Calculating p values with SAS
http://www.listserv.uga.edu/cgi-bin/wa?A2=ind0101e&L=sas-l&P=9455
*/
%let Distribution='NORMAL';
%let location=0; *Note: Location parameter, a.k.a. Mean;
%let scale=1.0; *Note: Scale parameter a.k.a. Std Dev;
%let z=1.625; * Example z value;
data junk;
do z=-5 to 5 by 0.01;
dens=pdf(&Distribution,z,&location,&scale);
cum=cdf(&Distribution,z,&location,&scale);
output;
end;
call
symput('cdf_value',put(cdf(&Distribution,&z,&location,&scale),best6.5));
if &z ge 0 then do;
phigher=1-cdf(&Distribution,&z,&location,&scale);
call symput('p_value',put(phigher,best6.5));
output;
end;
else do;
plower=cdf(&Distribution,&z,&location,&scale);
call symput('p_value',put(plower,best6.5));
output;
end;
run;
data labelDensity; * this is just an annotate dataset to label curves;
length text$ 32 color $8;
xsys='2'; * X scale is same as plot scale;
x=&z; * plot label at x=z value;
ysys='3'; * Y scale is in pct of plot area;
y=25; * plot label at 25% of y space;
color='Blue';
if &z ge 0 then text="Prob z > &z is &p_value";
else text="Prob z < &z is &p_value";
output;
run;
data labelCum; * define label for cum prob plot;
length text$ 32 color $8;
xsys='3'; * X scale is in pct of x space;
x=68; * plot label at x=z value;
ysys='2'; * Y scale is same as plot scale;
y=&cdf_value-0.0075; * plot label at value of cdf;
color='Green';
text="Cum. Prob at z=&z is &cdf_value";
output;
run;
goptions vsize=5.25 hsize=5.75;
goptions ftitle=swiss ftext=swiss htitle=4 pct;
axis1 label=(angle=90);
axis3 label=(angle=-90);
symbol1 c=blue i=spline v=none l=1;
symbol2 c=green i=spline v=none l=1;
title1 "&Distribution Distribution with Location=&location and Scale=&scale
";
proc gplot data=junk;
plot dens*z / href=&z chref=blue lhref=2
vaxis=axis1 annotate=labelDensity;
plot2 cum*z / vref=&cdf_value cvref=green lvref=2
vaxis=axis3 annotate=labelCum;
label dens="Probability Density Function (Blue Trace)"
cum="Cumulative Probability (Green Trace)";
run;
quit;
* Now we with play with another distribution.Let's try Weibull;
%let Distribution='WEIBULL';
%let location=1.0; *Define an arbitrary location param;
%let scale=2.0; *Choose a Scale Param;
%let z=5.0; *choose an example z value;
data junk;
do z=-1 to 10 by 0.01;
dens=pdf(&Distribution,z,&location,&scale);
cum=cdf(&Distribution,z,&location,&scale);
output;
end;
call
symput('cdf_value',put(cdf(&Distribution,&z,&location,&scale),best6.5));
if &z ge 0 then do;
phigher=1-cdf(&Distribution,&z,&location,&scale);
call symput('p_value',put(phigher,best6.5));
output;
end;
else do;
plower=cdf(&Distribution,&z,&location,&scale);
call symput('p_value',put(plower,best6.5));
output;
end;
run;
data labelDensity; * this is just an annotate dataset to label curves;
length text$ 32 color $8;
xsys='2'; * X scale is same as plot scale;
x=&z; * plot label at x=z value;
ysys='3'; * Y scale is in pct of plot area;
y=25; * plot label at 25% of y space;
color='Blue';
if &z ge 0 then text="Prob z > &z is &p_value";
else text="Prob z < &z is &p_value";
output;
run;
data labelCum; * define label for cum prob;
length text$ 32 color $8;
xsys='3'; * X scale is in pct of x space;
x=68; * plot label at x=z value;
ysys='2'; * Y scale is same as plot scale;
y=&cdf_value-0.0075; * plot label at value of cdf;
color='Green';
text="Cum. Prob at z=&z is &cdf_value";
output;
run;
goptions vsize=5.25 hsize=5.75;
goptions ftitle=swiss ftext=swiss htitle=4 pct;
axis1 label=(angle=90);
axis2 order=(-1 to 10 by 1); * note...this will have to change based on
data;
axis3 label=(angle=-90);
symbol1 c=blue i=spline v=none l=1;
symbol2 c=green i=spline v=none l=1;
title1 "&Distribution Distribution with Location=&location and Scale=&scale
";
proc gplot data=junk;
plot dens*z / href=&z chref=blue lhref=2
vaxis=axis1 haxis=axis2 annotate=labelDensity;
plot2 cum*z / vref=&cdf_value cvref=green lvref=2
vaxis=axis3 haxis=axis2 annotate=labelCum;
label dens="Probability Density Function (Blue Trace)"
cum="Cumulative Probability (Green Trace)";
run;
quit;
* Now we with play with another distribution.
