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

Degrees of freedom and standard error adjustments on all statistics for
models using DDFM=KR are not correct if your PROC MIXED model meets all
of the following criteria:

   The model contains RANDOM and REPEATED statements.
   No SOLUTION option is present on the RANDOM statement.
   SUBJECT= effects are present on the REPEATED statement and on all
      RANDOM statements.
   The REPEATED SUBJECT= effect is nested within the RANDOM SUBJECT=
      effect.
   The REPEATED subject effect levels are not balanced within the RANDOM
     subject effect levels.

The degrees of freedom for models using DDFM=SATTERTH are not correct if
your PROC MIXED model meets these criteria as well (there is no standard
error adjustment when Satterthwaite is used).

For example, suppose your RANDOM and REPEATED statements looked like

   random int / subject=person;
   repeated time / subject=eye(person) type=ar(1) ddfm=kr;

and a sample of your data looked like

                 Number of
   Person  Eye  Observations
     241    L        5
     241    R        4
     253    L        6
     253    R        3

then the data are unbalanced within the REPEATED levels, the remaining
criteria are also met, and the df along with the standard errors will be
incorrect.

To circumvent the problem, add the /S option to the RANDOM statement
(which will work only if the resulting G matrix is positive definite) or
rewrite the RANDOM statement so that it does not contain a SUBJECT=
option.  In the code above, we could write that RANDOM statement as

   random person;

to avoid the incorrect results.  You can also use PROC GLIMMIX instead
of PROC MIXED to avoid this problem.

Operating System and Release Information