Use a string variable (with one unique entry) to build a local

[Frank Taumann]

I'm trying to call Stata functions from R and I need to pass more information from one program to the other than just a simple dataset. As a workaround, I simply built string variables that hold the other pieces of information (for example, a variable name that changes dynamically in my R program).

In Stata now, I would like to get the value of this variable (which is a variable name actually) to use it in my regression. Here is an example:

sysuse auto, clear

* R runs through a loop where "somename" changes many times so I want to use
* it dynamically in Stata. Say here we are at this one:
gen somename="trunk"

* now I want to use the value of "somename" (which is the variable name "trunk")
* in a regression that should output:

reg price mpg trunk

* here is how I would usually do this
local myname trunk
reg price mpg `myname'

But I don't know how I can get the information from the "somename" variable to my "myname" local. I've tried this:

levelsof(somename)

which already looks like a macro but I don't know how to call it later. This post here tells me to use display but that doesn't seem to work with levelsof.

local myname levelsof(somename)
local myname display levelsof(somename)

Any help would be appreciated!

[Nick Cox]

 
 reg price mpg `= somename[1]'

As the value is the same, any observation will do for looking it up, and observation 1 will do as well as any other.

[Frank Taumann]

Is there a way to make this work if somename[1] is a missing and the missings change dynamically? Sorry that my example didn't include that. I have missings and the variable name in the variable I need to base the decision on.

edit: As a workaround, we can just make sure all the missings have the same values as in my original example:

replace somename= somename[_N]

this is from here and from here.

And the example where it would matter would look like this:

* the data could look like this:
sysuse auto, clear
gen somename="trunk"
replace somename = "" if foreign == 0

* and we could run:
replace somename = somename[_n-1] if missing(somename)
replace somename= somename[_N]
* I think this should work no matter where the missings are hiding (first, last, in between) as long as one value not-missing.

*then run Nicks regression:
reg price mpg `= somename[1]'

[Nick Cox]

Well, think it through: string missing means empty, so Stata sees nothing different once somename has been evaluated if its value is empty. I am taking you as meaning what you say, that there is one unique entry (in the jargon I recommend, that's one distinct value).

A rider: I don't have good advice on how to mix Stata and R like this. I know R exists and I've used it occasionally but that's where my knowledge starts and stops. My instinct is to look out for Stata programs to interface with R that others have written earlier. I think you may need a new thread with R in the title to catch better informed people.

[Frank Taumann]

Your answer has completely solved my problem! I just didn't prepare my dataset well (see my edit). I am running Stata out of R using the RStata package there and it works perfectly so far.

Thanks!