Normal probability plot with 95% confidence interval bands

[JuLePe]

I would like to make a probability plot with SAS just as the above (created with MINITAB). It's important to have 95% confidence interval bands and the same statistics in the inset.

Just in case, this is the data I used:

Obs	Bimodal_Var
1	11.18227
2	8.349733
3	14.40938
4	24.66227
5	9.279151
6	7.238792
7	15.46339
8	22.47347
9	15.132
10	21.10275
11	8.501284
12	15.77826
13	16.20075
14	17.67517
15	7.934397
16	18.39657
17	11.40699
18	8.399611
19	8.490487
20	22.70707
21	11.09486
22	11.00888
23	22.85953
24	6.783415
25	11.31967
26	25.19409
27	12.12261
28	10.3787
29	22.13491
30	20.40574
31	9.802019
32	15.45317
33	17.87584
34	10.8164
35	10.8377
36	11.4462
37	6.82155
38	10.80242
39	19.35702
40	17.50303
41	16.63369
42	21.50525
43	23.79174
44	10.77436
45	9.775864
46	10.62838
47	9.338187
48	8.149213
49	17.10799
50	26.5875

[PaigeMiller]

As it so often happens here in the SAS Communities, @Rick_SAS has written a blog post on how to compute these confidence intervals on the percentiles. https://blogs.sas.com/content/iml/2013/05/06/compute-confidence-intervals-for-percentiles-in-sas.htm...

You would have to merge these confidence intervals with the normal probability plot in order to get the plot you want. I believe PROC SGPLOT will allow you to draw these points and lines, after you merge the data sets.

[JuLePe]

Thanks, @PaigeMiller , I'll try that.

On the other hand, I just found out that PROC RELIABILITY produces a nice normal probability plot with 95% confidence bands. You have to treat the as uncensored. Basically,

ods graphics on;
ods select ProbabilityPlot;
proc reliability data=bimodal_data;
 probplot Bimodal_Var;
run;
ods select all;
ods graphics off;

It creates this plot:

Without the NOINSET option:

That's not the inset I need. I have to figure out how to create an inset as in the original MINITAB image. That is, with the statistics (and corresponding text) Mean, StDev, N, AD, p-Value (where AD is the Anderson-Darling statistic for the normality test).

[Ksharp · Accepted solution]

You want this ?

data have;
input Obs	Bimodal_Var;
cards;
1	11.18227
2	8.349733
3	14.40938
4	24.66227
5	9.279151
6	7.238792
7	15.46339
8	22.47347
9	15.132
10	21.10275
11	8.501284
12	15.77826
13	16.20075
14	17.67517
15	7.934397
16	18.39657
17	11.40699
18	8.399611
19	8.490487
20	22.70707
21	11.09486
22	11.00888
23	22.85953
24	6.783415
25	11.31967
26	25.19409
27	12.12261
28	10.3787
29	22.13491
30	20.40574
31	9.802019
32	15.45317
33	17.87584
34	10.8164
35	10.8377
36	11.4462
37	6.82155
38	10.80242
39	19.35702
40	17.50303
41	16.63369
42	21.50525
43	23.79174
44	10.77436
45	9.775864
46	10.62838
47	9.338187
48	8.149213
49	17.10799
50	26.5875
;

/*Get N Mean Std */
proc summary data=have  ;
var Bimodal_Var;
output out=mean_std_n n=n mean=mean std=std;
run;
data _null_;
set mean_std_n;
call symputx('mean',put(mean,8.2 -l));
call symputx('std' ,put(std,8.3 -l));
call symputx('n' ,n);
run;
%put &=mean &=std;

/*Get AD statistic and P-Value*/
ods select none;
ods output  TestsForNormality= TestsForNormality;
proc univariate data=have normal ;
var Bimodal_Var;
run;
ods select all;
data _null_;
 set TestsForNormality(where=(Test='Anderson-Darling'));
 call symputx('AD',put(Stat,8.3 -l));
 call symputx('pvalue',cats(pSign,vvalue(pValue)));
run;
%put &=ad &=pvalue;

