A common exercise in applied microeconomics is to assess the stability of a coefficient under different choices of controls. This can be a tedious task for the econometrician, particularly with a large set of controls. More importantly, it's difficult to concisely convey this information to the reader. Regression tables are useful for displaying a limited number of models, but they're less useful for demonstrating how coefficient estimates evolve under dozens or even hundreds of sets of controls. starbility provides a simple interface to create a "coefficient stability plot" (also known as a "specification curve") allowing both the econometrician and the reader to assess coefficient stability under different combinations of controls.
starbility builds upon lfe and ggplot2, allowing for fast estimation of models with many groups of fixed effects and for flexible plotting. IV estimation, sample weights, clustered standard errors, and custom models are supported. See the main vignette for an introduction to starbility, and see the advanced vignette for a discussion of custom models, manual plotting, etc.
devtools::install_github('https://github.com/AakaashRao/starbility')
library(starbility)
Comments, criticism, suggestions, pull requests, etc. are very much appreciated. Email: arao@g.harvard.edu.
Thanks to David Yanagizawa-Drott for suggesting the structure of the plot and to Ross Mattheis and Eric Karsten for very helpful discussions during development.
Looking for a Stata implementation? Martin Andresen provides a package here and Hans H. Sievertsen provides a detailed example .ado file here.
Ariel Ortiz Bobea provides an excellent alternative R implementation (with nicer aesthetics and easier customization) here.