Temporarily keep a country in a panel

[Evan Gensler]

Starting from a panel data I would like to estimate a reg for each country separately and produce a figure for each country then combining these figures.

The problem is that the estimation is producing weird results, if I simply run a regression and end it with e.g. if country=="USA". (I am running a non-linear regression by the way.)

Is there a way to temporarily keep one country similar to

keep if country=="USA"

Without Stata erasing the other panel from its memory. So I can jump to the next country and do:

keep if country=="JPN"

Sure, I can reload the panel data for each regression, but I would lose the graphs from the memory (I would think) and won't be able to combine them.

Thank you

[Carlo Lazzaro]

Evan:
wouldn't it be simpler going:

bysort country: regress <depvar> <predictors> <controls>

[William Lisowski]

The problem is that the estimation is producing weird results, if I simply run a regression and end it with e.g. if country=="USA". (I am running a non-linear regression by the way.

Perhaps instead of concluding that you must drop all the other countries before running your commands, you could instead correct your errors that are leading to "weird results" when you run your estimation using the if qualifier. As it stands, instead of asking us about your fundamental problem, you've guessed at a solution and asked us about it. Your solution may be correct, but some of us will be reluctant to point you in the direction you ask if we're not sure that it's the direction you need to go.

As it stands, though, your problem with the if qualifer isn't clear without more detail, or at a minimum it is too difficult to guess at a good answer from what little you have shared. Please help us help you. Show example data. Show your code. Show us what Stata told you. Tell us what precisely is weird. The Statalist FAQ provides advice on effectively posing your questions, posting data, and sharing Stata output.

[Evan Gensler]

Carlo,

Thanks for the answer. The reason I did not try this is because I am following each regression with a "predict" and while perhaps feasible I do not know how to do it.

My country-level code looks like this

use "panel.dta", clear
keep if country=="United Kingdom"
sort days
nl (testpc = ({b1=0.5}/(1 +{b2=10}*exp(-1*{b3=0.01}*days)))) if days<366
predict testpc_fit

The "testpc_fit" is what I want to graph at the end.

Now if I want to run it the way you are recommending, it would be:

use "panel.dta", clear
bysort country: nl (testpc = ({b1=0.5}/(1 +{b2=10}*exp(-1*{b3=0.01}*days)))) if days<366

How do I go about predicting in a way I can keep track of the country name?


William: My question is the way it is, because I was focused on resolving the "keep" issue which I thought was feasible and would help in the future.
Had my question been about why the regressions are not making sense, I would have included some details about that. But that wasn't my question.
Thanks for your civil comments nevertheless.

[Evan Gensler]

As a followup:

The bysort country: nl (testpc = ({b1=0.5}/(1 +{b2=10}*exp(-1*{b3=0.01}*days)))) if days<366

Also causes issues, in that the regression does not converge. I tried running a linear regression with no issues. If I do each country alone with the same initial values, I have no issues.

[Carlo Lazzaro]

Evan:
if the issue is that the same set of initial values is not "one size fits all" (the countries), probably regressing them one at a time is the only way to go.
That said, as William wisely pointed out, showing your code is always the way to go because:
1) it shows interested listers what the matter is with your data (but this was not your question);
2) can help interested listers to suggest a solution/trick/workaround on the grounds of other threads and/or experienced issues.

[Andrew Musau]

Supposing that you have an issue running the command on the full dataset, appending the results will recreate your dataset with the predictions. The following is not tested and may contain typos or errors.

use "panel.dta", clear
encode country, g(Country)
save "panel.dta", replace
tempfile myfile
qui sum Country
forval i=1/`r(max)'{
    use "panel.dta", clear
    keep if Country==`i'
    sort days
    nl (testpc = ({b1=0.5}/(1 +{b2=10}*exp(-1*{b3=0.01}*days)))) if days<366
    predict testpc_fit
    local next = cond(`i'==1, "save `myfile', replace",  "append using `myfile'")
    `next'
    save `myfile', replace
}
use `myfile', clear

[Evan Gensler]

Thank you both. Andrew, thanks a lot, I will try this solution.