1
$\begingroup$

This question is based on How to derive a regression formula

Start with a regression equation: $s_{ij} = \alpha + \beta \bar{s_j} + \epsilon_{ij},$

The goal is to get to this. *Update. I figured this part out. ***The remaining question is listed at the very bottom.

$\hat{\beta} = \frac{\sum_j \sum_i s_{ij} (\bar{s_j} - \bar{s})}{\sum_j n_j (\bar{s_j} - \bar{s})^2} = \frac{\sum_j (\bar{s_j} - \bar{s})(n_j \bar{s_j})}{\sum_j n_j( \bar{s_j} - \bar{s})^2} = 1 $

First, get $\hat{\alpha}$:

\begin{equation} argmin_{\hat{\alpha}}~~ \hat{\epsilon_{ij}^2} = argmin_{\hat{\alpha}}~ \sum_j \sum_i (s_{ij} - \hat{\alpha} - \hat{\beta} \bar{s_j})^2 \end{equation}

\begin{equation} \rightarrow \sum_j \sum_i (s_{ij} - \hat{\alpha} - \hat{\beta} \bar{s_j}) = 0 \end{equation}

\begin{equation} \rightarrow \sum_j \sum_i \hat{\alpha} = \sum_j \sum_i (s_{ij} - \bar{s_j} \hat{\beta}) \end{equation}

\begin{equation} \rightarrow \sum_j n_j \hat{\alpha} = \sum_j n_j \bar{s_j} - \hat{\beta} \sum_j n_j \bar{s_j} \end{equation}

\begin{equation} \rightarrow \hat{\alpha} = (1 - \hat{\beta}) \frac{\sum_j n_j \bar{s_j} }{\sum_j n_j} = (1 - \hat{\beta}) \bar{s} \end{equation}

Next, get $\hat{\beta}$.

\begin{equation} argmin_{\hat{\beta}}~~ \hat{\epsilon_{ij}^2} = argmin_{\hat{\beta}}~ \sum_j \sum_i (s_{ij} - \hat{\alpha} - \hat{\beta} \bar{s_j})^2 \end{equation}

\begin{equation} = argmin_{\hat{\beta}}~ \sum_j \sum_i (s_{ij} - (1 - \hat{\beta}) \bar{s} - \hat{\beta} \bar{s_j})^2 \end{equation}

\begin{equation} = argmin_{\hat{\beta}}~ \sum_j \sum_i ((s_{ij} - \bar{s}) - \hat{\beta}( \bar{s_j} - \bar{s}))^2 \end{equation}

\begin{equation} \rightarrow \sum_j \sum_i ((s_{ij} - \bar{s}) - \hat{\beta}( \bar{s_j} - \bar{s}))( \bar{s_j} - \bar{s})= 0 \end{equation}

\begin{equation} \rightarrow \sum_j \sum_i (s_{ij} - \bar{s})( \bar{s_j} - \bar{s}) = \sum_j \sum_i \hat{\beta}( \bar{s_j} - \bar{s})( \bar{s_j} - \bar{s}) \end{equation}

\begin{equation} \rightarrow \hat{\beta} = \frac{\sum_j \sum_i (s_{ij} - \bar{s})( \bar{s_j} - \bar{s}) } {\sum_j \sum_i ( \bar{s_j} - \bar{s})( \bar{s_j} - \bar{s})} \end{equation}

\begin{equation} \rightarrow \hat{\beta} = \frac{\sum_j \sum_i (s_{ij} - \bar{s})( \bar{s_j} - \bar{s}) } {\sum_j \sum_i ( \bar{s_j} - \bar{s})^2} \end{equation}

\begin{equation} \rightarrow \hat{\beta} = \frac{\sum_j \sum_i (s_{ij} - \bar{s})( \bar{s_j} - \bar{s}) } {\sum_j n_j ( \bar{s_j} - \bar{s})^2} \end{equation}

\begin{equation} \rightarrow \hat{\beta} = \frac{\sum_j \sum_i (s_{ij} - \bar{s})( \bar{s_j} - \bar{s}) } {\sum_j n_j ( \bar{s_j} - \bar{s})^2} \end{equation}

\begin{equation} \rightarrow \hat{\beta} = \frac{\sum_j \sum_i s_{ij}(\bar{s_j} - \bar{s}) - \sum_j \sum_i \bar{s}(\bar{s_j} - \bar{s}) } {\sum_j n_j ( \bar{s_j} - \bar{s})^2} \end{equation}

\begin{equation} \rightarrow \hat{\beta} = \frac{\sum_j \sum_i s_{ij}(\bar{s_j} - \bar{s}) - \sum_j \sum_i \bar{s}(\bar{s_j} - \bar{s}) } {\sum_j n_j ( \bar{s_j} - \bar{s})^2} \end{equation}

\begin{equation} \rightarrow \hat{\beta} = \frac{\sum_j \sum_i s_{ij}(\bar{s_j} - \bar{s}) - \sum_j \sum_i \bar{s}\bar{s_j} - \bar{s}^2 } {\sum_j n_j ( \bar{s_j} - \bar{s})^2} \end{equation}

\begin{equation} \rightarrow \hat{\beta} = \frac{\sum_j \sum_i s_{ij}(\bar{s_j} - \bar{s}) - \bar{s}\sum_j n_j(\bar{s_j} - \bar{s}) } {\sum_j n_j ( \bar{s_j} - \bar{s})^2} \end{equation}

\begin{equation} \rightarrow \hat{\beta} = \frac{\sum_j \sum_i s_{ij}(\bar{s_j} - \bar{s}) } {\sum_j n_j ( \bar{s_j} - \bar{s})^2} \end{equation}

because $\sum_j n_j(\bar{s_j} - \bar{s}) = 0$

\begin{equation} \rightarrow \hat{\beta} = \frac{\sum_j (\bar{s_j} - \bar{s})(n_j \bar{s_j})}{\sum_j n_j( \bar{s_j} - \bar{s})^2} \end{equation}

Why does the last equality produce 1 in:

\begin{equation} \hat{\beta} = \frac{\sum_j (\bar{s_j} - \bar{s})(n_j \bar{s_j})}{\sum_j n_j( \bar{s_j} - \bar{s})^2} = 1 \end{equation}

$\endgroup$

1 Answer 1

1
$\begingroup$

Here are my steps (started from Line 8 of your steps finding $\hat{\beta}$). Hope there is no mistake. $$ \hat{\beta} = \frac{\sum_j\sum_i (s_{ij}-\bar{s})(\bar{s}_j - \bar{s})}{\sum_j n_j (\bar{s}_j - \bar{s})^2} \\ \implies \hat{\beta} = \frac{\sum_j \left[(\bar{s}_j - \bar{s})\sum_i (s_{ij}-\bar{s})\right]}{\sum_j n_j (\bar{s}_j - \bar{s})^2} $$

$$ \because \sum_i (s_{ij} - \bar{s}) = \sum_i s_{ij} - n_j \bar{s} = n_j\bar{s}_j - n_j\bar{s} = n_j (\bar{s}_j - \bar{s}) $$

$$ \therefore \hat{\beta} = \frac{\sum_j \left[(\bar{s}_j - \bar{s})\sum_i (s_{ij}-\bar{s})\right]}{\sum_j n_j (\bar{s}_j - \bar{s})^2} = \frac{\sum_j \left[(\bar{s}_j - \bar{s})n_j (\bar{s}_j - \bar{s})\right]}{\sum_j n_j (\bar{s}_j - \bar{s})^2} =1 $$


use $\implies$ (\implies in LaTeX) (which means 'implies' in mathematics) instead of $\to$

$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.