/*Get the normal probability table*/
ods graphics /reset=index noborder;
ods listing gpath="%sysfunc(pathname(work))" style=htmlblue;; *Save this plot into a path;
ods select ProbabilityPlot;
proc reliability data=have ;
 probplot Bimodal_Var/NOINSET;
run;
ods select all;

%sganno
data sganno;
%SGIMAGE(IMAGE="%sysfunc(pathname(work))\ProbabilityPlot1.png",ANCHOR="topleft",BORDER="FALSE",DRAWSPACE="LAYOUTPERCENT" ,x1=-1,y1=101)
%SGTEXT(LABEL="Mean", BORDER= "FALSE",DRAWSPACE= "LAYOUTPERCENT" , LAYER= "FRONT",X1=10,Y1=90)
%SGTEXT(LABEL="&Mean", BORDER= "FALSE",DRAWSPACE= "LAYOUTPERCENT" , LAYER= "FRONT",X1=20,Y1=90)
%SGTEXT(LABEL="StDev", BORDER= "FALSE",DRAWSPACE= "LAYOUTPERCENT" , LAYER= "FRONT",X1=10,Y1=85)
%SGTEXT(LABEL="&std", BORDER= "FALSE",DRAWSPACE= "LAYOUTPERCENT" , LAYER= "FRONT",X1=20,Y1=85)
%SGTEXT(LABEL="N", BORDER= "FALSE",DRAWSPACE= "LAYOUTPERCENT" , LAYER= "FRONT",X1=10,Y1=80)
%SGTEXT(LABEL="&n", BORDER= "FALSE",DRAWSPACE= "LAYOUTPERCENT" , LAYER= "FRONT",X1=20,Y1=80)
%SGTEXT(LABEL="AD", BORDER= "FALSE",DRAWSPACE= "LAYOUTPERCENT" , LAYER= "FRONT",X1=10,Y1=75)
%SGTEXT(LABEL="&ad", BORDER= "FALSE",DRAWSPACE= "LAYOUTPERCENT" , LAYER= "FRONT",X1=20,Y1=75)
%SGTEXT(LABEL="P-Value", BORDER= "FALSE",DRAWSPACE= "LAYOUTPERCENT" , LAYER= "FRONT",X1=10,Y1=70)
%SGTEXT(LABEL="&pvalue", BORDER= "FALSE",DRAWSPACE= "LAYOUTPERCENT" , LAYER= "FRONT",X1=20,Y1=70)
run;

ods graphics/reset noborder;
proc sgplot data=sashelp.class sganno=sganno;
  scatter x=weight y=height;
run;

[JuLePe]

Thanks! That worked perfectly!!! I would have never figured it out on my own.

[Ksharp]

Yeah. The most difficult thing is the Y axis which is non-linear scale .
Otherwise , I would like to use PROC SGPLOT to replicate this graph .

[JuLePe]

@Ksharp thanks again. Your solution works perfectly for a full-size graph. I have a follow-up question:
I want the final graph scaled down to fit in layout region of "height=3 in width=4 in". (I found the hard way that preimage doesn't resize.)
How would you change you code so that the final plot is resized proportionally to "ods graphics / height=3 in width=4 in", including the (proportionally-resized) inset text? 

[Ksharp]

OK. Here is .

data have;
input Obs	Bimodal_Var;
cards;
1	11.18227
2	8.349733
3	14.40938
4	24.66227
5	9.279151
6	7.238792
7	15.46339
8	22.47347
9	15.132
10	21.10275
11	8.501284
12	15.77826
13	16.20075
14	17.67517
15	7.934397
16	18.39657
17	11.40699
18	8.399611
19	8.490487
20	22.70707
21	11.09486
22	11.00888
23	22.85953
24	6.783415
25	11.31967
26	25.19409
27	12.12261
28	10.3787
29	22.13491
30	20.40574
31	9.802019
32	15.45317
33	17.87584
34	10.8164
35	10.8377
36	11.4462
37	6.82155
38	10.80242
39	19.35702
40	17.50303
41	16.63369
42	21.50525
43	23.79174
44	10.77436
45	9.775864
46	10.62838
47	9.338187
48	8.149213
49	17.10799
50	26.5875
;

