CIMissTest

The main test class. Compares classifier loss for predicting missingness with and without the outcome variable, combining results across multiple imputations and cross-validation folds.

CIMissTest(dataset, imputer=MidasImputer, classifier=RFClassifier, m=10, n_folds=10, classifier_args={}, imputer_args={}, random_state=42, target_level='variable', variance_method='mi_crossfit', subsample_cap=2000)

Conditional-independence-of-missingness test.

Compares classifier loss for predicting the missingness indicator with and without the outcome variable. A significant difference implies missingness is not conditionally independent of the outcome.

Parameters:

Name	Type	Description	Default
dataset	Dataset	Populated Dataset (call Dataset.make first).	required
imputer	type[Imputer]	Imputer class to use (default MidasImputer).	MidasImputer
classifier	type[CIClassifier]	Classifier class to use (default RFClassifier).	RFClassifier
m	int	Number of multiply-imputed datasets.	10
n_folds	int	Number of cross-validation folds.	10
classifier_args	dict	Extra keyword arguments forwarded to the classifier.	{}
imputer_args	dict	Extra keyword arguments forwarded to the imputer.	{}
variance_method	str	'mi_crossfit' (default) or 'legacy_fold'.	'mi_crossfit'
subsample_cap	int or None	Maximum number of rows to subsample for testing. Set to None to disable subsampling (default 2000).	2000

run()

Run the conditional independence test

Having declared the test object, this method will run the test

summary()

Print a summary of the test results