causalis.scenarios.classic_rct.model

Difference-in-Means (RCT) scenario.

This module provides the DiffInMeans model, which implements standard inference methods for Randomized Controlled Trials (RCTs). It supports absolute and relative treatment effect estimation using t-tests, permutation tests, and z-tests for proportions.

Module Contents

Classes

DiffInMeans

Difference-in-means model for Randomized Controlled Trials (RCT).

API

class causalis.scenarios.classic_rct.model.DiffInMeans

Difference-in-means model for Randomized Controlled Trials (RCT).

The difference-in-means estimator is the simplest way to estimate the Average Treatment Effect (ATE) in a randomized experiment. Because treatment assignment is random, the difference in sample means between the treated and control groups is an unbiased estimator of the ATE.

Notes

The Average Treatment Effect (ATE) is defined as:

.. math::

\tau = E[Y(1) - Y(0)]

In an RCT, $D \perp (Y(0), Y(1))$, so:

.. math::

\tau = E[Y | D=1] - E[Y | D=0]

The estimator implemented here is the simple difference in sample means:

.. math::

\hat{\tau} = \frac{1}{n_1} \sum_{i: D_i=1} Y_i - \frac{1}{n_0} \sum_{i: D_i=0} Y_i

Standard errors and confidence intervals are computed using Welch’s t-test by default, which does not assume equal variances between groups:

.. math::

SE(\hat{\tau}) = \sqrt{\frac{s_1^2}{n_1} + \frac{s_0^2}{n_0}}

where $s_g^2$ is the sample variance in group $g$.

Examples

from causalis.scenarios.classic_rct.dgp import generate_classic_rct_26 from causalis.scenarios.classic_rct.model import DiffInMeans

Generate synthetic RCT data

data = generate_classic_rct_26(seed=42)

Fit the model

model = DiffInMeans().fit(data)

Estimate ATE using standard t-test

estimate = model.estimate(method=”ttest”) print(f”ATE: {estimate.value:.4f}”) ATE: 0.0339 print(f”P-value: {estimate.p_value:.4f}”) P-value: 0.0000

Attributes

data : CausalData or None The dataset used for fitting and estimation.

Initialization

fit(data: causalis.dgp.causaldata.CausalData) → causalis.scenarios.classic_rct.model.DiffInMeans

Fit the model by storing the CausalData object.

In the DiffInMeans scenario, “fitting” primarily involves validating the data structure and storing it for subsequent estimation.

Parameters

data : CausalData The CausalData object containing treatment and outcome variables. Treatment must be binary (0/1).

Returns

DiffInMeans The fitted model instance.

Raises

ValueError If the input is not a CausalData object.

estimate(method: Literal[causalis.scenarios.classic_rct.inference.ttest, causalis.scenarios.classic_rct.inference.welch_permutation_t_test, causalis.scenarios.classic_rct.inference.conversion_ztest] = 'ttest', alpha: float = 0.05, diagnostic_data: bool = True, **kwargs: Any) → causalis.data_contracts.causal_estimate.CausalEstimate

Compute the treatment effect using the specified method.

Parameters

method : {“ttest”, “welch_permutation_t_test”, “conversion_ztest”}, default “ttest” The inference method to use: - “ttest”: Standard Welch’s two-sample t-test (handles unequal variances). - “welch_permutation_t_test”: Permutation-based p-value using the Welch statistic. - “conversion_ztest”: Z-test for proportions, suitable for binary (0/1) outcomes. alpha : float, default 0.05 Significance level for calculating confidence intervals. diagnostic_data : bool, default True Whether to include diagnostic data in the result (e.g., covariate balance if confounders exist). **kwargs : Any Additional arguments passed to the underlying inference function: - For “welch_permutation_t_test”: B (iterations), alternative, seed. - For “conversion_ztest”: ci_method, se_for_test.

Returns

CausalEstimate A results object containing: - value: The estimated ATE (absolute difference). - p_value: The statistical significance. - ci_lower_absolute, ci_upper_absolute: Confidence interval bounds. - value_relative: The relative effect (% lift).

Raises

RuntimeError If the model has not been fitted yet. ValueError If an unsupported method is specified.

__repr__() → str