Capabilities

Capabilities(
    covariance_update,
    crv3_inference,
    hac_inference,
    multiway_clustering,
    wildboottest,
    cluster_causal_variance,
    decomposition,
    prediction,
    prediction_inference,
    fixed_effect_recovery,
    randomization_inference,
    sherman_morrison_update,
    anytime_valid_inference,
)

Inference and post-estimation features a fitted model supports.

Each model class declares one value, so a method can reject an unsupported estimator before it reads any estimation state rather than reinterpreting another estimator’s arrays.

Parameters

Name Type Description Default
covariance_update bool Whether post-estimation vcov() updates are available. did2s() fits require refitting with the desired cluster variable instead. required
crv3_inference bool Whether vcov() accepts "CRV3". The jackknife refits the model on leave-one-cluster-out samples, so it is restricted to estimators whose refit replays the original estimation contract. required
hac_inference bool Whether vcov() accepts the Newey-West and Driscoll-Kraay estimators. required
multiway_clustering bool Whether vcov() accepts more than one cluster variable. required
wildboottest bool Whether wildboottest() can resample the fit. Unweighted OLS only. required
cluster_causal_variance bool Whether ccv(), the causal cluster variance estimator, is available. required
decomposition bool Whether decompose(), the Gelbach decomposition, is available. required
prediction bool Whether predict() is available. IV and did2s() fits do not support it. required
prediction_inference bool Whether predict(se_fit=True) and predict(interval="prediction") are available. Only OLS fits support them, additionally requiring no weights or fixed effects. required
fixed_effect_recovery bool Whether the model class can recover fixed-effect estimates through fixef(). IV, quantreg(), and did2s() fits do not support it. fixef() additionally requires the fit to have fixed effects. required
randomization_inference bool Whether ritest(), randomization inference, is available. Only feols() fits without instruments and fepois() fits support it; did2s() and event_study() fits do not. required
sherman_morrison_update bool Whether update() can append observations to the fit through a Sherman-Morrison update of the OLS coefficients. Only feols() fits without instruments support it; did2s() and event_study() fits do not. required
anytime_valid_inference bool Whether evalue(), pvalue_savi(), and confint(inference_type="savi"), safe anytime-valid inference, are available. Only feols() fits without instruments support it; did2s() and event_study() fits do not. SAVI additionally requires no weights, no fixed effects, and an iid or heteroskedasticity-robust covariance. required

Examples

import pyfixest as pf

pf.feols("Y ~ X1 | f1", pf.get_data()).capabilities
Capabilities(covariance_update=True, crv3_inference=True, hac_inference=True, multiway_clustering=True, wildboottest=True, cluster_causal_variance=True, decomposition=True, prediction=True, prediction_inference=True, fixed_effect_recovery=True, randomization_inference=True, sherman_morrison_update=True, anytime_valid_inference=True)
pf.fepois("Y ~ X1 | f1", pf.get_data(model="Fepois")).capabilities
Capabilities(covariance_update=True, crv3_inference=True, hac_inference=True, multiway_clustering=True, wildboottest=False, cluster_causal_variance=False, decomposition=False, prediction=True, prediction_inference=False, fixed_effect_recovery=True, randomization_inference=True, sherman_morrison_update=False, anytime_valid_inference=False)