G-Computation
Causal inference
G-Computation
G-computation
- Fit a model for
y ~ x + z, where z is all covariates - Create a duplicate for each level of x
- Set the value of x to a single value for each cloned dataset:
x = 1for one,x = 0for the other - predict
- calculate estimate,
mean(x_1) - mean(x_0)
Advantages
- flexible
- precise (compared to propensity-score based methods)
- basis of other important models (e.g. TMLE)