Ever throw a dummy variable in a regression to see whether the effect the dummy variable is measuring has an impact on the dependent variable? Ever find that the dummy variable had the wrong sign, was of small magnitude or had vastly large variance so that you decided based on your data that the effect measured by the dummy variable had no effect? Of course you have, we all have.
But did you know that your data may be lying to you?
This presentation is an exploration of whether a time series changes based on an intervention that occurs half way through the data. Perhaps the intervention is a new law, or a treatment of some kind, did it have an effect. In our example, the dummy variable is insignificant in the first instance, model specification is in doubt and a full-on testing strategy is developed. That is, the test of whether D, the dummy variable, effects Y, the outcome measure is more than a p-test in a simple single regression, much more. Check out the classroom presentation and I will eventually load all the SAS code here to run all 8 regressions required and the multiple tests of each regression to answer the original hypothesis that D matters.