Statistics Glossary → Change-point tests (Pettitt, Buishand, SNHT)
Change-point tests (Pettitt, Buishand, SNHT)concept
Tests that determine whether the mean of a time series shifts abruptly at an unknown time and, if so, when that shift most likely occurred.
Pettitt's test is rank-based and nonparametric: for each possible break point it compares the parts of the series before and after with a Mann-Whitney-type statistic and proposes the time with the largest difference as the change point. Buishand's range test is based on cumulative sums of the deviations from the mean. Alexandersson's standard normal homogeneity test (SNHT) compares the standardized means before and after each possible point. Buishand and SNHT assume normality; SNHT is more sensitive to breaks near the beginning and end of the series.
These tests look for a single change point and assume independent observations. In autocorrelated series they often flag breaks that do not exist as 'significant'. A steady, gradual trend can also look like a shift, so the results should be read together with trend tests such as Mann-Kendall and with a plot. They are widely used in climate data to detect events that break homogeneity, such as a station move or a change of instrument.
An everyday example
In a weather station's 50-year series of annual mean temperatures, values suddenly rise by about 0.8 °C after 1987. Pettitt's test points to 1987 as the most likely change point, and the records reveal that the station was moved to the city centre that year.
In MerQur
In Mann-Kendall + Sen's Slope, the 'Additional output: Change point' option gives the Pettitt, Buishand and SNHT tests.