How to Double Sort?

[LanMin]

Hi,

1. I have a panel dataset (below is a small sample). I would like to sort the data independently by Uvalue, and Svalue into quartiles,

By independently, I mean if I use proc rank to output 4 ranks based on Uvalue, I do not consider Svalue. And vice versus if I output based on Svalue.

After the two sorts are done,

I expect 4 ranks within each variable category. 4*4 =16 cells,

My goal is to test whether Dvalue differs significantly within each cell.

2. That is does Dvalue in (Quartile 1 based on Uvalue rank) is significantly different from Dvalue in Quartile 1 based on Dvalue rank.

I know how to single sort, but need your help with double sort, and save the output for T-test in part 2.

thanks!

Lan

firmid	year	Uvalue	Svalue	Dvalue
1001	1992	0.3	10	100
1001	1993	0.6	11	90
1001	1994	1	11	90
1001	1995	0.9	12	100
1001	1996	1.1	12	110
1001	1997	0.92	12	110
1001	1998	1.1	13	111
1002	1993	0.2	9	20
1002	1994	0.3	9	22
1002	1995	0.4	9.1	23
1002	1996	0.9	10	25
1002	1997	0.7	10	30
1003	1995	1	6	15
1003	1996	1.1	5	15
1003	1997	1.4	7	16

[PGStats]

Proc RANK does its own sorting for each variable. You don't have to worry about it. So the answer to your first question is straightforward. The second question however isn't valid statistically as the two sets that you want to compare are not independent. I would simply look at the frequency and Dvalue mean for each combination of quartiles :

proc rank data=have out=haveQuart groups=4;
var Uvalue Svalue;
ranks Uquart Squart;
run;

proc means data=haveQuart n mean nonobs printaltypes;
class Uquart Squart;
var Dvalue;
run;

PG