|
Abstract:
|
Changepoint models are gaining popularity in ecological and environmental settings due to their ability to model thresholds and abrupt changes. While a variety of changepoint models are available, there are few studies evaluating them. Two such techniques, one based on a two-sample t-test and one based on reduction in model deviance, are discussed and their limited usefulness is investigated through simulation studies. Two competing methods are then explored via simulations: Horvath's likelihood method and a proposed bootstrap based technique. Both methods are shown to obtain close to nominal type I error rates while still having reasonable power over a range of scenarios. While Horvath's method is likelihood based, the bootstrap method is competitive in the majority of the considered cases and is superior in some. The methods are then compared on soil chemistry data from the Everglades.
|
