Providing software solutions since 1976

• Log in
• Create profile
• Welcome
  [Log out]
• Edit profile

• Print  |
• Feedback  |
• Share

  • E-mail

  
  
  

  E-mail This Page

  Your Name:*
  Your E-mail:*
  Recipient E-mail:*
  Send a copy to me
  Personal Message:
  
  

  * indicates a required field

Compute a diagnostic similar to Cook's D for generalized linear models

NOTE: Beginning in SAS 9.2, this statistic is available both for independent-observation data and clustered data. See the COOKD= and CLUSTERCOOKD= options in the OUTPUT statement in PROC GENMOD.

PURPOSE:

Illustrate computing a diagnostic similar to Cook's D statistic for generalized linear models. This diagnostic gives a measure of the influence of each observation on the model fit.

REQUIREMENTS:

Base SAS and SAS/STAT Software, Version 7 or later.

DETAILS:

Ideally, the influence of an observation on the model fit would be determined by fitting the model with and without the observation and computing twice the difference in the resulting log likelihoods, divided by the number of parameters in the model. However, this would require the model to be fit as many times as there are observations. The statistic computed here is an approximation that is considerably less expensive to compute.

---

These sample files and code examples are provided by SAS Institute Inc. "as is" without warranty of any kind, either express or implied, including but not limited to the implied warranties of merchantability and fitness for a particular purpose. Recipients acknowledge and agree that SAS Institute shall not be liable for any damages whatsoever arising out of their use of this material. In addition, SAS Institute will provide no support for the materials contained herein.

data berkeley;
  input dept sex yes no total @@;
  prop=yes/total;
  cards;
1 1 512 331 843 1 2  89  19 108
2 1 353 207 560 2 2  17   8  25
3 1 120 205 325 3 2 202 391 593
4 1 138 279 417 4 2 131 244 375
5 1  53 138 191 5 2  94 299 393
6 1  22 351 373 6 2  24 317 341
;

proc genmod data=berkeley;
  class dept sex;
  model yes/total=dept sex / dist=binomial link=logit;
  output out=obsout stdxbeta=std hesswgt=hesswgt
                    stdreschi=stdreschi p=pred;
  run;
  
data cookd; 
  set obsout;
  obs=_N_;
  /* p is number of parameters excluding the intercept:
     5 for DEPT, 1 for SEX */
  p=6;
  /* h is the leverage */
  h=Std*Hesswgt*Std;
  cookd=h*Stdreschi**2/((p+1)*(1-h));
  run;

proc plot; 
  plot cookd*obs;
  run;

proc print noobs; 
  var dept sex prop pred cookd;
  run;


/******************************************************************** 
   Note that an approximate Cook's D can also be obtained using PROC 
   LOGISTIC for binary response models.  The C= option on the OUTPUT 
   statement outputs a statistic that is the Cook's D approximation 
   times the number of parameters.
********************************************************************/
proc logistic data=berkeley;
  class dept sex / param=glm;
  model yes/total=dept sex;
  output out=lout p=pred c=c;
  run;

data lout; 
  set lout;
  /* Divide by the total number of parameters in the model */
  cookd=c/7;
  run;

proc print noobs; 
  var dept sex prop pred cookd;
  run;

---

These sample files and code examples are provided by SAS Institute Inc. "as is" without warranty of any kind, either express or implied, including but not limited to the implied warranties of merchantability and fitness for a particular purpose. Recipients acknowledge and agree that SAS Institute shall not be liable for any damages whatsoever arising out of their use of this material. In addition, SAS Institute will provide no support for the materials contained herein.

RESULTS:

Following is the output from the PRINT procedure below giving the values of the diagnostic under the variable COOKD:

               DEPT    SEX      PROP       PRED      COOKD

                 1      1     0.60735    0.62914    24.6544
                 1      2     0.82407    0.65402     0.8009
                 2      1     0.63036    0.63142     0.2322
                 2      2     0.68000    0.65622     0.0008
                 3      1     0.36923    0.33493     0.6589
                 3      2     0.34064    0.35944     1.7236
                 4      1     0.33094    0.32815     0.0117
                 4      2     0.34933    0.35243     0.0105
                 5      1     0.27749    0.23813     0.2929
                 5      2     0.23919    0.25831     1.1085
                 6      1     0.05898    0.06131     0.0123
                 6      2     0.07038    0.06784     0.0124

Operating System and Release Information