Causal effect estimation about dowhy HOT 2 CLOSED

That's a good question. In general, the treatment effect is ambiguous for a continuous variable. A convention is to estimate the difference in outcome between t=0 and t=1, but the exact values of t can change based on the requirement.

In the confounder example notebook , the treatment effect is estimated using a linear model, so the slope of the outcome wrt the treatment remains constant.

Thus, we consider E( outcome=1|t=1)-E(outcome|t=0)) which is equal to the coefficient of "t" in the linear model.

That said, this example is quite simple. In general, the relationship will be non-linear and the causal effect may vary based on value of the treatment. That is,
E( outcome=1|t=1)-E(outcome|t=0)) != E( outcome=1|t=k)-E(outcome|t=k-1)) for any k

Will be releasing an updated version soon that can handle these cases and allow effect estimation for any k.

from dowhy.

Hey @youssefbar I hope the answer was useful. Assuming that, I'm closing this issue, but if you have more questions, feel free to reply here.

from dowhy.