0
$\begingroup$

I'm just new to regression. I would like to know if I'm having a linear model. Does unbiasedness breakdown if i fail to account for heteroskedasticity?

$\endgroup$
  • 1
    $\begingroup$ Please explain your problem some more. What does "I'm having a linear model" mean? Bias towards what? What is your use case etc. $\endgroup$ – Ujjwal Kumar Jan 12 '17 at 12:39
  • $\begingroup$ I think you mean failure to have homoskedasticity? It is heteroskedasticity that causes the difficulty. $\endgroup$ – Michael R. Chernick Jan 12 '17 at 12:42
  • 1
    $\begingroup$ Okay, I would like to know if unbiasedness breaks down if one fails to account for heteroskedasticity in a given model. like why $\endgroup$ – Pai Neto Jan 12 '17 at 12:48
  • 2
    $\begingroup$ The answer is simple: unbiasedness is inherently different from heteroscedasticity, and unbiasedness in the usual linear model does not in anyway depend on homoscedasticity. So no, unbiasedness does not breakdown it remains $\endgroup$ – Repmat Jan 12 '17 at 13:26
  • 1
    $\begingroup$ Thanks a lot @Repmat, I think I relate to the answer. It was really of much help coz I was getting hard time finding the effects of unbiased estimators. $\endgroup$ – Pai Neto Jan 12 '17 at 14:06
2
$\begingroup$

The answer to your question is yes. If you apply ordinary least squares to obtain the parameter estimates of the regression coefficients in linear regression, estimates will all still be unbiased if all assumptions hold other than homoscedasticity. Heteroscedasticity causes model misspecification and can hurt predictions if not accounted for. But in the face of heteroscedasticity the least squares estimates remain unbiased. You can find this on wikipedia under the title of heteroscedasticity. A search of this site should turn up information on this also. You should check for possible duplicates to your question.

Another point about OLS, the Gauss-Markov theorem states that the least squares estimates of the regression parameters are the best linear unbiased estimates even when the error terms are correlated but the model is still homoscedastic.

| cite | improve this answer | |
$\endgroup$

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.