MerQuR+Statistics Glossary

Statistics Glossary → Ramsey RESET test

Ramsey RESET testconcept

A general specification test of whether the functional form of a linear regression model is wrong, carried out by adding powers of the fitted values to the model.

After the model is fitted, the squares (ŷ²) and cubes (ŷ³) of the fitted values are added as new predictors, and a partial F test checks whether these terms are jointly significant. If the model is correctly specified they should add no explanatory power. If they do, the relationship may be curved, an interaction may have been omitted, or the outcome may need a transformation.

RESET says that there is a problem but not where it comes from; a new model should be built using residual plots, curvilinear terms or transformations. It is not a general test for omitted-variable bias; it can only detect omitted effects that are related to a nonlinear function of the existing predictors. Unlike the lack-of-fit test, which needs replicated observations, it can be applied to any data set.

An everyday example

A company fits a linear model of sales on advertising spend. The RESET test is significant. The residual plot shows that the gain in sales slows down as spending rises: beyond a certain point extra advertising brings almost no extra sales. A model using the logarithm of spending captures this saturation.

In MerQur

In multiple linear regression, the 'Extra output (2): RESET test' option tests for functional-form misspecification with a partial F after adding ŷ² and ŷ³.

Analyses where this concept appears

Multiple Linear Regression

Related concepts

Lack-of-fit test