Date:         Sat, 4 May 1996 08:18:24 -0700
Reply-To:     Karsten Self <kmself@IX.NETCOM.COM>
Sender:       "SAS(r) Discussion" <SAS-L@UGA.CC.UGA.EDU>
From:         Karsten Self <kmself@IX.NETCOM.COM>
Subject:      Re: PROC FREQ limitation -Reply
Comments: To: Walter Scott <wscott@MAIL.STATE.TN.US>

As Walter wrote, this is one of several limitations to proc freq, the other
being that character values are truncated to length 16 in proc freq
tabulations.  For character vars longer than 16, FREQ returns the number of
unique values based only on the first 16 characters of the value.
 
Both of these issues may be overcome through SQL.  The printed report is not
the same, but equivalent output datasets may be generated if you are simply
interested in frequency, percent, or cumulative values of either.  Code
below generates a dataset with the value and frequency count in both sql and
freq.  Output is identical for variables with less than 32,767 values, or
length <= 16.  Performance in my experience is roughly comperable.
 
(adapted from working code but untested)
 
   proc freq data= mydata;
       tables myvar /out= freqout;
       run;
 
   proc sql noprint;
       create table sqlout as
       select myvar as myvar, count( * ) as count
       from mydata
       group by myvar
       ;
       quit;
 
You can run a data step on the sqlout to generate the appropriate additional
PROC FREQ statistics if you desire.
 
At 03:16 PM 5/3/96 -0500, Walter Scott wrote:
>     "The maximum number of levels allowed for any one  variable is
>32,767."     SAS Procedures Guide, Version 6, Third Edition  p#337.
>
>Walter Scott
>
>Original Question>>>>>>>>>>>>
>>>> abdu elnagheeb <Abdelmoneim.H.Elnagheeb@aexp.com> 08:48 Fri,
>3 May 96 >>>
>I have noticed that if there are too many UNIQUE values for a variable on
>which you run PROC FREQ, you will get an error from SAS.  I did this on
>a BIG data set and the error reads:
>
>ERROR: There are more than 32767 levels for variable ACCT9.
>
> The program was : PROC FREQ data=dataname noprint;
>                   Tables acct9 /out =xcount;
>                   proc sort data=dataname out=newname;
>                   by acct9;
>                   data temp;
>                    merge newname xcount; by acct9;  run;
>
>I wanted to use the data set xcount to create a new variable (count) that
>shows the freq of acct9 in each observation.  I got other codes from
>good SAS-Lers to do this.  But I wonder if this can be fixed.
>Anyone familiar with this limition and how to overcome.
>Thanks in advance (TIA.)
>
>Abdu
>
>
----------------------------------------
Karsten Self / kmself@ix.netcom.com
 
What part of gestalt don't you understand?