Math for Causal Inference
All simulators

The Roy Model of Self-Selection

Workers enrol when their gain beats the cost: decompose the naive gap into ATE, selection bias and heterogeneity bias.

Day 34: The Fundamental Problem & Selection Bias
-100105101520253035Y(0): earnings without traininggain Y(1) − Y(0)enrol above this line
ATE
2.02
ATT
3.19
theory 3.28
ATC
-1.80
naive
-0.06
selection bias
-3.25
heterogeneity bias
1.17
Check: ATE + SB + HB = -0.06 ≈ naive -0.06. Share enrolled 0.76.

Each dot is a worker: earnings without training (Y0) against the gain from training. Workers enrol when the gain exceeds the cost (dashed line). The naive comparison = ATE + selection bias (enrollees' Y0 differs) + heterogeneity bias (enrollees gain more).