RE: st: Residuals in Logistic Regression

 From "Nick Cox" <[email protected]> To <[email protected]> Subject RE: st: Residuals in Logistic Regression Date Fri, 9 Apr 2004 18:55:09 +0100

```To give this pot another stir, and to use
mutually accessible data, -logit- and -glm-
with logit link and binomial family give quite
different deviance residuals.

The pattern of -glm-'s makes sense,
but that of -logit-'s is more puzzling.

Something's wrong here; is it my understanding?

sysuse auto, clear
logit foreign mpg
predict logitdev, dev
predict glmdev, dev
scatter *dev
scatter logitdev mpg, mlabel(foreign)
scatter glmdev mpg, mlabel(foreign)

Nick
[email protected]

Richard Williams
>
> >I obviously don't understand how the residual statistics
> work in logistic
> >regression.  I have run a logistic regression of happymar
> (coded 0, 1) on
> >church and female (also both 0, 1) and educ (years of
> education).  I then
> >use predict to get the deviance residuals (I get similar
> results if I use
> >the rstandard or residuals options on predict).  I get the following:
> >
> >. extremes  dev p happymar church female educ, nolabel high
> >
> >   +----------------------------------------------------------------+
> >   | obs:        dev          p   happymar   church   female   educ |
> >   |----------------------------------------------------------------|
> >   |  43.   1.170689   .5039612          1        1        0     10 |
> >   |   2.   1.394511   .3782007          1        0        1     10 |
> >   |   6.     2.4859   .0858805          0        0        0     11 |
> >   |  13.     2.4859   .0858805          1        0        0     11 |
> >   |  36.     2.4859   .0858805          1        0        0     11 |
> >   +----------------------------------------------------------------+
>
> To follow up on my own message:  I ran the same logistic
> regression in
> SPSS, and its residuals statistics behave as I would expect,
> i.e. case 6 is
> not an outlier but cases 13 and 36 are.  As far as I can
> tell, with Stata's
> logistic regression, the residuals produced by the
> -deviance-, -rstandard-,
> and -deviance- parameters on -predict- will all be the same
> for cases that
> have the same values on the covariates, regardless of whether
> they have the
> same value on the outcome measure.  I imagine there is a
> logic behind this,
> but I don't know what it is, and it is not the same logic
> used by SPSS in
> computing residuals.

*
*   For searches and help try:
*   http://www.stata.com/support/faqs/res/findit.html
*   http://www.stata.com/support/statalist/faq
*   http://www.ats.ucla.edu/stat/stata/
```