1
$\begingroup$

Consider a simple linear model:

$Y = \beta_1 + \beta_2 MALE + u$

where $MALE$ is a dummy variable. $MALE = 0$ if female, $MALE = 1$ if male.

A model is fitted:

$\hat{Y} = \hat{\beta_1} + \hat{\beta_2} MALE$

$\bar{Y_F}$ and $\bar{Y_M}$ are the sample means of $Y$ for females and males respectively.

Prove that:

$\hat{\beta_1} = \bar{Y_F}$, $\hat{\beta_2} = \bar{Y_M} - \bar{Y_F}$

If the points are plotted in a scatter block, there will be two vertical clusters of points at $MALE = 0$ and $MALE = 1$, as shown:

Scatter plot and regression line

It seems intuitively obvious that the total residual sum of squares ($RSS$) is minimized by minimizing the $RSS$ separately for the female and male points. This is achieved by fitting a point at the mean of these points, $\bar{Y_F}$ and $\bar{Y_M}$. Fitting a line between these point yields the correct values for $\hat{\beta_1}$ and $\hat{\beta_2}$.

I wish to prove this algebraically.

$\endgroup$
  • $\begingroup$ This is not a place to get homeworks done. If you provide your solution (or at least a try) and explain your doubts, you can get help $\endgroup$ – PhDing Jun 6 '18 at 9:42
  • $\begingroup$ @Alessandro Thank you for your comment. I have gone away and attempted the question again. My answer is as below. $\endgroup$ – StevenRJClarke1985 Jun 6 '18 at 12:02
0
$\begingroup$

Find $\hat{\beta_2}$:

$$\hat{\beta_2}=\frac{\Sigma(Y_i-\bar{Y})(MALE - \overline{MALE})}{\Sigma(MALE - \overline{MALE})^2}$$

Assuming m males in a sample of n people, $$\overline{MALE}=\frac{m}{n}$$

We can say:

$$MALE - \overline{MALE}=\begin{cases}-\frac{m}{n} & \text{if }MALE=0\\\\\frac{n-m}{n} & \text{if }MALE=1\end{cases}$$

With some algebra, the numerator of $\hat{\beta_2}$ becomes:

$$\Sigma(Y_i-\bar{Y})(MALE - \overline{MALE})=\frac{m(n - m)}{n}(\bar{Y_M} - \bar{Y_F})$$

The numerator of $\hat{\beta_2}$ becomes:

$$\Sigma(MALE - \overline{MALE})^2=\frac{m(n - m)^2 + m^2(n - m)}{n^2}=\frac{m(n - m)}{n}$$

Dividing the two gives the correct answer:

$$\hat{\beta_2}=\bar{Y_M} - \bar{Y_F}$$

Find $\hat{\beta_1}$:

$$\hat{\beta_1}=\bar{Y}-\hat{\beta_2}\overline{MALE}=\frac{m\bar{Y_M}+(n - m)\bar{Y_F}}{n} - (\bar{Y_M} - \bar{Y_F})\frac{m}{n}=\bar{Y_F}$$

$\endgroup$

Your Answer

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

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