Estimation

class med_bench.estimation.mediation_coefficient_product.CoefficientProduct(regularize: bool, **kwargs)

Coefficient Product estimatation method class

estimate(*args, **kwargs)

fit(t, m, x, y)

Fits nuisance parameters to data

Parameters:

• array-like (y) – treatment value for each unit, binary

• (n_samples) (shape) – treatment value for each unit, binary

• array-like – mediator value for each unit, here m is necessary binary and uni- dimensional

• (n_samples) – mediator value for each unit, here m is necessary binary and uni- dimensional

• array-like – covariates (potential confounders) values

• (n_samples (shape) – covariates (potential confounders) values

• n_features_covariates) – covariates (potential confounders) values

• array-like – outcome value for each unit, continuous

• (n_samples) – outcome value for each unit, continuous

class med_bench.estimation.mediation_g_computation.GComputation(regressor, classifier, **kwargs)

GComputation estimation method class

estimate(*args, **kwargs)

fit(t, m, x, y)

Fits nuisance parameters to data

class med_bench.estimation.mediation_ipw.InversePropensityWeighting(classifier, clip: float, trim: float, **kwargs)

Inverse propensity weighting estimation method class

estimate(*args, **kwargs)

fit(t, m, x, y)

Fits nuisance parameters to data

class med_bench.estimation.mediation_mr.MultiplyRobust(regressor, classifier, ratio: str, normalized, **kwargs)

Iniitializes Multiply Robust estimatation method class

estimate(*args, **kwargs)

fit(t, m, x, y)

Fits nuisance parameters to data

class med_bench.estimation.mediation_dml.DoubleMachineLearning(regressor, classifier, normalized: bool, **kwargs)

Double Machine Learning estimation method class

estimate(t, m, x, y)

Estimates causal effect on data

fit(t, m, x, y)

Fits nuisance parameters to data

class med_bench.estimation.mediation_tmle.TMLE(regressor, classifier, ratio, **kwargs)

Implementation of targeted maximum likelihood estimation method class

estimate(*args, **kwargs)

fit(t, m, x, y)

Fits nuisance parameters to data

get_simulated_data

med_bench.get_simulated_data.simulate_data(n, rg, mis_spec_m=False, mis_spec_y=False, dim_x=1, dim_m=1, seed=None, type_m='binary', sigma_y=0.5, sigma_m=0.5, beta_t_factor=1, beta_m_factor=1)

Simulate data for mediation analysis

Parameters:

• n (int,) – Number of samples to generate.

• rg (RandomState instance,) – Controls the pseudo random number generator used to generate the data at fit time.

• mis_spec_m (obj:bool,) – Whether the mediator generation is misspecified or not defaults to False

• mis_spec_y (obj:bool,) – Whether the output model is misspecified or not defaults to False

• dim_x (int, optional,) – Number of covariates in the input. Defaults to 1

• dim_m (int, optional,) – Number of mediatiors to generate. Defaults to 1

• seed (int or None, optional,) – Controls the pseudo random number generator used to generate the coefficients of the model. Pass an int for reproducible output across multiple function calls. Defaults to None

• type_m (str,) – Whether the mediator is binary or continuous Defaults to ‘binary’,

• sigma_y (float,) – noise variance on outcome Defaults to 0.5,

:param sigma_m float: noise variance on mediator

Defaults to 0.5,

:paramnoise variance on mediator

Defaults to 0.5,

Parameters:

• beta_t_factor (float,) – scaling factor on treatment effect, Defaults to 1,

• beta_m_factor (float,) – scaling factor on mediator, Defaults to 1,

Returns:

• x (ndarray of shape (n, dim_x)) – the simulated covariates

• t (ndarray of shape (n, 1)) – the simulated treatment

• m (ndarray of shape (n, dim_m)) – the simulated mediators

• y (ndarray of shape (n, 1)) – the simulated outcome

• total (float,) – the total simulated effect

• theta_1 (float,) – the natural direct effect on the treated,

• theta_0 (float,) – the natural direct effect on the untreated,

• delta_1 (float,) – the natural indirect effect on the treated,

• delta_0 (float,) – the natural indirect effect on the untreated,

• p_t (ndarray of shape (n, 1),) – Propensity score

• th_p_t_mx (ndarray of shape (n, 1),) – overlap