Selecting variables with a target number of levels (categories)

[Sonnen Blume]

Hi,

This might be a rather unhappy question for the happy new year...

For better handling a large number of variables, I'd like to know the variables that are dichotomous, trichotomous and so on. For instance, I want to tabulate only the variables with 5 levels (very happy, happy, neutral, unhappy, very unhappy), so it'd be easier if they could identify all at once.

Thank you in advance!


Here is an example data:

clear
input float(v1 v2 v4 v5 v6 v7)
4 45 2 1 12 1
3 24 4 2 21 1
2 29 3 2 21 1
3 43 4 1 21 1
1 24 1 2 21 1
2 44 1 2 21 1
4 22 4 2 97 1
5 18 8 2 21 1
4 21 1 2 21 1
5 37 8 1 21 1
5 24 6 1 97 1
1 33 7 1 21 1
3 30 6 1 21 1
3 27 4 2 21 2
2 46 5 2 21 1
5 31 2 1 11 1
5 22 7 2 21 1
3 23 2 1 97 1
1 18 1 2 21 1
2 38 7 2 21 1
2 30 5 2 21 1
4 39 4 1 21 1
3 47 2 2 21 1
2 21 2 2 21 1
4 38 4 2 21 1
3 24 7 1 21 1
2 25 3 1 21 1
2 18 8 2 21 1
2 48 6 2 21 1
1 32 2 1 21 1
3 32 4 2 21 1
5 19 2 1 21 1
5 21 4 2 21 2
3 24 7 2 21 1
3 31 1 2 21 1
3 26 7 2 97 1
3 34 4 2 51 2
2 24 4 2 21 1
4 17 3 2 21 1
4 24 8 1 97 1
1 43 1 2 21 1
5 24 7 1 21 1
3 34 1 2 21 1
3 19 6 2 21 1
5 28 2 1 71 2
1 34 2 2 21 1
4 32 2 2 21 1
2 15 8 2 97 1
3 46 7 1 21 1
4 47 3 2 21 2
3 49 6 2 21 1
3 45 5 2 21 1
5 38 1 1 21 1
3 16 6 2 21 1
3 24 2 2 21 1
3 17 7 1 21 1
2 38 4 2 21 1
5 49 8 1 14 1
4 36 2 1 21 1
5 33 7 1 21 1
5 34 4 2 21 1
4 30 6 1 21 2
4 30 6 1 21 1
3 18 7 2 21 1
5 24 6 1 21 1
3 23 3 2 21 1
1 29 5 2 21 1
3 22 6 1 21 2
2 31 5 2 21 1
3 40 4 2 21 1
5 36 3 1 11 1
2 32 7 2 21 1
3 35 2 1 21 1
2 22 6 1 21 1
5 22 3 1 21 1
2 26 1 2 21 1
3 31 5 2 21 1
1 24 7 2 21 2
3 43 3 2 21 1
3 28 2 2 21 1
3 30 3 2 21 1
4 48 1 2 21 1
3 42 7 2 21 1
3 40 2 2 21 1
4 35 2 2 21 1
2 40 6 2 21 1
2 26 2 1 21 1
3 34 6 1 21 1
3 33 6 1 21 1
3 34 7 1 21 1
3 32 6 1 21 1
1 19 5 2 97 1
2 34 8 2 43 1
4 43 7 1 21 1
1 25 5 2 21 1
2 25 4 2 21 1
3 38 5 2 21 1
3 21 2 2 21 1
2 37 5 2 21 1
4 21 6 1 21 1
end
[/CODE]

[Clyde Schechter]

local desired_levels 2 // OR HOWEVER MANY LEVELS YOU WANT

gen long obs_no = _n  // TO NOTE ORIGINAL DATA ORDER

local vbles_with_desired_levels
foreach v of varlist _all {
    sort `v'
    gen long n_levels = sum(`v' != `v'[_n-1])
    if n_levels[_N] == `desired_levels' {
        local vbles_with_desired_levels `vbles_with_desired_levels' `v'
    }
    drop n_levels
}

sort obs_no // TO RESTORE ORIGINAL DATA ORDER

display "`vbles_with_desired_levels'"

[Nick Cox]

See also distinct from the Stata Journal which can give you a single sorted table.

[Nick Cox]

To the point of #3 here is a run on the data example in #1.

. distinct, min(5) max(5)

---------------------------
    |     total   distinct
----+----------------------
 v1 |       100          5
---------------------------

distinct is profoundly empirical. It can't identify variables where 5 distinct values are possible but do not all occur.


But if you have 5 distinct value labels defined, see findname from the Stata Journal as updated in 2020.

dm0048_4: Finding variables. N. J. Cox. Stata Journal 15: 605; 12: 167; 10: 691; 10: 281–296.

New options include columns() to find variables according to column position in the dataset, so, for example, columns(1 -1) finds the first and last variables in the dataset; and four new options to find variables with specified text in value labels, or specified numbers of value labels, either defined as value labels or used within the data. Options to do with value labels used within the data can be combined with if or in, or both.

[Sonnen Blume]

local desired_levels 2 // OR HOWEVER MANY LEVELS YOU WANT

gen long obs_no = _n // TO NOTE ORIGINAL DATA ORDER

local vbles_with_desired_levels
foreach v of varlist _all {
sort `v'
gen long n_levels = sum(`v' != `v'[_n-1])
if n_levels[_N] == `desired_levels' {
local vbles_with_desired_levels `vbles_with_desired_levels' `v'
}
drop n_levels
}

sort obs_no // TO RESTORE ORIGINAL DATA ORDER

display "`vbles_with_desired_levels'"

Thanks so much Clyde! Mighty solutions to might problems, as always! Happy new year!

[Sonnen Blume]

To the point of #3 here is a run on the data example in #1.

. distinct, min(5) max(5)

---------------------------
| total distinct
----+----------------------
v1 | 100 5
---------------------------

distinct is profoundly empirical. It can't identify variables where 5 distinct values are possible but do not all occur.


But if you have 5 distinct value labels defined, see findname from the Stata Journal as updated in 2020.

Thanks so much, professor, for this wonderful package! It gives a perfect view of what I was looking for.