/*Get N Mean Std */
proc summary data=have  ;
var Bimodal_Var;
output out=mean_std_n n=n mean=mean std=std;
run;
data _null_;
set mean_std_n;
call symputx('mean',put(mean,8.2 -l));
call symputx('std' ,put(std,8.3 -l));
call symputx('n' ,n);
run;
%put &=mean &=std;

/*Get AD statistic and P-Value*/
ods select none;
ods output  TestsForNormality= TestsForNormality;
proc univariate data=have normal ;
var Bimodal_Var;
run;
ods select all;
data _null_;
 set TestsForNormality(where=(Test='Anderson-Darling'));
 call symputx('AD',put(Stat,8.3 -l));
 call symputx('pvalue',cats(pSign,vvalue(pValue)));
run;
%put &=ad &=pvalue;

/*Get the normal probability table*/
ods graphics /reset=index noborder height=3in width=4in;
ods listing gpath="%sysfunc(pathname(work))" style=htmlblue;; *Save this plot into a path;
ods select ProbabilityPlot;
proc reliability data=have ;
 probplot Bimodal_Var/NOINSET;
run;
ods select all;

%sganno
data sganno;
%SGIMAGE(IMAGE="%sysfunc(pathname(work))\ProbabilityPlot1.png",ANCHOR="topleft",BORDER="FALSE",DRAWSPACE="LAYOUTPERCENT" ,x1=-1,y1=100,
      WIDTH=103,
      WIDTHUNIT="PERCENT",
	  HEIGHT=103,
      HEIGHTUNIT="PERCENT"
)
%SGTEXT(LABEL="Mean", BORDER= "FALSE",DRAWSPACE= "LAYOUTPERCENT" ,TEXTSIZE=8 , LAYER= "FRONT",X1=15,Y1=90)
%SGTEXT(LABEL="&Mean", BORDER= "FALSE",DRAWSPACE= "LAYOUTPERCENT" ,TEXTSIZE=8, LAYER= "FRONT",X1=25,Y1=90)
%SGTEXT(LABEL="StDev", BORDER= "FALSE",DRAWSPACE= "LAYOUTPERCENT" ,TEXTSIZE=8, LAYER= "FRONT",X1=15,Y1=85)
%SGTEXT(LABEL="&std", BORDER= "FALSE",DRAWSPACE= "LAYOUTPERCENT" ,TEXTSIZE=8, LAYER= "FRONT",X1=25,Y1=85)
%SGTEXT(LABEL="N", BORDER= "FALSE",DRAWSPACE= "LAYOUTPERCENT" ,TEXTSIZE=8, LAYER= "FRONT",X1=15,Y1=80)
%SGTEXT(LABEL="&n", BORDER= "FALSE",DRAWSPACE= "LAYOUTPERCENT" ,TEXTSIZE=8, LAYER= "FRONT",X1=25,Y1=80)
%SGTEXT(LABEL="AD", BORDER= "FALSE",DRAWSPACE= "LAYOUTPERCENT" ,TEXTSIZE=8, LAYER= "FRONT",X1=15,Y1=75)
%SGTEXT(LABEL="&ad", BORDER= "FALSE",DRAWSPACE= "LAYOUTPERCENT" ,TEXTSIZE=8, LAYER= "FRONT",X1=25,Y1=75)
%SGTEXT(LABEL="P-Value", BORDER= "FALSE",DRAWSPACE= "LAYOUTPERCENT" ,TEXTSIZE=8, LAYER= "FRONT",X1=15,Y1=70)
%SGTEXT(LABEL="&pvalue", BORDER= "FALSE",DRAWSPACE= "LAYOUTPERCENT" ,TEXTSIZE=8, LAYER= "FRONT",X1=25,Y1=70)
run;

ods graphics/reset noborder height=3in width=4in;
proc sgplot data=sashelp.class sganno=sganno noborder;
  scatter x=weight y=height;
run;

[JuLePe]

@Ksharp PERFECT!!!. Plus I learned a few tricks. Thanks!