topic Re: Calculate percentile based on Z-score in SAS Programming

Calculate percentile based on Z-score

Cruise — Tue, 08 May 2018 13:15:00 GMT

Hi guys, I'd like to calculate percentile based on z-score. My data is massive where using z-score-percentile conversion table is futile. Please help? Below is first few observation of my data. Thanks!!!!

data p.temp; 
input zcore;
cards;
0.16
3.1
-0.33
1.72
;

Re: Calculate percentile based on Z-score

PaigeMiller — Tue, 08 May 2018 13:22:46 GMT

You can use the PROBNORM function

http://documentation.sas.com/?cdcId=pgmmvacdc&cdcVersion=9.4&docsetId=lefunctionsref&docsetTarget=p0am6dtkvnrb09n132vioa1z6oen.htm&locale=en

Re: Calculate percentile based on Z-score

Cruise — Tue, 08 May 2018 13:34:20 GMT

Thanks a lot @PaigeMiller.

data temp1; set temp;
pctl=probnorm(zcore);
run;

zcore pctl

0.16 0.56

3.1 0.99

-0.33 0.37

1.7 0.96

Please correct me if I'm wrong. But it looks like it. I will double check with the conversion table as well.

Re: Calculate percentile based on Z-score

Ksharp — Tue, 08 May 2018 14:10:25 GMT

Function PROBNORM() is too old ,Check CDF().

data temp; 
input zcore;
p=cdf('normal',zcore);
cards;
0.16
3.1
-0.33
1.72
;

proc print;run;

Re: Calculate percentile based on Z-score

PaigeMiller — Tue, 08 May 2018 14:16:07 GMT

Too old? What risk does this pose to the user? The function is still supported by SAS, and the probabilities of a normal distribution haven't changed.

Re: Calculate percentile based on Z-score

Ksharp — Tue, 08 May 2018 14:41:11 GMT

It is better to explain by @Rick_SAS

Re: Calculate percentile based on Z-score

FreelanceReinh — Tue, 08 May 2018 17:22:32 GMT

Yes, there is a caveat regarding the "older" PROBXXX functions in Rick's Blog ("The DO Loop"). Just found it again: https://blogs.sas.com/content/iml/2013/07/10/stop-using-ranuni.html (at the bottom of the blog entry).

Re: Calculate percentile based on Z-score

PaigeMiller — Tue, 08 May 2018 17:34:29 GMT

Thank you for the link, @FreelanceReinh. That's very interesting.

If I am understanding properly, the new functions will perform better when you have very small or very large values, but in the range of –10 to +10, I just wrote a program and PROBNORM(x) and CDF('normal',x) give the same answers (at least to 14 decimal places).

In any event, SAS ought to update the PROBNORM (and similar functions) documentation to specifically say that users ought to use the newer CDF function, and state the reasons.