Applied economics studies target effects that can be relatively small. This dissertation delves into some statistical obstacles to the accurate estimation of such effects, with a particular focus on the concepts of statistical power and exaggeration---imprecise studies tend to produce inflated estimates of the effect of interest. It explores implications of low power and exaggeration that are specific to applied economics studies and their design.
Through the example of studies on the acute health effects of air pollution, the first chapter identifies tangible drivers of exaggeration that extend beyond small effects and a limited sample size. This analysis uncovers an overarching mechanism, studied in Chapter 2, that induces exaggeration when using causal identification strategies. This subsequent chapter emphasizes that causal approaches only focus on a subset of the variation---the exogenous part---reducing the precision of the study and increasing risks of exaggeration.
The final chapter further broadens the discussion to analyze design choices in light of the multiple goals of causal inference studies; these studies aim not only to identify an average effect but also differentiated effects across subgroups, as well as producing insights that extend beyond the population considered. Overall, this dissertation underlines the manifold implications of design choices on non-experimental economic studies, with the aim of contributing to more accurate estimations of effects to better inform policymaking.
| Identifer | oai:union.ndltd.org:columbia.edu/oai:academiccommons.columbia.edu:10.7916/we1k-9w19 |
| Date | January 2024 |
| Creators | Bagilet, Vincent |
| Source Sets | Columbia University |
| Language | English |
| Detected Language | English |
| Type | Theses |
Page generated in 0.0017 seconds