No common causes/confounders present.

[sinya2]

Hi! I've tried to investigate causality between lung cancer and smoking using this dataset.
I've built such casual graph:

But I have an error Exception: No common causes/confounders present. Propensity score based methods are not applicable
after applying this code:

causal_graph =  '''digraph { 
                    CHRONIC_DISEASE;
                    AGE;
                    GENDER;
                    PEER_PRESSURE;
                    CHEST_PAIN;
                    SMOKING -> {YELLOW_FINGERS, COUGHING, LUNG_CANCER, CHRONIC_DISEASE};
                    LUNG_CANCER  -> {CHEST_PAIN,SHORTNESS_OF_BREATH, FATIGUE, WHEEZING};
                    CHRONIC_DISEASE -> {LUNG_CANCER, PEER_PRESSURE, FATIGUE, WHEEZING};
                    ALCOHOL_CONSUMING -> {CHRONIC_DISEASE, FATIGUE};
                    AGE -> {CHRONIC_DISEASE, FATIGUE, LUNG_CANCER};
                    GENDER  -> {CHRONIC_DISEASE, FATIGUE};
                    }'''

model= CausalModel(
    data = df,
    graph=causal_graph.replace("\n", " "),
    treatment="SMOKING",
    outcome='LUNG_CANCER')

model.view_model()
identified_estimand = model.identify_effect(proceed_when_unidentifiable=True)
print(identified_estimand)
estimate = model.estimate_effect(identified_estimand, 
                             method_name="backdoor.propensity_score_weighting",target_units="ate")

Also I am confused by having such estimands. Do not it suppose also to show what variables have to be controlled?

Estimand type: EstimandType.NONPARAMETRIC_ATE
### Estimand : 1
Estimand name: backdoor
Estimand expression:
    d                     
──────────(E[LUNG_CANCER])
d[SMOKING]                
Estimand assumption 1, Unconfoundedness: If U→{SMOKING} and U→LUNG_CANCER then P(LUNG_CANCER|SMOKING,,U) = P(LUNG_CANCER|SMOKING,)

Version information:

• DoWhy version [e.g. 0.7]

I have dowhy version 0.10.1.

[amit-sharma]

@sinya2 in your causal graph, the correct way to estimate causal effect is to simply compute $\mathbf{E}[Lung Cancer |Smoking]$. There is no confounding common cause. That is what the estimand is describing. Usually it shows which variables need to be controlled, but here you do not need to control any variable.

That is also the reason why propensity score methods are throwing an error. They depend on creating a treatment propensity model using the confounding common causes. For your problem, you can simply use linear regression estimator or S-learner from EconML to compute the causal effect.

[github-actions]

This issue is stale because it has been open for 14 days with no activity.

[github-actions]

This issue was closed because it has been inactive for 7 days since being marked as stale.