Esttab Using Loop Not Working Properly for Categorical Variables in Logit Models

[Kirk Harvey]

Hi,

I read posts related to the esttab carefully, but I did not find the exact answer to my question.

I am running a simple loop on many countries with a logit model that uses the categorical variables country of birth, age group, education, and their interactions. I am interested in the odds ratios for birth-related variables across different countries. Therefore, I use esttab to obtain these odds ratios

The categorical variable country of birth is not balanced across countries, resulting in varying numbers of categories. Consequently, esttab returns a table that uses only the labels for the categorical variable birth from the last model/country. Is there a simple way to prevent this and use the correct labels for the variable?

Loop:

cd "....."
local files: dir . files "*dta"

foreach file of local files {
    use "`file'", clear
    quietly: logit employ  i.birth i.edu  i.age i.birth##i.edu, or noconstant baselevels nolog
    eststo
}
esttab, label eform nostar keep (*birth *birth#*edu)

[Andrew Musau]

estout is from SSC (FAQ Advice #12). The labels must come from somewhere. It appears you are not running the regressions using a common dataset. Therefore, you should load the dataset with no missing countries before running the final esttab command. Secondly, ensure that the labels are consistently defined across the datasets. If you encoded the country variable independently in each dataset, you risk having mismatches since some countries may be missing in some datasets.

cd "....."
estimates clear
local files: dir . files "*dta"

foreach file of local files {
    use "`file'", clear
    quietly: logit employ  i.birth i.edu  i.age i.birth##i.edu, or noconstant baselevels nolog
    eststo
}
use dataset_with_all_countries_labeled, clear
esttab est*, label eform nostar keep (*birth *birth#*edu)

[Andrew Musau]

ensure that the labels are consistently defined across the datasets

use dataset_with_all_countries_labeled, clear
label save `:val lab birth' using mylabels

Then open mylabels.do saved in your current directory, get the labels and relabel the birth variable within the loop.

[Kirk Harvey]

Andrew, thank you very much for your help. However, I believe my problem arises from a different source.

The issue is that the categorical variable -birth- (representing regions of birth) covers different regions in various countries. For example, there are 13 regions in country A but 18 regions in country B.

I assume my problem is somewhat similar to the problem mentioned here: https://www.statalist.org/forums/forum/general-stata-discussion/general/1692-preventing-estout-from-overwriting-e-labels-in-successive-runs , but I have no idea how to approach that in my case.

When I run the loop, the -Esttab- output uses labels from the last dataset, and I do not know how to prevent this. One solution is to append datasets into one file, but the file is too large to handle. Given that no dataset has all labels for categorical variable birth, I can not use your "use dataset_with_all_countries_labeled, clear command".

[Andrew Musau]

Do you need to report the coefficients on the region dummies or are these just control variables? If not, don't include them in the tables and just indicate their presence. Otherwise, it is difficult to imagine how a table reporting regional dummies from different countries (based on the model) would look like.

For example, there are 13 regions in country A but 18 regions in country B.

It is possible to write code that will handle this, but first let me know what kind of table you want to produce. If you can post an example of such a table, that would be helpful.

[George Ford]

This worked for me.

clear all

sysuse auto, clear
drop if foreign
save auto_nf, replace

sysuse auto, clear
keep if foreign
save auto_f, replace

foreach file in auto_nf auto_f {
    use "`file'", clear
    qui reg mpg weight i.rep78 i.rep78#c.turn
    eststo
}

esttab, label

Have you tried it without the "keep" option? I wonder if something funky is going on there.