Re: st: FW: DFBetas after GEE model

[Steve Samuels <sjsamuels@gmail.com>]

Tim,

Here's a direct approach. It utilizes psuedo-values from Stata's
-jackknife- prefix command to calculate coefficients when a set of
observations is is omitted. The obvious disadvantage is the time needed
for large data sets. For -xtgee- you'll  have to decide whether to
omit one individual or one panel at a time (cluster() option in
-jacknife-).

If:
     b = coefficient for entire sample
    b_i = coefficient omitting obs i
    se_i = standard error omitting i

The jackknife pseudovalue is: bpv_i = n*b - (n-1)*b_i

Thus: b_i = (bpv_i -n*b)/(n-1)

The standard error  se_i that omits i is
computed in the same way. Then:

    DFBETAS_i = (b - b_i)/se_i


**********CODE BEGINS***********************
sysuse auto, clear
local xvars trunk turn  //predictors
gen y = foreign       // short name

jackknife _b _se, keep: logit y  `xvars'
sum y_*  //pseudovalues

logit y `xvars'  //entire sample

foreach z of varlist `xvars'{
gen double b_`z' = (e(N)*_b[`z'] - y_b_`z')/(e(N) -1)
gen double se_`z' = (e(N)*_se[`z'] - y_se_`z')/(e(N)-1)
gen double dfb_`z' = (_b[`z']-b_`z')/se_`z'
label var dfb_`z' "DFBETAS `z'"
}
**************CODE ENDS**************

Steve

On Aug 13, 2013, at 6:02 AM, <T.J.Laing@lse.ac.uk> <T.J.Laing@lse.ac.uk> wrote:

Hi

Thanks - will take a look. 

Is there anyone who has done any of these in Stata, the commands that work for other types of regression models (inc. GLM) are not available for the xtgee command.

Thanks

Tim Laing
________________________________________
From: owner-statalist@hsphsun2.harvard.edu [owner-statalist@hsphsun2.harvard.edu] on behalf of Najib Mozahem [najib.mozahem@gmail.com]
Sent: 12 August 2013 17:09
To: statalist@hsphsun2.harvard.edu
Subject: Re: st: FW: DFBetas after GEE model

"Generalized Estimating Equations" by Hardin and Hilbe says that Cooks, distance, and DFBETA and DFFIT residuals can be used with GEE models as well. They provide a brief summary about these techniques on pages 188-194.

Najib
On Aug 12, 2013, at 12:51 PM, <T.J.Laing@lse.ac.uk> wrote:

> Hi,
> 
> I've been running some GEE models and would like to do some checking for outliers and was wondering if anyone had tried to calculate Cook's distance or DFBetas after such models as is suggested by the literature.
> 
> Thanks
> 
> Tim Laing
> t.j.laing@lse.ac.uk
> 
> Please access the attached hyperlink for an important electronic communications disclaimer: http://lse.ac.uk/emailDisclaimer
> 
> *
> *   For searches and help try:
> *   http://www.stata.com/help.cgi?search
> *   http://www.stata.com/support/faqs/resources/statalist-faq/
> *   http://www.ats.ucla.edu/stat/stata/


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

Please access the attached hyperlink for an important electronic communications disclaimer: http://lse.ac.uk/emailDisclaimer

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


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