>>>>>>>> SAS Notes -- Copyright (c) 1991 - 1999 by SAS Institute Inc. <<<<<<<<
V7-CATMOD-0347 *** ALERT NOTE *** PRODUCT: STAT
TITLE: Incorrect tests of between*within effects in repeated measures models
KEYS: INCORRSTAT REPEATED INTERACTIONS BETWEEN BY WITHIN ORDER ESTIMATES
If you specify more than one within-subjects factor in the REPEATED
statement (e.g. REPEATED TIME 2, TRIAL 2;), and if you specify
interaction(s) of between- and within-subjects factors on the MODEL
statement (e.g. group*_response_), then the tests of these interactions
will be incorrect in the Analysis of Variance table. Also, the
parameter estimates, while correct, are not correctly organized in the
Analysis of Weighted-Least-Squares Estimates table. PROC CATMOD
generates the correct design matrix columns, but if a between*within
interaction requires more than one column, those columns are not
consecutive in the matrix. Consequently, they are not in the order
stated in the Estimates table and the wrong contrast of parameters is
tested in the Analysis of Variance table. By examining the design
matrix, you can find the columns belonging to the interaction and then
produce a correct test of it using the CONTRAST statement. One symptom
of this problem is that tests of these interactions change if you change
the order of the within-subjects factors in the _RESPONSE_= option of
the REPEATED statement.
SYSTEM RELEASE LEVEL RELEASE LEVEL
AIX/R 7.00 0P1 .
CMS 7.00 0P1 .
DigitUnx 7.00 0P1 .
HP800 7.00 0P1 .
MVS 7.00 0P1 .
OS/2 7.00 0P1 .
VMS 7.00 0P1 .
This archive was generated by hypermail 2b29 : Mon Jan 10 2000 - 10:10:19 EST