How come label changes aren't registered by Stata

[Niels Henrik Bruun]

Hi
I just discovered that changes in labels is not captured by either c(changed) or describe (In the manual they are said work similarly) as seen by:

cls
sysuse auto, clear
label variable rep78 "Change label"
describe
display "`c(changed)'"

sysuse auto, clear
gen x = 1
label variable rep78 "Change label"
describe
display "`c(changed)'"

Is there a way to detect all changes in a dataset?

Looking forward to hear from you

[daniel klein]

Is there a way to detect all changes in a dataset?

I am not aware of anything beyond cf, compare, dta_equal, etc., all of which are concerned with values in variables.

But then, I find it hard to define what all changes would include. Does this include changes to the order of variables? Does it include changes to variable's display format? Does it include changes to characteristics and notes?

Best
Daniel

[Niels Henrik Bruun]

Dear Daniel
Thank you for your answer.

Yes, I would have expected changes in order and label to so important that I should be able to detect if such changes has happened.
And I would have liked to be warned if all my labelling work was about to disappear without warnings when importing another dataset.

But that is just me.

Kind regards nhb

[daniel klein]

Niels

you are not the only one wondering about what exactly is flagged as a change to the dataset in c(changed). A while ago, I was surprised that neither reshaping the dataset nor dropping all variables would be flagged as changes to the dataset. Try

webuse judges
reshape wide rating , i(target) j(judge)
display c(changed)
drop _all
display c(changed)

I think that StataCorp should include a technical note in the documentation that clearly describes what constitutes a change to the data in memory and what does not.

Best
Daniel

[Niels Henrik Bruun]

Dear Daniel
I totally agree

[wbuchanan]

I believe reshape adds a characteristic to the data set to indicate that reshape was used. That said, if the concern is limited to the domain of a data set on disk, then a version control program could indicate if there were any changes. Displaying the results in a useful way would probably be a bit more changing since the change detection would rely on a binary diff.

[Bjarte Aagnes]

-xmlsave- still working, but no longer an official part of Stata, may be used to save text files with metadata only to be compared with a diff tool, (or parsed).

sysuse auto
xmlsave A.xml in 1 if _n < 1 , doctype(dta) legible replace /* dtd */
lab var make "make mistake"
xmlsave B.xml in 1 if _n < 1 , doctype(dta) legible replace /* dtd */  
* then compare A.xml with B.xml using your favorite tool

* Simple use, but not covering value label values and characteristics
* using possibly a second Stata session

xmluse B.xml , clear
describe , replace
save B , replace
xmluse A.xml , clear
describe , replace
cf * using B , all verbose

. cf * using B , all verbose
        position:  match
            name:  match
            type:  match
       isnumeric:  match
          format:  match
          vallab:  match
          varlab:  1 mismatch
                   obs  1. Make and Model in master; make mistake in using
r(9);

[Niels Henrik Bruun]

Dear Bjarte
Thank you very much for your suggestion. I'll look into it.
Sorry, that I didn't answer before - Holiday