How to prove this propensity score weighting leads to averaged treatment effect among the treated?

31 Views Asked by At

Let $Y(1)$ and $Y(0)$ be the potential outcomes under the treatment and the control. $T$ represents the treatment status. We let

$ATT=E(Y(1)-Y(0)|T=1)$

Next, we let $e(x)$ denote the propensity score. One type of propensity score weighting scheme is: if treated, they get a weight of 1; if not treated, they get a weight of $\frac{e(x)}{1-e(x)}$. We then let

$\Delta=\frac{TY}{1}-\frac{(1-T)Y}{1/\frac{e(x)}{1-e(x)}}$

Now how can we we show that

$E(\Delta)=E(\frac{TY}{1}-\frac{(1-T)Y}{1/\frac{e(x)}{1-e(x)}})=E(Y(1)-Y(0)|T=1)$. So that we know $E(\Delta)$ is unbiased for ATT and weights $(1,\frac{e(x)}{1-e(x)})$ is for ATT.?

I can show that

\begin{align} E(I(T=1)Y) &=E(I(T=1)Y(1)) \, \hspace{1cm} Y(1)=Y \, \text{because of consistency}\\ \end{align}

Now I have a hard time to understand how can we get \begin{align} E(I(T=1)Y(1))=E(Y(1)|T=1) \end{align}

If so, it follows that \begin{align} E(I(T=1)Y(0))=E(Y(0)|T=1) \end{align}

Any suggestion will be appreciated.