[Sergio Goldbaum]

please disregard the previous post. I found out what happened. sorry for the inconvenience.

[Clyde Schechter]

It would be a contribution to our community if you posted what you found out. You probably will not be the last person to encounter the same difficulty, and others may benefit from learning what happened in your case. Your original question was no inconvenience to anybody. Not telling what you found is.

[Jonathan Mickels]

Hey, I have the identical problem. I thought setting a seed is enough to solve it. But unfortunately it didn't work. Probably since I did something wrong while setting the seed? I'd be grateful if somebody could take a look at my code below and help me out!

Thank you!!

use data.dta
set seed 1
collapse (sum) size, by(firm year)
save test.dta

use data.dta
set seed 1
collapse (sum) size, by(firm year)
cf size using test.dta

[Leonardo Guizzetti]

Using a "seed" is not the same as a "sortseed", the former is used for random number generation, and the later is used for random sort order. In either case, this can't be the solution since -collapse- will sort the data according to groups specified by the -by()- variables anyway.

Please, if you are going to request help with code, provide us a data sample that illustrates the problem. We don't have your data.dta, so it's unhelpful for us to troubleshoot.

In any case, the code you have present performs the same way on the same dataset so they should be identical.

[Jonathan Mickels]

Hey Leonardo,

thank you so much for your quick response. Actually, it helped me spotting my mistake. Sortseed does the trick. So the solution proposed in earlier posts works, I just didn't know that sortseed for sorting exists.

[Leonardo Guizzetti]

Hey Leonardo,

thank you so much for your quick response. Actually, it helped me spotting my mistake. Sortseed does the trick. So the solution proposed in earlier posts works, I just didn't know that sortseed for sorting exists.

Thanks for closing the loop on the issue and letting us know how you solved the problem. I'd be curious if you provided a sample of your data that illustrates why sortseed matters here.

[Tim Huegerich]

Here's a replicable example of a basic collapse result dependent on sortseed:

clear all
sysuse nlsw88, clear
keep wage c_city
recast double wage
replace wage = wage + 1000000*wage
expand 1000
preserve
set sortseed 1
collapse (mean) wage, by(c_city)
save "$temp_dir/collapse_test.dta", replace
restore

preserve
set sortseed 1 // needed to replicate exactly
collapse (mean) wage, by(c_city)
cf _all using "$temp_dir/collapse_test.dta", verbose
restore

preserve
collapse (mean) wage, by(c_city)
cf _all using "$temp_dir/collapse_test.dta", verbose // fails
restore

Output of that last cf shows two mismatches, so close (in this case) that the displayed digits are identical, though I have seen larger differences in real-world use cases with confidential data:

            wage: 2 mismatches
                  obs 1. 7599984.6 in master; 7599984.6 in using
                  obs 2. 8172533.6 in master; 8172533.6 in using

Stata/MP 19.0 for Windows (64-bit x86-64) - 6 core license
Revision 28 Jan 2026