Let's try Laplace;
%let Distribution='LAPLACE';
%let location=1.0; *Choose a location param;
%let scale=2.0; *Choose a scale param;
%let z=-5.0; *Choose an example z ;
data junk;
do z=-10 to 10 by 0.01;
dens=pdf(&Distribution,z,&location,&scale);
cum=cdf(&Distribution,z,&location,&scale);
output;
end;
call
symput('cdf_value',put(cdf(&Distribution,&z,&location,&scale),best6.5));
if &z ge 0 then do;
phigher=1-cdf(&Distribution,&z,&location,&scale);
call symput('p_value',put(phigher,best6.5));
output;
end;
else do;
plower=cdf(&Distribution,&z,&location,&scale);
call symput('p_value',put(plower,best6.5));
output;
end;
run;
data labelDensity; * this is just an annotate dataset to label curves;
length text$ 32 color $8;
xsys='2'; * X scale is same as plot scale;
x=&z; * plot label at x=z value;
ysys='3'; * Y scale is in pct of plot area;
y=25; * plot label at 25% of y space;
color='Blue';
if &z ge 0 then text="Prob z > &z is &p_value";
else text="Prob z < &z is &p_value";
output;
run;
data labelCum; * define label for cum prob;
length text$ 32 color $8;
xsys='3'; * X scale is in pct of x space;
x=68; * plot label at x=z value;
ysys='2'; * Y scale is same as plot scale;
y=&cdf_value-0.0075; * plot label at value of cdf;
color='Green';
text="Cum. Prob at z=&z is &cdf_value";
output;
run;
goptions vsize=5.25 hsize=5.75;
goptions ftitle=swiss ftext=swiss htitle=4 pct;
axis1 label=(angle=90);
axis2 order=(-10 to 10 by 2); * note this will have to change based on
data;
axis3 label=(angle=-90);
symbol1 c=blue i=spline v=none l=1;
symbol2 c=green i=spline v=none l=1;
title1 "&Distribution Distribution with Location=&location and Scale=&scale
";
proc gplot data=junk;
plot dens*z / href=&z chref=blue lhref=2
vaxis=axis1 haxis=axis2 annotate=labelDensity;
plot2 cum*z / vref=&cdf_value cvref=green lvref=2
vaxis=axis3 haxis=axis2 annotate=labelCum;
label dens="Probability Density Function (Blue Trace)"
cum="Cumulative Probability (Green Trace)";
run;
quit;
* This time we play with a discrete distribution;
%let Distribution='BINOMIAL';
%let p=0.01; *Prob of success;
%let n=100; *Number of trials ;
%let m=3; *Number of 'successes';;
%let tail='upper';
*%let plotstop=5;
data junk;
do m=0 to &n;
dens=pdf(&Distribution,m,&p,&n);
cum=cdf(&Distribution,m,&p,&n);
output;
end;
call symput('cdf_value',put(cdf(&Distribution,&m,&p,&n),best6.5));
call symput('plotstop',put(&m+3,best6.0));
call symput('refline',put(&m+1,best6.0));
if &tail='lower' then do;
phigher=cdf(&Distribution,&m,&p,&n);
call symput('p_value',put(phigher,best6.5));
output;
end;
else do;
plower=1-cdf(&Distribution,&m,&p,&n);
call symput('p_value',put(plower,best6.5));
output;
end;
run;
data labelDensity; * this is just an annotate dataset to label curves;
length text$ 32 color $8;
xsys='2'; * X scale is same as plot scale;
x=&refline; * plot label at x=z value;
ysys='3'; * Y scale is in pct of plot area;
y=25; * plot label at 25% of y space;
color='Blue';
if &tail='upper' then text="Prob > &m successes=&p_value";
if &tail='lower' then text="Prob <= &m successes=&p_value";
output;
run;
data labelCum; * define label for cum prob;
length text$ 32 color $8;
xsys='3'; * X scale is in pct of x space;
x=68; * plot label at x=z value;
ysys='2'; * Y scale is same as plot scale;
y=&cdf_value-0.0075; * plot label at value of cdf;
color='Green';
text="Cum. Prob for m<=&m is &cdf_value";
output;
run;
goptions vsize=5.25 hsize=5.75;
goptions ftitle=swiss ftext=swiss htitle=4 pct;
axis1 label=(angle=90);
axis2 order=(0 to &plotstop); * note this will have to change based on
data;
axis3 label=(angle=-90);
symbol1 c=blue i=stepj v=none l=1;
symbol2 c=green i=spline v=none l=1;
title1 "&Distribution Distribution";
title2 "Probability of success=&p, Number of Trials=&n, m=&m succeses";
proc gplot data=junk;
plot dens*m / href=&refline chref=blue lhref=2
vaxis=axis1 haxis=axis2 annotate=labelDensity;
plot2 cum*m / vref=&cdf_value cvref=green lvref=2
vaxis=axis3 haxis=axis2 annotate=labelCum;
label dens="Probability Density Function (Blue Trace)"
cum="Cumulative Probability (Green Trace)";
run;
quit;