Replication Data for: Treating Time With All Due Seriousness

Time series techniques see widespread use in political science. In De Boef and Keele (2008) we outlined a set of statistical methods for stationary data. Those methods have come to be widely used. Grant and Lebo (2015) contend that one of the methods we discussed, the error correction model, should generally not be used with political data. They argue that the error correction model leads to both interpretational and inferential mistakes by applied analysts. While we agree with their statements about equation balance, we show that the error correction model leads to the same inferences as the autoregressive distributed lag model when the data are stationary. We also demonstrate that careful use of an error correction model can help diagnose model misspecification when the equation is unbalanced. Such techniques are useful since pretesting for integration and fractional integration is often a highly uncertain process, which we demonstrate through a simulation exercise. We also highlight two related but often ignored complications in time series: low power and overfitting. We argue that the statistical tests used in time series analyses have little power to detect differences in many of the sample sizes typical in political science. Moreover, given small sample sizes, many analysts overfit their time series models. We argue that the results in the Grant and Lebo replications stem from inadequate sample sizes that make it difficult to conclusively use any time series model.