Standardizing var gives multiple 0's and 1's instead of a continuous var

[Hannah Pareik]

Hi everyone,

I am cleaning survey data from the Asianbarometer and want to standardize some variables to range from 0-1.
This is the code I usually use, in this case for the variable executive approval, that ranges from 1-4. I want it to range from 0-1. The variable educ ranges from 1-9. I also want it to range from 0 to 1.

sum execapproval
replace execapproval = (execapproval - `r(min)') / (`r(max)'-`r(min)')

Here is the data:

* Example generated by -dataex-. For more info, type help dataex
clear
input byte(educ execapproval)
7 2
8 3
5 3
8 2
7 3
7 3
5 3
3 1
9 .
8 1
7 3
6 2
6 3
5 2
6 2
7 1
4 2
7 2
8 3
5 3
7 3
1 .
1 3
3 4
1 .
8 3
end
label values educ labels10
label def labels10 1 "No formal", modify
label def labels10 3 "Complete primary /elementary", modify
label def labels10 4 "Incomplete secondary/high", modify
label def labels10 5 "Complete secondary/high", modify
label def labels10 6 "Incomplete secondary/high school", modify
label def labels10 7 "Complete secondary/high school", modify
label def labels10 8 "Some university/college-level, with diploma", modify
label def labels10 9 "With University/College degree", modify
label values execapproval labels56
label def labels56 1 "Not at all satisfied", modify
label def labels56 2 "Not very satisfied", modify
label def labels56 3 "Fairly satisfied", modify
label def labels56 4 "Very satisfied", modify

for some reason the code neither works for educ nor execapproval, something seems to be up with the labels but i cannot figure out what it is. It works if I create a new variable, for example:

gen educn = educ
sum educn
replace educn = (educn - `r(min)') / (`r(max)'-`r(min)')
label var educn "Education - normalized"

But not for educ.

Can someone help?

Thank you so much!!

Best,
Hannah

[Clyde Schechter]

You start out with your variables having value labels attached to them. When you rescale them, most of the resulting values are not integers. And the ones that are, namely 0 and 1, do not correspond correctly to the labels. (0 isn't part of either label. And 1 now represents 4 or 9 in the original variable and is labeled accordingly--which is wrong, of course.) The problem goes away when you remove the value label.

sum execapproval
replace execapproval = (execapproval - `r(min)') / (`r(max)'-`r(min)')
list, clean

label values execapproval    // REMOVE THE VALUE LABEL
list, clean

The problem you are having is merely an illusion. Internally, Stata has the correctly calculated numbers--but because you haven't removed the value labels, it is displaying them in a misleading way. Even if you left the labels on, which, to be clear, I don't recommend, all calculations will be done correctly with the new values--the labels just mislead your eyes into seeing something different from what Stata is doing.

[Hannah Pareik]

Hi Clyde,
thank you so much for your quick help on this matter!
I have tried to run your code, unfortunately I still only get multiple 0's and 1's.
Do i need to modifiy the code somehow? I also tried to the code "label val varname" before running sum execapproval...

Thank you!
Kind regards,
Hannah

[Nick Cox]

I can't reproduce the problem reported. Here is a minimal test.

* Example generated by -dataex-. For more info, type help dataex
clear
input byte educ
1
5
9
end

su educ
gen educn = (educ - r(min)) / (r(max) - r(min))
label var educn "Education - normalized"

list

Values are returned as 0, 0.5, 1, as they should be.


The last few lines of code are just a slightly shorter variant of yours. Your code gives the same result.

[Clyde Schechter]

I wonder if the display format is somehow messed up. Try running your code again, getting the 0's and 1's. Then run -format execapproval educ %3.2f- and -list- or -browse- the data. See if that shows the numbers correctly.