How to make this correspondence dta?

[Chingyuan Kao]

Original data has a string variable "country"

country
US
US
UK
US
Canada
UK
US
Canada

To reduce file size, I want to turn this into a numeric variable, something like
country
1
1
2
1
3
2
1
3
while also separately saving a file correspondence.dta that looks like
numeric country
1 US
2 UK
3 Canada

Changing it to numeric to reduce the file size could be done by doing this.
encode country, gen(country2)
drop country
gen country3=country2
drop country2
rename country3 country

But since I am using country3, but not country2 (which has string label), this data now do not contain string information that I need although file size has been successfully reduced.
I need that correspondence.dta file.
How can I generate that while also simultaneously reducing file size like this?

[Leonardo Guizzetti]

Welcome to the Statalist. You may want to reconsider using label, rather than a "look-up" dataset (or translation table). If you do need this table, it's not clear why you need it from just your description of your problem, but then again, you probably don't need it.

First off, Stata allows up to 65,536 unique label value levels, which can have a maximum length of 80 characters. If your data are within these limits (such as country names in your toy data example), then labels are the way to go.

clear
input str6(country_name)
US
US
UK
US
Canada
UK
US
Canada
end
list

encode country_name, gen(country)
compress country
label list country

This creates a new variable called -country- taking integer values from 1 to 3, which have the following label attached.

. label list country
country:
           1 Canada
           2 UK
           3 US

You could also supply your own label if you cared about which values get mapped to which country.

From here you can generate a dataset for the label mapping of -country- which can serve as your translation table.

uselabel country, clear
list

Result

     +----------------------------------+
     |   lname   value    label   trunc |
     |----------------------------------|
  1. | country       1   Canada       0 |
  2. | country       2       UK       0 |
  3. | country       3       US       0 |
     +----------------------------------+

If your data fall outside of the limits of what a label could hold, then the most direct way may be to use -egen, tag()-.

egen byte unique = tag(country_name)
bysort unique (country_name) : gen country = sum(unique)
drop unique
keep if country
qui compress
list, abbrev(12)

Result

     +------------------------+
     | country_name   country |
     |------------------------|
  1. |       Canada         1 |
  2. |           UK         2 |
  3. |           US         3 |
     +------------------------+

[Chingyuan Kao]

Thanks for your reply, and I wonder how can I save the table below to another dta file?

[Leonardo Guizzetti]

Read the next sentence....

[Bjarte Aagnes]

First off, Stata allows up to 65,536 unique label value levels, which can have a maximum length of 80 characters.

label define creates a value label named lblname, which is a set of individual numeric values
and their corresponding labels. lblname can contain up to 65,536 individual labels; each individual
label can be up to 32,000 characters long.

[Leonardo Guizzetti]

label define creates a value label named lblname, which is a set of individual numeric values
and their corresponding labels. lblname can contain up to 65,536 individual labels; each individual
label can be up to 32,000 characters long.

Thanks for the correction, Bjarte. I misread the label length. Indeed 32,000 characters expands the utility of labels even more.

[Clyde Schechter]

Indeed 32,000 characters expands the utility of labels even more.

Well, maybe. Doesn't the argument against Stata allowing variable names to be longer than 32 characters based on the difficulty/impossibility of displaying them in output apply equally well to value labels? And that argument has been raised against requests to go to 64 characters. It would be much more problematic with 32,000. Is there some use of value labels that I'm just not aware of which makes extremely long value labels more useful?

[Leonardo Guizzetti]

Well, maybe. Doesn't the argument against Stata allowing variable names to be longer than 32 characters based on the difficulty/impossibility of displaying them in output apply equally well to value labels? And that argument has been raised against requests to go to 64 characters. It would be much more problematic with 32,000. Is there some use of value labels that I'm just not aware of which makes extremely long value labels more useful?

I agree with you, Clyde, and my remarks in #6 were restricted to what I believe to be the use case in this thread. Specifically, I understood the use case to be (1) create a dictionary/look-up dataset that (2) uses minimal disk space. I don't believe that long labels or variable names are terribly useful for any display purposes, in general. Here, OP does not say that they want to display this information, only to store it. I thought to use labels as a concise way of doing this that wouldn't later require some use of -merge- (or the like) to recover the encoded data.

[Nick Cox]

Cross-posted at https://stackoverflow.com/questions/...-following-dta

Please note our longstanding request to tell people about cross-posting elsewhere.