0

Generally speaking, meaning the exponential family of distributions allowed under a GLM:

Is the distribution of the standardized deviance residuals approximately normally distributed?

At my place of business I see our homegrown GLM software plotting standardized deviance residuals for all types of GLMs in a QQ plot comparing to a normal distribution.

If they are asympotically normal, is there a standard reference I can refer to for this?

Thanks.

Stephen Lien
  • 233
  • It's a common thing to do (a normal QQ plot of one of the 'tend to be more normal' residuals like deviance residuals), but note that residuals are mostly affected by single observations (indeed, that's the whole point)... asymptotics is not usually something we try to invoke for $n=1$, and in this case, it doesn't apply. In some cases, residuals may be more-or-less normal, but in other cases, there's really nothing you can do. This answer recommends simulation to see what the distribution could look like in a particular instance. – Glen_b Jun 08 '13 at 00:45
  • (That question I link to in my previous comment might even be close enough to this one to count as a duplicate.) – Glen_b Jun 08 '13 at 00:46

0 Answers0