st: Residual deviance analysis after GLM-Negative Binomial

["Abdul Q Memon" <ucesaqm@ucl.ac.uk>]

Hi All

Your response to this will be appreciated;

I am using Generalised Linear model with negative binomial for count data.
For residual analysis I am using following four plots:

?        Deviance residuals vs estimated values
?        Normal qq plot
?        Scale location plot
?        Cooks distance plot

Could i please ask when we use negative binomial it is assumed the error
terms will come from negative binomial distribution. But we are comparing
the residual deviance in qq plot to follow assumptions of normailit???
Normal distribution. Is It due to

1.        We are comparing residual deviance instead of deviance which is
standardised some how
2.        May be due to central limit theorem which suggests that with high
number of observations the distribution will be close to normal.
3.        In scale location plot we are observing the data to be around 1
(ABS(SQUAREROOT(deviance residuals) on y axis. Is it because the we want
the deviance residual to be between +2 and -2 and its root will be 1.


With Regards

Memon








*
*   For searches and help try:
*   http://www.stata.com/help.cgi?search
*   http://www.stata.com/support/statalist/faq
*   http://www.ats.ucla.edu/stat/stata/