Demo for the DoWhy causal API

We show a simple example of adding a causal extension to any dataframe.

import dowhy.datasets
import dowhy.api
from dowhy.graph import build_graph_from_str

import numpy as np
import pandas as pd

from statsmodels.api import OLS

data = dowhy.datasets.linear_dataset(beta=5,
        num_common_causes=1,
        num_instruments = 0,
        num_samples=1000,
        treatment_is_binary=True)
df = data['df']
df['y'] = df['y'] + np.random.normal(size=len(df)) # Adding noise to data. Without noise, the variance in Y|X, Z is zero, and mcmc fails.
nx_graph = build_graph_from_str(data["dot_graph"])

treatment= data["treatment_name"][0]
outcome = data["outcome_name"][0]
common_cause = data["common_causes_names"][0]
df

	W0	v0	y
0	-1.297527	False	-2.444230
1	-0.750903	True	4.586710
2	0.423814	True	4.597054
3	-0.492698	True	3.996149
4	-1.196889	False	-0.901873
...	...	...	...
995	-1.008442	True	4.409140
996	-0.418388	False	1.331345
997	-0.773559	True	4.927800
998	-0.630556	False	-1.122099
999	0.105905	False	0.811834

1000 rows × 3 columns

# data['df'] is just a regular pandas.DataFrame
df.causal.do(x=treatment,
             variable_types={treatment: 'b', outcome: 'c', common_cause: 'c'},
             outcome=outcome,
             common_causes=[common_cause],
            ).groupby(treatment).mean().plot(y=outcome, kind='bar')

<Axes: xlabel='v0'>

df.causal.do(x={treatment: 1},
              variable_types={treatment:'b', outcome: 'c', common_cause: 'c'},
              outcome=outcome,
              method='weighting',
              common_causes=[common_cause]
              ).groupby(treatment).mean().plot(y=outcome, kind='bar')

<Axes: xlabel='v0'>

cdf_1 = df.causal.do(x={treatment: 1},
              variable_types={treatment: 'b', outcome: 'c', common_cause: 'c'},
              outcome=outcome,
              graph=nx_graph
              )

cdf_0 = df.causal.do(x={treatment: 0},
              variable_types={treatment: 'b', outcome: 'c', common_cause: 'c'},
              outcome=outcome,
              graph=nx_graph
              )

cdf_0

	W0	v0	y	propensity_score	weight
0	-1.504645	False	-1.690800	0.881423	1.134529
1	-1.271635	False	-0.719883	0.844585	1.184014
2	-1.536436	False	-1.601381	0.885817	1.128901
3	-0.521900	False	0.575721	0.664831	1.504141
4	-0.473614	False	-1.010226	0.650216	1.537949
...	...	...	...	...	...
995	-1.851640	False	-1.504164	0.922186	1.084380
996	-1.108646	False	0.707090	0.813611	1.229089
997	-1.870935	False	-2.870493	0.924027	1.082219
998	-1.140591	False	0.611924	0.820036	1.219459
999	-0.544243	False	-0.518798	0.671491	1.489224

1000 rows × 5 columns

cdf_1

	W0	v0	y	propensity_score	weight
0	-0.292275	True	4.049878	0.407038	2.456772
1	-1.833597	True	4.706013	0.079572	12.567210
2	-2.424888	True	4.041703	0.037578	26.611164
3	-1.285458	True	3.614333	0.152992	6.536308
4	0.053056	True	6.841924	0.521985	1.915765
...	...	...	...	...	...
995	-1.326290	True	1.728799	0.146013	6.848683
996	-1.630039	True	2.912754	0.102060	9.798120
997	-0.887897	True	4.014498	0.235610	4.244296
998	-0.546227	True	3.670496	0.327921	3.049514
999	-0.669044	True	5.264007	0.292618	3.417420

1000 rows × 5 columns

Comparing the estimate to Linear Regression

First, estimating the effect using the causal data frame, and the 95% confidence interval.

(cdf_1['y'] - cdf_0['y']).mean()

$\displaystyle 5.12975076619075$

1.96*(cdf_1['y'] - cdf_0['y']).std() / np.sqrt(len(df))

$\displaystyle 0.099627983505489$

Comparing to the estimate from OLS.

model = OLS(np.asarray(df[outcome]), np.asarray(df[[common_cause, treatment]], dtype=np.float64))
result = model.fit()
result.summary()

OLS Regression Results

Dep. Variable:	y	R-squared (uncentered):	0.895
Model:	OLS	Adj. R-squared (uncentered):	0.894
Method:	Least Squares	F-statistic:	4238.
Date:	Mon, 11 Aug 2025	Prob (F-statistic):	0.00
Time:	11:05:29	Log-Likelihood:	-1420.5
No. Observations:	1000	AIC:	2845.
Df Residuals:	998	BIC:	2855.
Df Model:	2
Covariance Type:	nonrobust

	coef	std err	t	P>|t|	[0.025	0.975]
x1	0.5590	0.026	21.719	0.000	0.508	0.609
x2	5.0079	0.056	90.015	0.000	4.899	5.117

Omnibus:	5.525	Durbin-Watson:	1.952
Prob(Omnibus):	0.063	Jarque-Bera (JB):	5.699
Skew:	0.131	Prob(JB):	0.0579
Kurtosis:	3.261	Cond. No.	2.16

Notes:
[1] R² is computed without centering (uncentered) since the model does not contain a constant.
[2] Standard Errors assume that the covariance matrix of the errors is correctly specified.