11/10/2022 | Press release | Distributed by Public on 11/10/2022 08:52

Weekly links November 10: causal inference from text data, exposure designs, being more creative, and

· Stanford Grit and Growth podcast with Steve Anderson about three of his experiments with businesses in different African countries: "you'll hear why having a coach, getting classroom training, and learning how to delegate can drive growth and impact your bottom line".· Doing causal inference from text-based data - a new paper in Science Advances discusses issues that arise and recommends a split-sample approach. "Texts are complex, high-dimensional objects; thus, researchers must find simpler, lower-dimensional representations for their texts to use them in scientific analyses. This simplification can be intuitive and familiar. For example, we might take a collection of emails and divide them into "spam" and "not spam." We call the function that maps the documents into our measure of interest g…Our workflow highlights an identification and estimation problem that arises from a common source-using the same documents to both discover measures and estimate causal effects. By using the same documents to discover g and estimate effects, the analyst creates a problem where the categories obtained depend on the particular randomization. Consequently, each randomization could create a different codebook function g….researchers can simply divide one dataset into a training set for discovering measures and a test set for estimating causal effects"· In the latest issue of the Journal of Economic Perspectives, Liyang Sun and Jesse Shapiro provide a short introduction to analyzing exposure designs - e.g. when there is a national policy change, but some regions are more exposed to it than others, and the new literature on the problems with two-way fixed effects models for this when there are heterogeneous impacts - and they note when there is no totally unaffected region or unit, then no estimator can be guaranteed to be centered in a range around the true coefficients. · On the Brookings blog, Susanna Loeb and Carly Robinson discuss the good and bad of virtual on-demand tutoring, based on an experiment in middle schools in California: "take-up was low. Only 19% of students ever logged on and received help from a tutor. What's more, struggling students were only half as likely to ever use the resource than their higher-achieving peers…We tested the effectiveness of the program and found positive effects for the students who used it. Accessing the opt-in tutoring increased the likelihood of students passing all of their courses." They then did encouragement experiments to try to increase take-up, and found that engaging both students and parents simultaneously was the most effective, and helped get struggling students to use the tutoring.· Ethan Mollick shares a lot of interesting research findings on twitter. He has now started a substack, with his first post on how to be …more creative. "Among entrepreneurs, within-person differences in creativity over the course of a week are greater than across-person differences. That means that maximizing your own creativity may be more important than being a "naturally creative" person. So how can you maximize your creativity? The paper finds the two key determinants are sleep and time to reflect….Aside from rest, reflection and a wandering mind seem to help make you more creative. A study of scientists and artists found that 20% of their breakthrough ideas came while their mind was wandering, resulting in surprise "aha!" moments. Good ideas may actually come while commuting, letting your mind wander, and while showering" - not sure how credible many of these findings are, but an interesting starting point and justification for going to bed early, going for a run, then showering and calling it working on your next paper.· The Institute for Replication is looking for PhD students (and post-docs and professors) interested in taking part in a one-day "Replication games" either at the University of Calgary or virtually on November 26. Participants will get co-authorship on a meta-paper combining the replications. See here for details and to apply.· Conference call for papers: The MidWest International Economic Development Conference has submissions due Jan 5, and seems to have expanded the MidWest to now include Houston - which I guess is Florida's MidWest.