topic Re: SAS QLIM with random effects on two subject classes in Statistical Procedures

SAS QLIM with random effects on two subject classes

sasgenie — Mon, 11 Sep 2023 21:35:37 GMT

I'm running a QLIM analysis to analyze a panel data (shopper x store x time). I need to specify the random effects from two sources: shoppers and stores. I recognize that QLIM does not allow more than one random statement.

I need "shopper_id" and "store_id" to both act as subjects for the random effects. Is there any way to trick QLIM to run the following analysis?

proc qlim data=indata heckit; class shopper_id store_id; model choice = price promotion iv / DISCRETE; model y = price promotion / SELECT(choice=1) ; random Int / type=un subject=shopper_id store_id; run;

Re: SAS QLIM with random effects on two subject classes

sbxkoenk — Mon, 11 Sep 2023 22:49:43 GMT

Hello,

Are shoppers hierarchically nested within stores (not one shopper pops up in more than one store)?
In that case, you could try :
subject = shopper_id(store_id);

If not , can't you make a new variable to identify the cross-sectional units (an id for every unique shopper*store combination)?

Like here :

data _null_;
   separator='%%$%%';
   shopper_id='The Olympic  '; 
   store_id  ='   Arts Festival ';
   shopXstore_ID=catx(separator, shopper_id, store_id);
   put shopXstore_ID $char.; 
run;

Then do :

subject = shopXstore_ID;

BR,
Koen

Re: SAS QLIM with random effects on two subject classes

sasgenie — Mon, 11 Sep 2023 22:53:47 GMT

Thanks for the ideas. There is no nesting relationship between the two. Creating a cross variable would results in too many combinations (instead of m + n levels, this will create m*n levels), and my data does not have enough degrees of freedom to estimate it (imagine each shopper shops at each store at most once).

Re: SAS QLIM with random effects on two subject classes

sbxkoenk — Mon, 11 Sep 2023 23:14:53 GMT

Hello,

I looked

at PROC HPQLIM (SAS 9.4 multi-threaded) and
at PROC CQLIM (SAS Viya using CAS engine)

, but these will not help you out.

Following support note will not help you out either , but it's a nice read :

Usage Note 67037: Estimating limited and discrete dependent variable models with random parameters or random effects using PROC QLIM
https://support.sas.com/kb/67/037.html

Koen