statsmodels.stats.outliers_influence.MLEInfluence¶
- class statsmodels.stats.outliers_influence.MLEInfluence(results, resid=None, endog=None, exog=None, hat_matrix_diag=None, cov_params=None, scale=None)[source]¶
Global Influence and outlier measures (experimental)
- Parameters
- results
instance
of
results
class
This only works for model and results classes that have the necessary helper methods.
- other arguments :
Those are only available to override default behavior and are used instead of the corresponding attribute of the results class. By default resid_pearson is used as resid.
- results
Notes
MLEInfluence uses generic definitions based on maximum likelihood models.
MLEInfluence produces the same results as GLMInfluence for canonical links (verified for GLM Binomial, Poisson and Gaussian). There will be some differences for non-canonical links or if a robust cov_type is used. For example, the generalized leverage differs from the definition of the GLM hat matrix in the case of Probit, which corresponds to family Binomial with a non-canonical link.
The extension to non-standard models, e.g. multi-link model like BetaModel and the ZeroInflated models is still experimental and might still change. Additonally, ZeroInflated and some threshold models have a nondifferentiability in the generalized leverage. How this case is treated might also change.
Warning: This does currently not work for constrained or penalized models, e.g. models estimated with fit_constrained or fit_regularized.
This has not yet been tested for correctness when offset or exposure are used, although they should be supported by the code.
status: experimental, This class will need changes to support different kinds of models, e.g. extra parameters in discrete.NegativeBinomial or two-part models like ZeroInflatedPoisson.
- Attributes
- hat_matrix_diag (hii)
This
is
the
generalized
leverage
computed
as
the
local derivative of fittedvalues (predicted mean) with respect to the observed response for each observation. Not available for ZeroInflated models because of nondifferentiability.
- d_params
Change
in
parameters
computed
with
one
Newton
step
using
the
full Hessian corrected by division by (1 - hii). If hat_matrix_diag is not available, then the division by (1 - hii) is not included.
- dbetas
change
in
parameters
divided
by
the
standard
error
of
parameters
from the full model results,
bse
.- cooks_distance
quadratic
form
for
change
in
parameters
weighted
by
cov_params
from the full model divided by the number of variables. It includes p-values based on the F-distribution which are only approximate outside of linear Gaussian models.- resid_studentized
In
the
general
MLE
case
resid_studentized
are
computed from the score residuals scaled by hessian factor and leverage. This does not use
cov_params
.- d_fittedvalues
local
change
of
expected
mean
given
the
change
in
the
parameters as computed in
d_params
.d_fittedvalues_scaled
same
as
d_fittedvalues
but
scaled
by
the
standard
Change in fittedvalues scaled by standard errors.
- params_one
is
the
one
step
parameter
estimate
computed
as
params
from the full sample minus
d_params
.
- hat_matrix_diag (hii)
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.
Score residual divided by sqrt of hessian factor.
Creates a DataFrame with influence results.
Properties
Cook's distance and p-values.
Change in expected response, fittedvalues.
Change in fittedvalues scaled by standard errors.
Approximate change in parameter estimates when dropping observation.
Scaled change in parameter estimates.
Diagonal of the generalized leverage
Diagonal of the hat_matrix using only exog as in OLS
Parameter estimate based on one-step approximation.
studentized default residuals.