statsmodels.stats.outliers_influence.MLEInfluence.resid_score

MLEInfluence.resid_score(joint=True, index=None, studentize=False)[source]

Score observations scaled by inverse hessian.

Score residual in resid_score are defined in analogy to a score test statistic for each observation.

Parameters:
jointbool

If joint is true, then a quadratic form similar to score_test is returned for each observation. If joint is false, then standardized score_obs are returned. The returned array is two-dimensional

indexndarray (optional)

Optional index to select a subset of score_obs columns. By default, all columns of score_obs will be used.

studentizebool

If studentize is true, the the scaled residuals are also studentized using the generalized leverage.

Returns:
array1-D or 2-D residuals

Notes

Status: experimental

Because of the one srep approacimation of d_params, score residuals are identical to cooks_distance, except for

  • cooks_distance is normalized by the number of parameters

  • cooks_distance uses cov_params, resid_score is based on Hessian. This will make them differ in the case of robust cov_params.

Last update: Oct 03, 2024