Abstract
A statistical test for detecting a change in the behavior of an annual temperature series is presented. The test is based on the two-phase regression model. By trading the hypothesized time of change as an unknown parameter, the approach allows an inference to be made about the time of change. The approach also avoids a serious problem, called data-dredging, that can arise in testing for change occurring at a specified time. The test is applied to a series of Southern Hemisphere temperatures, and the hypothesis of no change cannot be rejected.
