SAS Institute Alert Note SN-006025

From: SAS Technical Support (LISTSERV@VM.SAS.COM)
Date: Fri Oct 12 2001 - 11:01:30 EDT

  • Next message: SAS Technical Support: "SAS Institute Technical Support Hot Fix 82IH02"

    SN-006025 ***Alert Note***

    Interval censoring syntax may lead to incorrect results from the
    PROBPLOT statement

      Product: SAS/STAT
    Component: LIFEREG procedure
     Priority: ALERT

      If there are any uncensored observations in the input dataset and you
      use interval censoring syntax (variable1 variable2) and a PROBPLOT
      statement, incorrect results will occur. The Turnbull algorithm for the
      computation of the empirical distribution function estimates may fail to
      iterate.

      With interval censoring syntax, uncensored observations are those for
      which variable1=variable2. Such observations represent exact failure
      times in the data.

      Results produced by the ITPRINTEM (interation history for the Turnbull
      algorithm) and the PPOUT (table of cumulative probabilities) options of
      the PROBPLOT statement will be in error. Plots produced by the PROBPLOT
      statement will also be incorrect. The MODEL statement does not use the
      Turnbull algorithm and is therefore unaffected.

      To circumvent the problem, make another run of PROC LIFEREG, recoding
      any uncensored observations (exact failure times) as interval censored
      observations on a very small interval. Recode the data so that the new
      intervals representing the uncensored observations do not overlap the
      endpoints of any other interval. For uncensored observations, those
      where variable1 and variable2 are equal, subtract a small positive value
      from variable1 and/or add a small positive value to variable2 keeping in
      mind the proviso about overlapping endpoints.

      Recoding the data in this manner will produce correct output from the
      ITPRINTEM and PPOUT options and correct plots. However, when recoding
      the data the likelihood will differ and results stemming from the
      likelihood will change, e.g. the maximum likelihood value, parameter
      estimates for the fitted distribution, parameter estimates for any
      covariates (independent variables) listed in any MODEL statements,
      percentile estimates from the PCTLIST option, etc.

    Keywords:
      procedure proc lifereg stat sas/stat saspts incorrout incorrstat
      turnbull algorithm does not iterate fails to iterate probplot statement
      incorrect results interval censoring censored syntax interval exact
      failure times (variable1 variable2) variable1 equals variable2
      variable1=variable2 (lower upper) lower equals upper lower=upper
      plots graphs incorrect computation of plotting positions
      interval censoring syntax may lead to incorrect results from the
      probplot statement
      006025 6025

    SAS Note Revised On: Thu, 11 Oct 2001

    System Version Reported Version Fixed

    AIX 64 bit 8.2 TS2M0
    AIX/6000 8.2 TS2M0
    VM/ESA (CMS) 8.2 TS2M0
    Compaq Tru64 UNIX 8.2 TS2M0
    HP-UX Operating Systems 8.2 TS2M0
    HP-UX 64 bit 8.2 TS2M0
    Intel ABI 8.2 TS2M0
    Linux 8.2 TS2M0
    MIPS ABI 8.2 TS2M0
    OS/390 (MVS) 8.2 TS2M0
    IBM OS/2 8.2 TS2M0
    Solaris 8.2 TS2M0
    Solaris 64 bit 8.2 TS2M0
    OpenVMS VAX 8.2 TS2M0
    OpenVMS Alpha 8.2 TS2M0
    Windows NT 8.2 TS2M0
    Windows 2000 Datacenter Server 8.2 TS2M0
    Windows 2000 Professional 8.2 TS2M0
    Windows 2000 Server and Advanced Server 8.2 TS2M0
    Windows 95 8.2 TS2M0
    Windows 98 8.2 TS2M0
    Windows ME 8.2 TS2M0

    No Fixes Available

    +-------------------------------------------------------------------------
    + NOTE: To unsubscribe you can reply to this mail with:
    + "SIGNOFF tsnews-l"
    + as the only text in the body of the message (without the double quotes).
    + If you have any questions please send them to SUPPORT@SAS.COM
    +-------------------------------------------------------------------------



    This archive was generated by hypermail 2b29 : Fri Oct 12 2001 - 12:09:42 EDT