Sample size calculations for a one-sided tolerance limit

[djgetty]

I need to determine the minimum sample size/create a power curve for computing a 95%-95% upper tolerance limit. I see that proc power has functionality for many power computations but not for tolerance intervals.  From what I have investigated thus far, I would need to set up monte carlo simulations to create the power curve. Is there an easier way? Minitab and other packages have this functionality through menu driven platforms so I am hoping that with the inclusiveness of SAS procedures there is a way to perform this task.

Any advice would be appreciated.

[sbxkoenk]

Hello,

I believe in JMP you can also do this through a menu driven platform.

(JMP is a wholly owned subsidiary of SAS --> https://community.jmp.com/)

In SAS, you will need some programming.

Start here :

• Using simulation to estimate the power of a statistical test
  By Rick Wicklin on The DO Loop May 30, 2013
  https://blogs.sas.com/content/iml/2013/05/30/simulation-power.html
• Using simulation to compute a power curve
  By Rick Wicklin on The DO Loop June 5, 2013
  https://blogs.sas.com/content/iml/2013/06/05/simulation-power-curve.html
• Estimate a power curve in parallel in SAS Viya
  By Rick Wicklin on The DO Loop August 17, 2020
  https://blogs.sas.com/content/iml/2020/08/17/power-curve-parallel-sas-viya.html

Koen

[FreelanceReinh · Accepted solution]

Hello @djgetty and welcome to the SAS Support Communities!

Looking at the formulas in the Minitab® 20 documentation Methods and formulas for Sample Size for Tolerance Intervals, I think you can compute the sample size in a DATA step. You can also define a function (PROC FCMP) for this purpose. The latter will be particularly convenient when it comes to creating graphs visualizing the relationship between sample size and other statistics in the context of tolerance intervals. (How do you define a "power curve" for a tolerance interval rather than a hypothesis test?)

Here is an example for the case of a normal distribution: The DATA step below simply passes through the sample sizes n=2, 3, 4, ..., until the criterion for the one-sided tolerance interval is met (see the Minitab documentation linked above; I denote their parameters a, a* and e by a, ax and e, respectively).

data want;
a=0.05;
p=0.95;
e=0.03;
ax=0.05;
do n=2 to 1e6 until(q1<=q2);
  q1=quantile('t',1-a,n-1,probit(p)*sqrt(n));
  q2=quantile('t',ax,n-1,probit(p+e)*sqrt(n));
end;
put n=;
run;

The same result, n=177, can be obtained with the user-defined function ntolu defined below:

proc fcmp outlib=work.funcs.test;
function f(a,p,e,ax,n);
  return(quantile('t', ax,n-1,sqrt(n)*probit(p+e))
        -quantile('t',1-a,n-1,sqrt(n)*probit(p)));
endsub;

function ntolu(a,p,e,ax,init);
  n=ceil(solve('f',{init},0,a,p,e,ax,.));
  return(n);
endsub;
run;

options cmplib=work.funcs;

data _null_;
n=ntolu(0.05,0.95,0.03,0.05,100);
put n=;
run;

For other parameter values you may or may not need to adjust the initial value (init=100) in the last argument of the function.

Simulated data could be used to verify that the sample size is sufficient to achieve the target values specified in the parameters.

[djgetty]

Thank you. This is the type of solution I was looking for. I will give it a try and see where it goes. I appreciate you taking the time to help me with this issue!