global or local seed setting in loop

[Toni Moreno]

Dear all,

I want to run a loop in which the seed changes for every single run:

forvalues mc = 1/100 {


    * global - seed setter! - same seed in every dofile
    global mc `mc'
    set seed `mc'

    use "${data}run_monte_carlo.dta", clear
    
    xtset pid syear
    
    *** run do-files
    
    do "${do}dofile_1.do"
    
    do "${do}dofile_2.do"
    
        save "${data}loop.dta", replace
    
        use "${data}monte_carlo.dta", clear
    
        merge 1:1 pid syear using "${data}loop.dta", keep (1 3) keepusing (varnames) nogen
        
        save "${data}monte_carlo.dta", replace
        }

where the seed should be equal to `mc'.

I also want to set the seed in the dofiles 1 and 2 to `mc'.

How can I do that? Any help would be really appreciated.
Thank your very much in advance.

[Joro Kolev]

I am afraid I do not understand.

mc is defined somewhere out...
mc is not changing every round
mc is dereferences as a local, not a global macro...

[daniel klein]

forvalues mc = 1/100 {


* global - seed setter! - same seed in every dofile
global mc `mc'
set seed `mc'

Setting the seed here is unnecessary if it is to be reset in the do-files later because there is nothing random in the code that follows.

You can pass arguments to do-files much like you pass arguments to Stata commands; try something like

...
*** run do-files

do "${do}dofile_1.do" `mc'

do "${do}dofile_2.do" `mc'
...

the in the do-files, refer to `1', e.g.

*** first line of do-file

version 15

set seed `1'

...

Best
Daniel

[Toni Moreno]

I thought that since the code start with:

 forvalues mc = 1/100 {

that mc would change from being 1 in the first round to 2 in the second round and so on ...

And in every round I would like to set the seed to mc - so to 1 in the first round, to 2 in the second round and so on.

However, since I use the global "do" to run my dofiles, I am not sure how to make sure that the seed for the estimations in the dofiles 1 and 2 is also set to mc.

[Joro Kolev]

Toni, does then not what you are doing work by itself? I mean if you have defined a global macro, in my understanding then it is available everywhere, so you just need to reference to it in the sub-do files, some somewhere in both do files you should have the seed set to the dereferenced value of the global?

This is my understanding of how globals work from reading the manual, this is why they are global because they are globally available... But to be honest I never use globals, so I dont have much experience. Here is a link explaining the difference between global and local.

https://blog.stata.com/2015/11/03/pr...-local-macros/

[Toni Moreno]

Thank you Daniel!

Maybe it was misleading - the dofiles are different from each other and they do different things, however the seed in the dofiles should be `mc' and should therefore change in every round.

So instead of dofile_1 and dofile_2, imagine them as "dofile_a" and "dofile_b"

[Joro Kolev]

Notwithstanding the curious fact of Stata that Daniel just taught us (I did not know that one can pass arguments to do files...)...

I re-read the Stata blog post I referenced, and it seems to me that what I was saying is correct.

You just start defining the global macro as you have done in your code in #1 and then all you have to have is one line of code before the relevant randomisation starts in both

dofile_1.do and
dofile_2.do saying

 set seed $mc

[daniel klein]

Toni

I am sorry, I cannot make sense of your latest post; nothing of what I have written concerns the do-file names. I was trying to say:

1. You do not need to define global macros; local macro will suffice.
2. You do not need to set the seed inside the loop you show in your original post since none of the commands that you show relies on any random numbers
3. You pass arguments, in your case the contents of local macro mc, to the sub-do-files by just adding them after the do-file's names (whatever these names may be).
4. You refer to these arguments inside the sub-do-files by position: `1' is the first argument, `2' is the second, and so on; `0' contains all arguments.

If you prefer global macros (these can mess things up and lead to hard to identify bugs, especially when used with cryptic, simple names such as mc) that is your choice. In this case, Joro is right, and you can just reference the global macros inside your sub-do-files. There is still no need to set the seed inside teh loop, though.

Best
Daniel

[daniel klein]

Notwithstanding the curious fact of Stata that Daniel just taught us (I did not know that one can pass arguments to do files...)...

It is documented quite clearly. I see that this is easily overlooked because you typically do not bother reading

help do

as its common use is pretty intuitive.

Best
Daniel

[William Lisowski]

Toni Moreno Without understanding anything about the objective of your code, I feel obliged to point out the following discussion about the advisability, or lack of advisability, of setting the seed repeatedly, in the output of help set seed (for Stata 15.1, at least).

If you have not reviewed that material with reference to the work you describe in post #1, I recommend it to your attention. Or better yet, the full discussion in the Stata Base Reference Manual PDF included in your Stata installation and accessible from Stata's Help menu.

[Toni Moreno]

Thank you very much everybody for your help. And thank you William for pointing out the discussion about the advisability of setting the seed repeatedly. I will have a read.