I want to continue on the work here, and come to a condensed form of the estimated variance of the average treatment effect.


\begin{align*} Y_{i} = \beta_{0} + \beta_{1}D_{i} + \varepsilon_{i} \end{align*}

where $D_{i}=1$ if the $i$th unit is in the treatment group and $0$ otherwise (the controlled group).

What is an estimation for $\text{Var}(\widehat{\beta_{1}})$, with respect to the sample variances of the treatment and controlled groups?

May someone please look over my work and logic?

I begin by recalling that

\begin{align*} \text{Var}(\widehat{\beta_{1}}) = \frac{\sigma_{\varepsilon}^{2}}{\sum_{i=1}^{N}(D_{i} - \overline{D})^{2}} \end{align*}

Let $p$ be the proportion of $N$ in the treatment group. Therefore $1-p$ is the proportion in the controlled group. It can be shown that

\begin{align*} \sum_{i=1}^{N}(D_{i} - \overline{D})^{2} = Np(1-p) \end{align*}

We are left to find the numerator, which can be estimated as

\begin{align*} \sigma_{\varepsilon}^{2} = \frac{\sum_{i=1}^{N}(Y_{i} - (\widehat{\beta_{0}}+\widehat{\beta_{1}}D_{i}))^{2}}{N-2} \end{align*}

After doing the algebra - I reach that

\begin{align*} \widehat{\text{Var}(\widehat{\beta_{1}})} = \frac{\widehat{\text{Var}(\overline{Y_{T}})}}{(N-2)(1-p)} + \frac{\widehat{\text{Var}(\overline{Y_{C}})}}{(N-2)(p)} \end{align*}

where $\overline{Y_{T}}$ is the sample average of the treatment group; notation generalised to $\overline{Y_{C}}$.

May someone please see if they reach the same conclusion? I used the following results

\begin{align} \text{Var}(X) &= \mathbb{E}(X^{2}) - \mathbb{E}(X)^{2} \\ \\ \overline{Y} &= p\overline{Y_{T}} + (1-p)\overline{Y_{C}} \\ \\ \overline{Y_{T}} &= \frac{1}{Np}\sum_{i=1}^{N}Y_{i|D_{i}=1} \\ \\ \overline{Y_{T}} &= \frac{1}{N(1-p)}\sum_{i=1}^{N}Y_{i|D_{i}=0} \end{align}

Thank you very much for your help



Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Browse other questions tagged or ask your own question.