Stratified Random Sampling by Group

[HoaTruong]

Hi,

I'd like to request some assistance with the following programming query.

I have an input dataset (please see attachement) - small sample provided below.  It consists of one row per person.  Each row belongs to a specific campaign cell and is given a model score (values from 1 - 10).

Cell	Individual	ModelScore
100	Person1	1
100	Person2	2
100	Person3	3
100	Person4	4
100	Person5	5
100	Person6	6
100	Person7	7
100	Person8	8
100	Person9	9
100	Person10	10
100	Person11	1
100	Person12	2
101	Person13	3
101	Person14	4
101	Person15	5
101	Person16	6
101	Person17	7
102	Person18	8
102	Person19	9
102	Person20	10

For each value of Cell, I would like to perform a random stratified (strata variable: ModelScore) sample to split the data into 20 roughly equal-sized groups.  For example, the subset of rows that belong to cell 100 will be split into 20 groups, using the ModelScore variable. I would like this to be repeated for each value of Cell in the input dataset.

I was thinking of using PROC SURVEYSELECT with the GROUPS= option, but wasn't sure if there is a better approach. 

Many thanks,

Hoa

[Ksharp · Accepted solution]

What do you mean by " using the ModelScore variable. " ?
you want group it from smallest to largest ?

proc ranks data=have groups=20;
by cell;
var modelscore;
rank rank;
run;