Normality Assumption

[shanehall.m]

Say there is 100 observations, but these observations are equally broken into 2 genders and 5 cities. When creating a qq plot or just checking normality by the S-W statistic, can the 100 observations be tested all at once or does the data need to be broken out by genders or even the five cities?

 

[Dason]

What are you trying to do? What are you going to do after testing normality? Are you creating a model? It would probably be better to just test the normality of the residuals of whatever model you create instead.

 

[shanehall.m]

Sorry for the lack of description... yes I would be creating a model. But to follow up on your question, you are saying run the model and get the residuals from the model as say part of the output, then test for normality on those residuals. But those residuals are calculated using all of the data as one.

 

[Dason]

But those residuals are calculated using all of the data as one.

Well yes you don't break the residuals up into groups - but the assumption is that the data for each group is normally distributed with a shared variance for most linear models so this is ok.