Consider the standard simple regression model $y= \beta_o + \beta_1 x +u$ under the Gauss-Markov Assumptions SLR.1 through SLR.5.
Let $\tilde{\beta_1}$ be the estimator for $\beta_1$ obtained by assuming that the intercept is 0. Find $E[\tilde{\beta_1}]$ in terms of the $x_i$, $\beta_0$, and $\beta_1$. Verify that $\tilde{\beta_1}$ is an unbiased estimator of $\beta_1$ obtained by assuming intercept is zero. Are there any other cases when $\tilde{\beta_1}$ is unbiased?


We need to prove that $E[\tilde{\beta_1}] = E[\beta_1]$

Using least squares, we find that $\tilde{\beta_1} = \dfrac{\sum{x_iy_i}}{\sum{(x_i)^2}}$

Then, $ \tilde{\beta_1} = \dfrac{\sum{x_i(\beta_0 +\beta_1x_i +u)}}{\sum{(x_i)^2}}$

$\implies \tilde{\beta_1} = \beta_0\dfrac{\sum{x_i}}{\sum{(x_i)^2}} +\beta_1 +\dfrac{\sum{x_iu_i}}{\sum{(x_i)^2}}$

Taking Expectayion on both sides:

$\implies E[\tilde{\beta_1}] = \beta_0E[\dfrac{\sum{x_i}}{\sum{(x_i)^2}}]+ \beta_1 +\dfrac{\sum{E(x_iu_i)}}{E[\sum{(x_i)^2}]}$ (since summation and expectation operators are interchangeable)

Then, we have that $E[x_iu_i]=0$ by assumption (results from the assumption that $E[u|x]=0$

$\implies E[\tilde{\beta_1}] = \beta_0E[\dfrac{\sum{x_i}}{\sum{(x_i)^2}}]+ \beta_1 +0$

Now, the only problem we have is with the $\beta_0$ term.

If we have that $\beta_0 =0$ or $\sum{x_i}=0$, then $\tilde{\beta_1}$ is an unbiased estimator of $\beta_1$/

Can anyone please verify this proof? Also, why don't we write $y= \beta_1x +u$ instead of $y= \beta_0 +\beta_1x +u$ if we're assuming that $\beta_0 =0$ anyway?

Please let me know if my reasoning is valid and if there are any errors.

Thank you.


Here's where I got the slope estimate from:

enter image description here


You are to show $E(\tilde\beta_1)=\beta_1$ but your formula for $\tilde\beta_1$ is not correct. It should be: \begin{align*} \tilde\beta_1&=\frac{\sum_i(x_i-\bar{x})y_i}{\sum_i(x_i-\bar{x})^2}\quad\text{where}\quad\bar{x}=\frac{1}{n}\sum_ix_i. \end{align*} Now, first, observe $\sum_i(x_i-\bar{x})^2=\sum_i(x_i-\bar{x})x_i$. Second, using $y_i=\beta_0+\beta_1x_i+u_i$, we have: $$ \sum_i(x_i-\bar{x})y_i=\sum_i(x_i-\bar{x})(\beta_0+\beta_1x_i+u_i)=\beta_1\sum_i(x_i-\bar{x})x_i+\sum_i(x_i-\bar{x})u_i. $$ Therefore, $$ \tilde\beta_1=\beta_1+\frac{\sum_i(x_i-\bar{x})u_i}{\sum_i(x_i-\bar{x})^2}=\beta_1+\sum_i\lambda_iu_i,\quad\text{with}\quad\lambda_i=\frac{x_i-\bar{x}}{\sum_j(x_j-\bar{x})^2}. $$ Under the Gauss-Markov's assumptions, the $\lambda_i$ are not random. And so: $$ E(\tilde\beta_1)=\beta_1+E\left(\sum_i\lambda_iu_i\right)=\beta_1+\sum_i\lambda_iE(u_i)=\beta_1. $$

  • $\begingroup$ @A.Asad: Some algebra simplification steps are skipped, intentionally. Fill in the blanks yourself if something is unclear. $\endgroup$ – yurnero Oct 6 '18 at 14:52
  • $\begingroup$ Can you please check my edit? I've added a picture from the book I'm using and it says that the slope estimate is the one which I have used. $\endgroup$ – A.Asad Oct 6 '18 at 17:34
  • $\begingroup$ @A.Asad They are imposing zero intercept (read their first sentence). The formula in my answer works whether the intercept is zero or not. $\endgroup$ – yurnero Oct 6 '18 at 18:01

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.