implications of residual heteroskedasticity

#1
Hello,

I'm running a multiple regression with 4 predictors, and when I examined the Standardized residuals.

a Graph is below.

I'm just curious what this means for my regression. Can I not interpret the Beta values the same?

Do I need to fix this problem?

What I am interested in is showing the effects of each of my predictors on my DV.

what sorts of biases might this lead too and do I need to fix this?

Thank you.
 
#3
Re: implications of residual heteroskadasticity

The scatter plot shows a Cone shaped distribution of the residuals.

I know typically for a regression you want to see the data points scattered around, as they are standardized.

Isn't this a problem for interpretation? as variances may not be equal across the regression line?
 

Dason

Ambassador to the humans
#4
Re: implications of residual heteroskadasticity

I don't really see a cone shape. I think you're allowing the far left point to influence you too much.
 
#5
Re: implications of residual heteroskadasticity

If you look at the difference between the points around XY coordinates (0, 1.5) vs. those around coordinates (0, -1.5) You can see a more wide distribution at the latter coordinates yes?

Thanks for taking a look with me.
 

Dason

Ambassador to the humans
#7
Re: implications of residual heteroskadasticity

I can see what you're getting at but it just doesn't look that severe to me. The methods you're using are relatively robust to small departures from the assumptions. In any case you still get unbiased estimates even if there is a nonconstant variance.

If I was in your situation I wouldn't worry too much.
 
#8
Re: implications of residual heteroskadasticity

I can see why you might be slightly concerned, but at the saem time I agree with Dason that I wouldnt worry too much. There is no definite upward or downward trend and these tests can cope quite well even when assumptions arent satisfied brilliantly.
 
#9
Re: implications of residual heteroskadasticity

I think the randomness is being prevented due to the outlier to the far left. Look at that datapoint and see if it makes sense. I dont reco you delete data but if it is not in tune with the rest of the population --delete it and rerun the model.. your residuals looks close but that outlier seems to be influencing the model
 

Dason

Ambassador to the humans
#10
Re: implications of residual heteroskadasticity

I think the randomness is being prevented due to the outlier to the far left.
What does that even mean?

Look at that datapoint and see if it makes sense. I dont reco you delete data but if it is not in tune with the rest of the population --delete it and rerun the model.. your residuals looks close but that outlier seems to be influencing the model
The point does seem like it might have a high leverage but that doesn't mean you should just willy nilly throw it out of the model. It's not a bad idea to fit a model without it and see if that point significantly changes anything. If it doesn't then who cares, keep it in and don't worry about it. If it does change quite a bit then you might want to see if the point is justifiable. If something happened during the data collection process that invalidates that point then you have the option of removing it. But don't just throw points out because you don't like them. That's just bad science and bad statistics.