Statistics Glossary → Chow structural break test
Chow structural break testconcept
An F test of whether a regression relationship is the same before and after a pre-specified point (e.g. a date).
The data are split into two parts, before and after the break point. First a single model is fitted to all the data (the restricted model), then a separate model is fitted to each part. If the separate models reduce the residual sum of squares significantly, the relationship has changed at the break point: F = [(RSS_all - (RSS₁ + RSS₂))/k] / [(RSS₁ + RSS₂)/(n - 2k)], where k is the number of parameters including the intercept.
The Chow test assumes that the break point was chosen without looking at the data, from an external event (a change in the law, a crisis, a policy). If the point is chosen after looking at the data, where the largest difference appears, the test comes out significant too often; for unknown break points, supremum tests such as Quandt-Andrews or change-point methods are used. The test also assumes equal error variance in the two periods.
An everyday example
Public transport use in a city has been rising steadily for years, and a new metro line opens in 2020. The Chow test checks whether the trend line (intercept and slope) stayed the same before and after the opening. It gives F = 9.4 and p < 0.001: after the opening both the level and the growth rate changed.
In MerQur
In the Time Series analysis the 'Chow break observation' option tests, with the classical F test, whether there is a structural break after the given observation in the linear trend y = a + b·t. In multiple linear regression it is also available under 'Model diagnostics' as the 'Chow break test' (None, Observation order, Grouping variable).
Analyses where this concept appears
Multiple Linear RegressionTime Series