Creating a Sample from two categorical variables. Looking for replication.

[Felix Quinones]

Hello Statalist,

This is my second post, sorry in advance if I broke a rule.

I have a data that contains 140 observations with over 30 variables, but three of those variables are the important ones: corpname (Corporation Name), LPLLC and PBC. The last two variables are categorical (my individual groups).

I'm trying to make a sample that can be replicated within the individual groups (categorical variables), but every time I run the sample, I get different observations. What can I do to get a sample that can be replicated over time?

I'm using the next code to generate the specific samples from every subgroup.

set seed 988
set obs 140

// Create a Random Sample of LPLLC

sort LPLLC
by LPLLC: count

sample 2 if LPLLC==1, count
sample 2 if LPLLC==2, count
sample 2 if LPLLC==3, count
sample 2 if LPLLC==4, count
sample 2 if LPLLC==5, count
sample 2 if LPLLC==6, count

// Create a Random Sample of PBC

sort PBC
by PBC: count

sample 2 if PBC==1, count
sample 1 if PBC==2, count
sample 1 if PBC==3, count
sample 2 if PBC==4, count
sample 2 if PBC==5, count
sample 1 if PBC==6, count

//Show the List for LPLLC

list corpname LPLLC if LPLLC1==1
list corpname LPLLC if LPLLC2==1
list corpname LPLLC if LPLLC3==1
list corpname LPLLC if LPLLC4==1
list corpname LPLLC if LPLLC5==1
list corpname LPLLC if LPLLC6==1

//Show the List for PBC

list corpname PBC if PBC1==1
list corpname PBC if PBC2==1
list corpname PBC if PBC3==1
list corpname PBC if PBC4==1
list corpname PBC if PBC5==1
list corpname PBC if PBC6==1

An example of my data is as follow (I believed I did not add this correctly here):

* Example generated by -dataex-. To install: ssc install dataex
clear
input float(LPLLC PBC) str89 corpname
5 . "2730 L3C"                                                                                
4 . "AQUARYTHM LPLLC"                                                                          
2 . "PANADERIA Y CAFETERIA DON NESTOR LOW PROFIT LIMITE..."                                    
6 . "PAWSOME BLISS LPLLC"                                                                      
2 . "PLANACTION L3C"                                                                          
4 . "PORTAL SOCIAL L3C"                                                                        
4 . "PRCOMMERCE L3C"                                                                          
3 . "PROSEDUC COMPAÑÍA DE RESPONSABILIDAD LIMITADA CON..."                                  
. 1 "PROYECTO BUCARABON C.B.S."                                                                
. 1 "PROYECTO COMUNITARIO PORTUGUÉS CBS"                                                      
1 . "PUBLLISHERS EDITORIAL BIEKE LIBRE LPLLC"                                                  
4 . "PUERTA AZUL L.P.L.L.C."                                                                  
2 . "ARAYAN CONSTRUCTION LOW PROFIT LIMITED LIABILITY C..."                                    
. 1 "PUERTO RICO SPECIALTY COFFEE ASOCIATION C.B.S."                                          
. 3 "RECICLA TU CEL, INC. C.B.S."                                                              
                                 
end
label values LPLLC LPLLC
label def LPLLC 1 "0y LPLLC !=SJ", modify
label def LPLLC 2 "0y LPLLC ==SJ", modify
label def LPLLC 3 "1y LPLLC !=SJ", modify
label def LPLLC 4 "1y LPLLC ==SJ", modify
label def LPLLC 5 "2y+ LPLLC !=SJ", modify
label def LPLLC 6 "2y+ LPLLC ==SJ", modify
label values PBC PBC
label def PBC 1 "0y PBC !=SJ", modify
label def PBC 2 "0y PBC ==SJ", modify
label def PBC 3 "1y PBC !=SJ", modify
label def PBC 4 "1y PBC ==SJ", modify
label def PBC 5 "2y+ PBC !=SJ", modify
label def PBC 6 "2y+ PBC ==SJ", modify

Thanks,

Félix Quiñones.

[Andrew Musau]

help sortseed

[Maarten Buis]

If you sort on a variable with ties, i.e. a variable where multiple observations have the same value, then there is a problem: How do you sort the numbers 2 and 2? There is no way to decide which 2 is larger, so Stata sorts them randomly. Moreover, the randomness is determined by its own seed. So set seed won't force the ordering to be stable across iterations, as you found out. I would say that normally that is a good thing, but if you wish to fix the order you could add the stable option to sort, see help sort.

[Felix Quinones]

Maarten,

Thanks for your reply.

The numbers 2 and 2 are the number of sample corporation (observations) that I want to see by every "bucket" (subgroup). This is because I need a representative sample with the same number of observations across years of establishment, type of organization, and location. That constraints are already added in the categorical variable.

Your comment about the - stable option - in sort works perfect for me.

Thanks.