Categorical data coding

[Margo Long]

I have a variable 'fav_color' with categorical data. Is there a STATA command to convert them to factors so blue will instead be 0, green will be 1, red will be 2 and yellow will be 3 so a regression will be easier to interpret (the same for fav_food)?

fav_color fav_food
blue apples
green soup
red steak
yellow fish
no answer left blank

I tried these commands but STATA prints an error message:

encode fav_color, generate(f_color)
error: not possible with numeric variable

destring fav_color, generate(f_color)
f_color already numeric; no generate

This works but it changes fav_color variable to a string which I do not want.
decode fav_color, gen(f_color)

Thanks

[Ali Atia]

Stata is telling you that the variable is already numeric. You may be seeing what appear to be strings in the data editor because the variable has a value label applied. However, that will not affect any analysis you conduct on the underlying numeric data (e.g., a regression).

If you really want to remove the value label (though I see no reason for doing so, since value labels have no impact on the results of your analysis and are generally a very helpful way of identifying what categorical indicators represent), you can use:

label drop _all

[Margo Long]

That makes sense, I can keep the labels, but I have another question.

If I have these observations:

ID fav_color fav_food
1 blue soup
2 green not important
3 blue steak
4 no answer apples
5 yellow soup

I want to eliminate observations where fav_color is no answer or where fav_food is not important. I tried these commands but all of them printed 'type mismatch'
drop if strmatch(fav_color, "*no answer*")
drop if strmatch(fav_color, "no answer")
drop if strmatch(fav_color, 'no answer')

How can I eliminate them from my data?

Thanks

[Nick Cox]

It's the same answer as #2. These variables are numeric; they have value labels but that is cosmetic. There is a way to select observations according to value label, but it is simpler just to work with numeric values:

drop if fav_color == 4 | fav_food == 2 

But I would go

gen OK = fav_color != 4 & fav_food != 2 

and then carry our analyses with a qualifier if OK because you might mess up the drop or change your mind.

[Margo Long]

Thank you both for the information, it worked!