topic Re: Proc Boxplot for percintiles only in Graphics Programming

Proc Boxplot for percintiles only

DmytroYermak — Mon, 18 May 2020 08:51:38 GMT

Hi all,

Can you please help with proc boxplot. I have the dataset where there are just calculated percentiles 10, 25, 50, 75, 90:

data WORK.MOTOR;
infile datalines dsd truncover;
input LEVEL:32. MOTOR_TYPE:$14. SUB_TYPE_N:32. P10:32. P25:32. P50:32. P75:32. P90:32.;
label LEVEL="Level" MOTOR_TYPE="Motor Type" SUB_TYPE_N="Motor Type(N)" P10="10th Percentile" P25="25th Percentile" P50="50th Percentile" P75="75th Percentile" P90="90th Percentile";
datalines4;
1,M14,1,24.7,31,33.15,36.3,36.34
2,M14,1,14.3,22.5,28,29.15,32.72
3,M14,1,14.3,22.5,28,29.5,32.72
4,M14,1,15,18,18,26.15,32.78
5,M14,1,25.2,28.5,31,34.3,35.2
6,M14,1,16.2,18,18.3,20,26
1,M15,2,53.8,78,96.5,157,188.14
2,M15,2,58.2,84.8,120.2,159,227.9
3,M15,2,52.6,72.33,91.3,114.8,166.4
4,M15,2,52.6,72.32,91.3,114.8,166.4
5,M15,2,42,47,64.5,92.3,150.5
6,M15,2,52.6,72.33,91.3,114.8,166.4
;;;;

I need to generate boxplots just for percentiles (for each motor_type)

Is it possible with proc boxplot and what is the syntax? Thank you!

Re: Proc Boxplot for percintiles only

PaigeMiller — Mon, 18 May 2020 10:37:09 GMT

As far as I know, PROC BOXPLOT (and also PROC SGPLOT) doesn't work on pre-computed percentiles. You have to provide the raw data to PROC BOXPLOT (and PROC SGPLOT).

Re: Proc Boxplot for percintiles only

FreelanceReinh — Mon, 18 May 2020 12:53:07 GMT

Hi @DmytroYermak,

PROC BOXPLOT has an option HISTORY= where you can provide an input dataset containing group summary statistics. However, it must contain a mean value per group (and other statistics not contained in your sample data). Of course, you could set "mean=median," but the mean would still be plotted by default and you would need to modify ODS graphics settings to hide it because for a horizontal box plot you must use ODS graphics. To avoid these complications, I would suggest that you create an ordinary input dataset from your statistics dataset and use PROC SGPLOT, as shown below.

data inds(drop=p:);
set motor;
array p p:;
do over p;
  x=p; output;
end;
run;

title 'Box Plot From Summary Statistics';
%let attrs=thickness=2px color=CXE2B977;
proc sgplot data=inds;
by motor_type;
xaxis min=0 label='Parameter';
hbox x / category=level whiskerpct=0 nomean nofill
         lineattrs   =(&attrs)
         medianattrs =(&attrs)
         whiskerattrs=(&attrs);
run;

Edit: This assumes that you want the whiskers to extend to the 10% and 90% quantile (you don't have minimum and maximum in your data anyway). PROC SGPLOT achieves this (using the whiskerpct=0 or extreme option of the HBOX statement) by treating P10 and P90 as minimum and maximum, respectively. P25, P50 and P75 are the three quartiles of the five data points per category (level) and BY group in dataset INDS.

Re: Proc Boxplot for percintiles only

DanH_sas — Mon, 18 May 2020 14:15:43 GMT

If your data is pre-computed, you will need to use a BOXPLOTPARM statement in GTL. The data structure should look something like the following:

Statistics Computed for Two Days
Day	PowerOutputs	Statistic
04JUL	3180.00	MIN
04JUL	3340.00	Q1
04JUL	3487.40	MEAN
04JUL	3490.00	MEDIAN
04JUL	3610.00	Q3
04JUL	4050.00	MAX
04JUL	20.00	N
05JUL	3179.00	MIN
05JUL	3333.50	Q1
05JUL	3471.65	MEAN
05JUL	3419.50	MEDIAN
05JUL	3605.00	Q3
05JUL	3849.00	MAX
05JUL	20.00	N

and a simple template would look like the following:

proc template;
  define statgraph boxplotparm1;
    begingraph;
      entrytitle "City Mileage for Vehicle Types";
      layout overlay;
       boxplotparm y=value x=x stat=stat /
         datalabel=datalabel spread=true ;     
      endlayout;
    endgraph;
  end;
run;

/* Generate the plot. */
proc sgrender data=boxdata template=boxplotparm1;
run;

See https://go.documentation.sas.com/?docsetId=grstatgraph&docsetTarget=p1db7ll5bzyo4nn1pv31g04etput.htm&docsetVersion=9.4&locale=en for more details.

Hope this helps!

Dan

Re: Proc Boxplot for percintiles only

DmytroYermak — Mon, 18 May 2020 14:44:32 GMT

Thank you, Dan!

Re: Proc Boxplot for percintiles only

DmytroYermak — Mon, 18 May 2020 14:49:00 GMT

Thank you, Reinhard! Can you please help with option that put both Motor_type on one page? Like this:

Re: Proc Boxplot for percintiles only

FreelanceReinh — Mon, 18 May 2020 16:04:25 GMT

Then use the GROUP= option instead of a BY statement:

title 'Box Plot From Summary Statistics';
%let attrs=thickness=2px;
proc sgplot data=inds;
styleattrs datacontrastcolors=(gray CXE2B977) datalinepatterns=(1 1);
xaxis min=0 label='Parameter';
hbox x / category=level whiskerpct=0 nomean nofill group=motor_type
         lineattrs   =(&attrs)
         medianattrs =(&attrs)
         whiskerattrs=(&attrs);
run;

Re: Proc Boxplot for percintiles only

DmytroYermak — Mon, 18 May 2020 17:19:49 GMT

It is undoubtedly a solution to my case. I just wanted to add that I was able to adapt the code https://blogs.sas.com/content/graphicallyspeaking/2013/03/24/custom-box-plots/#prettyPhoto while trying to sort it out:


proc sgplot data=motor nocycleattrs;
  highlow y=LEVEL high=p90 low=p10 / group=MOTOR_TYPE groupdisplay=cluster
      clusterwidth=0.7;
  highlow y=LEVEL high=p75 low=p25 / group=MOTOR_TYPE type=bar
      groupdisplay=cluster grouporder=ascending clusterwidth=0.7
      barwidth=0.7 name='a';
  highlow y=LEVEL high=p50 low=p50 / group=MOTOR_TYPE type=bar
      groupdisplay=cluster grouporder=ascending clusterwidth=0.7
      barwidth=0.7;
  keylegend 'a';
  xaxis label= "Parameter";
  yaxis label= "Level" reverse;
  run;

There is no color and width solution here.