1. Most empirical models that employ a treatment-effect strategy to test the impact of particular policies are based on policy environments in which there has been a clear binary change in a policy that has affected a well-defined, exogenously determined subset of the population and thus a clearly distinguishable before and after.
2. The rate of U.S. natural gas production from shale has risen rapidly over the last decade, growing from 4.1 percent of total U.S. production in 2005 to 23.1 percent in 2010 (Wang and Krupnick 2013).
3. For example, in 2013 in the midst of a (failed) re-election campaign, Governor Tom Corbett claimed that 200,000 Pennsylvanians had jobs or were made more prosperous because of the industry (T. Puko, Pittsburgh Tribune Review, November 14, 2013).
4. We thank an anonymous reviewer for pointing out that our DDD terminology is the same as saying that we are estimating a difference-in-differences model in growth rates instead of levels. Given the close empirical connection between our models and those of Banzhaf and Lavery (2010), we continue to use DDD terminology but this key distinction should be kept in mind.