Non normality in residuals in SLR

[starpen]

Hey guys.

I've done a SLR and after testing the residuals for normality using the Shapiro-Wilk test and I've learned that the residuals are not normal. What does this mean? Is the test useless or can it be saved? Should I use another test to test my two variables?

 

[ledzep]

First of all, I wouldn't take the results of Shapiro-Wilk test as gospel. Instead, or together with S-W, I will use a Q-Q plot to visually inspect the normality of residuals.
Usually, ANOVA is pretty robust to non-normality.
If the residuals are not normal, this means the data is slightly heavier tailed or skewed (which depends on the shape of the Q-Q plot).
Usually a log transformation does the trick.

This might help you
http://people.duke.edu/~rnau/testing.htm

 

[starpen]

Okay, so my problem is that I've already transformed my data. I've never worked with anova before, but I'll give it a try. Is it possible to make some sort of graphical representation as with regression? I had such a pretty line in my regression when doing a scatterplot with fitted line.

 

[starpen]

Also, I get this error when doing anova: "factor variable may not contain noninteger values".

 

[Dason]

Also, I get this error when doing anova: "factor variable may not contain noninteger values".

What program are you using?

 

[starpen]

I'm using stata.

Can someone explain to me, if I can do SLR even if residuals are non normal and total N is 31.

 

[starpen]

Perhaps if I run a robust regression?

 

[starpen]

P.s. here is a picture of a q-plot with normal line (in the residuals).