SAS Note 001436

From: tsdlist@UNX.SAS.COM
Date: Mon Dec 20 1999 - 10:12:16 EST


>>>>>>>> SAS Notes -- Copyright (c) 1991 - 1999 by SAS Institute Inc. <<<<<<<<

SAS Note: 001436 ***ALERT NOTE*** Product: STAT

Title: Observations with missing values incorrectly excluded from analysis

Keywords:
INCORRSTAT INCORROUT MISSING VALUES CLASS LISTWISE DELETION EXCLUDED
CATEGORICAL CHARACTER PATTERN

When the MISSING option is not specified on the PROC SURVEYMEANS
statement and more than one class or categorical variable is listed on
the VAR statement, observations with a missing value for any class
variable will be incorrectly excluded from estimates for all of the
class variables. As a result, if the pattern of missing values is
different across the class variables then all of their output statistics
will be incorrect. In addition, if one or more continuous variables are
also specified on the VAR statement, then the degrees of freedom,
p-values and confidence intervals will be incorrect for all of the class
variables even if there is a consistent pattern of missing values in the
class variables.

In all of these situations, corresponding observations are not excluded
from continuous variables on the VAR statement, so the results for these
variables are correct.

To circumvent this problem, run any categorical variables with missing
values by themselves in separate executions of the procedure.

Reported Releases:
+------------REPORTED------------+ +------FIXED-----+
System Version Level Version Level

AIX/R 7 P1
AIX/R 8 M0
CMS 7 P1
CMS 8 M0
DigitUnx 7 P1
DigitUnx 8 M0
HP800 7 P1
HP800 8 M0
IABI 8 M0
MIPS ABI 8 M0
MVS 7 P1
MVS 8 M0
OS/2 7 P1
OS/2 8 M0
Solaris 7 P1
Solaris 8 M0
VMS 7 P1
VMS 8 M0
VMS/AXP 7 P1
VMS/AXP 8 M0
WIN/NT 7 P1
WIN/NT 8 M0
Win95 7 P1
Win95 8 M0
Win98 7 P1
Win98 8 M0



This archive was generated by hypermail 2b29 : Mon Jan 10 2000 - 10:10:21 EST