topic Re: Quantile for chi square in SAS Programming

Quantile for chi square

sthudson0 — Sat, 20 Feb 2016 23:16:47 GMT

Hello,

I am trying to use SAS for the first time and I am stuck. My first homework problem is : Find the quantile for chi squared with 4 df at a probablity of .47.

How can I write this program?

Thank you fo rnay help you can give me.

sthudson0

Re: Quantile for chi square

Reeza — Sat, 20 Feb 2016 23:39:14 GMT

My stats is getting rusty these days, but I think you're looking for one of the following:

cumulative distribution function for chi square (CDF) - not likely

probability distribution function for chi square (PDF) - possible

Cumulative inverse - quantile (CINV) - likely

data want;
prob=0.47;
df=4;
y=probchi(0.47, 4);
z=cinv(0.47, 4); *<-most likely this number;
run;

Re: Quantile for chi square

sthudson0 — Sat, 20 Feb 2016 23:57:17 GMT

Thank you very much. I wil lgive this a try.

Re: Quantile for chi square

PGStats — Sun, 21 Feb 2016 05:14:06 GMT

quantile('CHISQ', .47, 4); /* definitely */

Re: Quantile for chi square

FreelanceReinh — Tue, 23 Feb 2016 12:54:13 GMT

Just a side note: It seems that the two functions are implemented differently. In this particular case the results differ in the last bit (i.e. 2^-51) on my Windows machine. (Irrelevant for practical purposes, of course.)

data _null_;
c=cinv(.47, 4);
q=quantile('CHISQ', .47, 4);
if c ne q then do;
  put 'Not exactly equal.';
  d=c-q;
  put 'Difference: ' d;
  put (c q) (hex16. /);
end;
run;

So, which one is closer to the "true" result?

Well, it depends ... For the exact argument of 0.47=47/100 c is closer, but for the internal 64-bit floating-point representation of 0.47 (which equals 0.47 - 0.96*2^-55) q is closer, says my computer algebra software.