iArt.test
- iArt.test(*, Z, X, Y, G='iterative+linear', S=None, L=10000, threshold_covariate_median_imputation=0.1, randomization_design='strata', verbose=False, covariate_adjustment=0, random_state=None, alternative='greater', alpha=0.05)
Imputation-Assisted Randomization Tests (iArt) for testing the null hypothesis that the treatment has no effect on the outcome.
Parameters
- Zarray_like
Z is the array of observed treatment indicators
- X, Yarray_like
X is 2D array of observed covariates, Y is 2D array of observed outcomes,
- Sarray_like, default: None
S is the array of observed strata indicators
- threshold_covariate_median_imputationfloat, default: 0.1
The threshhold for missing covariate to be imputed with median in advance for performance improvement
- Gstr or function, default: ‘iterative+linear’
A string for the eight available choice or a function that takes (Z, M, Y_k) as input and returns the imputed complete values
- Lint, default: 10000
The number of Monte Carlo simulations
- randomization_design{‘strata’,’cluster’}, default: ‘strata’
A string indicating the randomization design
- verbosebool, default: False
A boolean indicating whether to print training start and end
- covarite_adjustmentint, default: 0
if 0, covariate adjustment is not used if linear, linear covariate adjustment is used if xgboost, xgboost covariate adjustment is used if lightgbm, lightgbm covariate adjustment is used
- random_state{None, int, numpy.random.Generator,`numpy.random.RandomState`}, default: None
If seed is None (or np.random), the numpy.random.RandomState singleton is used. If seed is an int, a new
RandomStateinstance is used, seeded with seed. If seed is already aGeneratororRandomStateinstance then that instance is used.- alternative{‘greater’,’less’,’two-sided’}, default: ‘greater’
A string indicating the alternative hypothesis
- alphafloat, default: 0.05
Significance level
Returns
- p_valuesarray_like
1D array of p-values for lenY outcomes
- rejectarray_like
A boolean indicating whether the null hypothesis is rejected for each outcome