DoWhy for explicit causal analysis

How it works

DoWhy separates causal modeling, identification, estimation and refutation. The supplied graph or assumptions determine which effect can be identified from the available data.

What you need

Define the intervention, outcome and assumptions before choosing an estimator. Record which variables were measured and what unmeasured confounding could affect the analysis.

What to try first

Use simulated data with a known treatment effect. Draw the causal graph, estimate the effect and try a refutation test before analyzing observations.

Specify a variable that should be unrelated to treatment before testing refutation. Record which assumptions the check probes and which ones still need domain evidence.

Where to look for failures

A refutation check probes one assumption or failure mode. The result depends on whether the causal structure and measurement assumptions hold.

Related Franklin pages

An explanation of a prediction and an estimate of a causal effect ask different questions. Inspect the data and assumptions required by the selected analysis before interpreting its output.

• Fit explainable models and inspect feature effects with shared visualization tools. InterpretML for explainable predictive models
• Attribute a prediction to input features using explainer methods suited to the model. SHAP for explaining model predictions

For a primer on learning from data and evaluating predictions, see Franklin’s glossary. Machine learning

Project documentation

• Current project README
• README at the reviewed revision