In this visualization, we'll use a sample dataset about a tutoring program (our treatment) and how it affects student exam scores (our outcome).
Our dataset has the following information about 40 students:
The Average Treatment Effect (ATE) measures the average difference in outcomes between all treated units and all control units in the entire population.
In our example, it's the average difference in math scores between students who received tutoring and students who didn't.
ATE = Average Outcome(Treated) - Average Outcome(Control)
The Conditional Average Treatment Effect (CATE) is the average treatment effect for a specific subgroup of the population.
In our example, we can see how tutoring affects students differently based on their grade level or previous performance.
Heterogeneous Treatment Effects (HTE) means that the treatment has different effects for different individuals or groups.
When we see different CATEs across subgroups, we have heterogeneous treatment effects.
The Local Average Treatment Effect (LATE) is the average treatment effect for "compliers" - units that receive treatment when encouraged but wouldn't otherwise.
In our example, imagine some students were offered tutoring (encouragement), but not all accepted it. LATE measures the effect for those who took the tutoring because they were encouraged.
SHAP (SHapley Additive exPlanations) values explain how each feature contributes to a prediction.
For our tutoring example, SHAP values can show how much each factor (tutoring, grade level, previous performance) contributed to a student's predicted math score.
Select a student to see how different factors affect their predicted math score: