Abstract
A researcher choosing between models ordered by some criterion may seek the best specification that satisfies a testable assumption. In this scenario, sequential specification tests (SSTs) are hypothesis tests of that assumption for each model in the sequence. We introduce a method using the p-values from SSTs to estimate the point in the sequence where the assumption ceases to hold. Unlike alternative approaches, this method is robust to individual errant p-values and does not require choosing a test level or tuning parameter. We demonstrate the method’s properties with a simulation study, and illustrate it by choosing a bandwidth in a regression discontinuity design and a lag order for a time series model.
2010 Mathematics Subject Classification:
Notes
1 This very helpful citation was suggested by an anonymous reviewer.