Getting started with CausalEstimate

Install the Python package, prepare a pandas DataFrame, and compute a first causal effect estimate from your own propensity scores.

Installation

pip install CausalEstimate

For local development:

git clone https://github.com/kirilklein/CausalEstimate.git
cd CausalEstimate
pip install -e .

The input-data contract

Every estimator works on a plain pandas DataFrame. You tell the estimator which columns to use in its constructor; compute_effect(df) does the rest. CausalEstimate does not fit propensity or outcome models for you — you bring those predictions as columns:

Column	Meaning	Needed by
treatment	Binary treatment assignment (0/1)	all estimators
outcome	Observed outcome	all estimators
ps	Propensity score, your estimate of P(treatment = 1 given covariates)	IPW, AIPW, TMLE, Matching
predicted_outcome	Predicted outcome under the observed treatment	TMLE
predicted_outcome_treated	Predicted outcome if treated	AIPW, TMLE
predicted_outcome_control	Predicted outcome if untreated	AIPW, TMLE

Column names are arbitrary — pass them as treatment_col=..., ps_col=..., etc.

First estimate

import numpy as np
import pandas as pd
from CausalEstimate.estimators import IPW

np.random.seed(42)
n = 1000
ps = np.random.uniform(0, 1, n)          # true propensity for treatment
treatment = np.random.binomial(1, ps)    # actual treatment assignment
outcome = 2 + 0.5 * treatment + np.random.normal(0, 1, n)

df = pd.DataFrame({"ps": ps, "treatment": treatment, "outcome": outcome})

ipw_estimator = IPW(
    effect_type="ATE",
    treatment_col="treatment",
    outcome_col="outcome",
    ps_col="ps",
)

results = ipw_estimator.compute_effect(df)
print(results)

Output:

{'effect': 0.5518, 'effect_1': 2.5260, 'effect_0': 1.9742}

effect is the estimated treatment effect; effect_1 and effect_0 are the mean potential outcomes under treatment and control. For standard errors and confidence intervals, see Multiple Estimators & Bootstrap.