topic Re: PSI stability index in SAS Programming

PSI stability index

Ronein — Tue, 15 Mar 2022 10:39:22 GMT

Hello

I would like to calculate PSI stability index for the following data.

May anyone show the code in SAS to calculate it.

PCT_DEV is percentile of customers in category from all customers (Development time)

PCT_last is percentile of customers in category from all customers (last time)

Data have;
input category  PCT_Dev  PCT_last;
cards;
1 0.9864649963 0.9923499412
2 0.0135350037 0.0076500588
;
Run;

Re: PSI stability index

Amir — Tue, 15 Mar 2022 11:16:59 GMT

Hi,

Does this paper help?

Kind regards,

Amir.

Re: PSI stability index

Ronein — Tue, 15 Mar 2022 11:28:57 GMT

May you also show here ? Please direct the code to the example I sent,cheers

Re: PSI stability index

Amir — Tue, 15 Mar 2022 12:23:38 GMT

Hi,

I had never heard of Population Stability Index until your post, so I don't think I'd be best placed to assist.

I just Googled "PSI site:sas.com" and the first hit is the paper I sent in my post.

If the paper is not helpful then I would suggest you try performing other searches to get results from any site. Or even try posting to statistical forums here and elsewhere, or whatever PSI would be considered a sub-topic of.

If you have tried going through the paper and attempted a solution, but are getting errors then try posting the errors here to see if anyone else can help. Alternatively, are you able to ask a colleague of yours?

Lastly, I'm not sure if your confusion is over the general method of how to calculate PSI (which one can Google).

HTH.

Kind regards,

Amir.

Re: PSI stability index

Ksharp — Wed, 16 Mar 2022 12:35:59 GMT


/*********** Group=10 ****************/
data have;
input n min max n_good_real n_good_pred  ;
cards;
0 406 468 24 20
1 470 484 45 42 
2 485 495 43 48
3 496 508 64 66
4 509 520 80 78
5 521 531 80 84
6 532 544 85 84
7 545 556 89 87
8 557 571 97 95 
9 572 624 93 96
;
run;

title 'Group=10';
proc sql;
create table temp as
select n label='组号',min label='最小得分' ,max label='最大得分',
n_good_real label='实际好客户个数',
n_good_pred label='预测好客户个数',
 n_good_real/(select sum(n_good_real) from have) as A label='实际好客户占比 A' format=percent7.2,
 n_good_pred/(select sum(n_good_pred) from have) as E label='预测好客户占比 E' format=percent7.2,
 calculated A - calculated E as A_E label='A-E',
 log(calculated A/calculated E) as logAE label='LOG(A/E)' ,
 calculated A_E *calculated logAE as PSI
   from have ;

select * from temp;
select sum(PSI) as PSI from temp;

select n_good_pred into : list separated by ' ' from have;
quit;

%put &list ;

title2 'Chi-Square Goodness-of-Fit Tests';
proc freq data=have order=data;
tables n / nocum chisq testf=( &list ) plots(only)=deviationplot(type=dotplot);
weight n_good_real;
run;

Re: PSI stability index

Ronein — Wed, 16 Mar 2022 05:58:42 GMT

Thanks and great.
I think that you forgot to calculate the Multiplication of (A-E)*(Log (A/E) ) and then sum this values to get PSI.

calculated A - calculated E as A_E label='A-E',
log(calculated A/calculated E) as logAE label='LOG(A/E)',
calculated A_E *calculated logAE as PSI

Re: PSI stability index

Ksharp — Wed, 16 Mar 2022 12:36:33 GMT

OK. I changed my code. Hope someone need it .

Re: PSI stability index

jorgelobin — Mon, 08 Apr 2024 08:30:09 GMT

On this page there is a complete example with all the steps:
https://www.sasdesdecero.com/implementacion-en-sas-del-population-stability-index/