statsmodels.stats.diagnostic.linear_reset

statsmodels.stats.diagnostic.linear_reset(res, power=3, test_type='fitted', use_f=False, cov_type='nonrobust', cov_kwargs=None)[source]

Ramsey’s RESET test for neglected nonlinearity

Parameters:
resRegressionResults

A results instance from a linear regression.

power{int, List[int]}, default 3

The maximum power to include in the model, if an integer. Includes powers 2, 3, …, power. If an list of integers, includes all powers in the list.

test_typestr, default “fitted”

The type of augmentation to use:

  • “fitted” : (default) Augment regressors with powers of fitted values.

  • “exog” : Augment exog with powers of exog. Excludes binary regressors.

  • “princomp”: Augment exog with powers of first principal component of exog.

use_fbool, default False

Flag indicating whether an F-test should be used (True) or a chi-square test (False).

cov_typestr, default “nonrobust

Covariance type. The default is “nonrobust` which uses the classic OLS covariance estimator. Specify one of “HC0”, “HC1”, “HC2”, “HC3” to use White’s covariance estimator. All covariance types supported by OLS.fit are accepted.

cov_kwargsdict, default None

Dictionary of covariance options passed to OLS.fit. See OLS.fit for more details.

Returns:
ContrastResults

Test results for Ramsey’s Reset test. See notes for implementation details.

Notes

The RESET test uses an augmented regression of the form

\[Y = X\beta + Z\gamma + \epsilon\]

where \(Z\) are a set of regressors that are one of:

  • Powers of \(X\hat{\beta}\) from the original regression.

  • Powers of \(X\), excluding the constant and binary regressors.

  • Powers of the first principal component of \(X\). If the model includes a constant, this column is dropped before computing the principal component. In either case, the principal component is extracted from the correlation matrix of remaining columns.

The test is a Wald test of the null \(H_0:\gamma=0\). If use_f is True, then the quadratic-form test statistic is divided by the number of restrictions and the F distribution is used to compute the critical value.