Effect of heteroskedasticity on Regression analysis

[nokia]

Hi Friends, Need you help in understanding these: Schweser says 10.k there are 4 effect of hetere… on regression analysis you should be aware of: - The standard errors are usually reliable estimates Reliable estimate of what??? what is it trying to explain here? - The coefficient estimates aren’t effected? Totally lost myself on this one too! - The standard errors are too small but the coefficient estimates themselves are not effected, the t-statistic will be too large and the null hypothesis is rejected too often. Opposite is true if SEs are too large This one further Increased by stress level Can any one please explain this ??? TIA

[125mph]

For hete… and serial correlation, the coefficients aren’t affected. Meaning the regression could still be good. What’s affected us standard errors may be small, which causes inflated tstat. If tstat is inflated, you may think it’s significant when it’s not. This is known as type I errors.

multicolinarity is opposite. It has type II errors.

[nokia]

Thanks 125mph slightly resolved my confusion

Standard error may b small or large than the ‘should be’ standard error, because of negative(small SE) or positive (large SE) correlation bw the standard error and the independent variable? Am I right when I say this statement?

Also you said:

@125mph

If tstat is inflated, you may think it’s significant when it’s not. This is known as type I errors.

If you reject the null when it should not be rejected , wouldn’t it b a type two error???

[Tactics]

heteroskedasticity makes the SE lower. Since T = B/SE, T would be inflated and make you think the coefficients are significant. btw, it has nothing to do with the sign or magnitude of the betas. heteroskedasticity simply lowers the standard errors.

[nokia]

Schweser says standard error may also be higher?

Also i want to know if there is a higher or lower standard error, how is the the slope of the line (coefficient ) not affected? Isn’t the heteroscedasticity mainly due to the fact that the slope line is not correct or not a best fit line???

Sorry for if it is something stupid… I feel like almost zero x zero in stat after this

[nokia]

@nokia

Thanks 125mph slightly resolved my confusion

Standard error may b small or large than the ‘should be’ standard error, because of negative(small SE) or positive (large SE) correlation bw the standard error and the independent variable? Am I right when I say this statement?

Also you said:

@125mph

If tstat is inflated, you may think it’s significant when it’s not. This is known as type I errors.

If you reject the null when it should not be rejected , wouldn’t it b a type two error???

Sorry my bad… That’s type1

[Tactics]

@nokia

Schweser says standard error may also be higher?

Also i want to know if there is a higher or lower standard error, how is the the slope of the line (coefficient ) not affected? Isn’t the heteroscedasticity mainly due to the fact that the slope line is not correct or not a best fit line???

Sorry for if it is something stupid… I feel like almost zero x zero in stat after this

SE will be higher for multicolinearity, not for heteroskedasticity. Heteroskedasticity is due to the error variance. It doesn’t affect the betas (coefficients). Omitted variable bias would affect the betas.