Date:         Fri, 19 Apr 1996 03:54:48 -0400
Reply-To:     PHIL_G@DELTA.PRA-WW.COM
Sender:       "SAS(r) Discussion" <SAS-L@UGA.CC.UGA.EDU>
From:         Phil Gallagher <PHIL_G@DELTA.PRA-WW.COM>
Subject:      Graphs of Empirical Distributions

Kouros Owzar asked how to graph empirical distributions.
 
I do this all the time, but I do not have a sample of tested code with me.
I will show the approach, but all should recognize that I have not tested
it, and I am bound to have made the usual keying errors.
 
*  to graph the empirical distribution of GOODSTUF;
 
PROC FREQ DATA= yourdataset;
    TABLES GOODSTUF / OUT=STATS;
    *  Output dataset will have the vars GOODSTUF COUNT PERCENT;
    *  See the PROC FREQ documentation;
RUN;
 
DATA CUMULATE(LABEL='Create the cumulative percents');
    SET STATS;
    RETAIN CUM_PCT 0;
    IF (_N_ EQ 1) THEN CUM_PCT = 0;
        *  in principle this is unnecessary, but it hints at what to do
           in more complicated situations, such as when using a BY-stmnt;
    CUM_PCT = SUM(CUM_PCT,PERCENT);
    LABEL CUM_PCT = 'Empirical Distribution (%)';
        *  a place to be careful - sooner or later one almost always invokes
           FREQ with the missing option on the TABLES stmnt, and it is very
           embarassing to have used a simple "+" and wind up with all missing
           CUM_PCTs;
RUN;
 
PROC PLOT DATA=CUMULATE;  *  or GPLOT, of course;
    PLOT CUM_PCT * GOODSTUF / options;
    TITLE 'Empirical distribution of GOODSTUF';
RUN;
 
*  this design may be extended in a straightforward way to use with
   BY-stmnts, and the use of the OVERLAY option on the plot stmnt.  Only a
   little complicated.  If you absolutely cannot do it yourself, query me.;
 
                                    Philip Gallagher