statsmodels.stats.outliers_influence.GLMInfluence

class statsmodels.stats.outliers_influence.GLMInfluence(results, resid=None, endog=None, exog=None, hat_matrix_diag=None, cov_params=None, scale=None)

Influence and outlier measures (experimental)

This uses partly formulas specific to GLM, specifically cooks_distance is based on the hessian, i.e. observed or expected information matrix and not on cov_params, in contrast to MLEInfluence. Standardization for changes in parameters, in fittedvalues and in the linear predictor are based on cov_params.

Parameters:

resultsinstance of results class

This only works for model and results classes that have the necessary helper methods.

residndarray, optional

Residuals to use instead of the default resid_pearson computed from the results instance.

endogndarray, optional

Endogenous (response) variable to use instead of results.model.endog.

exogndarray, optional

Exogenous (design) variable to use instead of results.model.exog.

hat_matrix_diagndarray, optional

Diagonal of the generalized leverage to use instead of the value computed internally.

cov_paramsndarray, optional

Covariance matrix of the parameter estimates to use instead of results.cov_params().

scalefloat, optional

Scale parameter to use instead of results.scale.

Attributes:

dbetas

change in parameters divided by the standard error of parameters from the full model results, bse.

d_fittedvalues_scaled

Change in fittedvalues scaled by standard errors

d_linpred

Change in linear prediction

d_linpred_scale

local change in linear prediction scaled by the standard errors for the prediction based on cov_params.

Methods

plot_index([y_var, threshold, title, ax, idx])	Index plot for influence attributes
plot_influence([external, alpha, criterion, ...])	Plot of influence in regression.
resid_score([joint, index, studentize])	Score observations scaled by inverse hessian
resid_score_factor()	Score residual divided by sqrt of hessian factor
summary_frame()	Creates a DataFrame with influence results

Notes

This has not yet been tested for correctness when offset or exposure are used, although they should be supported by the code.

Some GLM specific measures like d_deviance are still missing.

Computing an explicit leave-one-observation-out (LOOO) loop is included but no influence measures are currently computed from it.

Methods

plot_index([y_var, threshold, title, ax, idx])	Index plot for influence attributes
plot_influence([external, alpha, criterion, ...])	Plot of influence in regression.
resid_score([joint, index, studentize])	Score observations scaled by inverse hessian
resid_score_factor()	Score residual divided by sqrt of hessian factor
summary_frame()	Creates a DataFrame with influence results

Properties

cooks_distance	Cook's distance
d_fittedvalues	Change in expected response, fittedvalues
d_fittedvalues_scaled	Change in fittedvalues scaled by standard errors
d_linpred	Change in linear prediction
d_linpred_scaled	Change in linpred scaled by standard errors
d_params	Change in parameter estimates
dfbetas	Scaled change in parameter estimates
hat_matrix_diag	Diagonal of the hat_matrix for GLM
hat_matrix_exog_diag	Diagonal of the hat_matrix using only exog as in OLS
params_one	Parameter estimate based on one-step approximation
resid_studentized	Internally studentized pearson residuals