Statistics Glossary → Breusch-Pagan and White tests
Breusch-Pagan and White testsconcept
Tests of whether the variance of regression residuals changes with the explanatory variables (heteroscedasticity).
In the Breusch-Pagan test the squared residuals are regressed on the explanatory variables; if this auxiliary regression explains a lot, the variance is not constant. In the common (Koenker) version the statistic is LM = n·R², compared with a chi-square distribution whose degrees of freedom equal the number of explanatory variables; this version is more robust to non-normality. White's test also adds the squares and cross-products of the explanatory variables to the auxiliary regression, so it catches more general patterns, but its degrees of freedom grow quickly with many predictors.
A significant result does not mean that the coefficients are biased; it means that the classical standard errors, confidence intervals and p values cannot be trusted. Remedies include heteroscedasticity-consistent (HC) standard errors, weighted least squares or a transformation of the dependent variable. The test can also come out significant when the model is misspecified (for example, a missing curved term). In panel data, the Breusch-Pagan LM test of the same name is a different test, for random unit effects.
An everyday example
The relationship between household income and food spending is examined. Low-income households spend similar amounts, while high-income households vary widely: some eat at expensive restaurants, others are frugal. Because the spread of the residuals grows with income, the Breusch-Pagan test is significant and robust standard errors are used.
In MerQur
Under 'Model diagnostics: Heteroscedasticity tests (Breusch-Pagan, White)' in Multiple Linear Regression, Two-Way ANOVA and ANCOVA, in its original, Koenker and F forms; and among the 'OLS assumption diagnostics (Extra output)' of SAR. The Breusch-Pagan LM reported in Panel Data Regression is the unit-effects test.
Analyses where this concept appears
Two-Way ANOVAANCOVAMultiple Linear RegressionPanel Data RegressionSAR (Spatial Autoregressive Lag)
Related concepts
HeteroscedasticityRobust (sandwich) standard errors (HC0-HC3)