Creating composite variable leveraging three existing variables/timepoints

[Leigh Forgues]

Hello,

I have a dataset with three variables/timepoints (as below). I need to create a composite variable, Midlife Music Play (midmusic), based on the following:

Midlife Music Play:
---Continuous Play (value==1) is instrument engagement at all follow-up time points;
---Intermittent Play (value==2) is play at one point or another, A or B or C (but not continuously); and
---Never Play (value==3) is no play at any of the three timepoints
---Unknown (value==.)

Note: I cannot use dataex because my data is stored on a confidential server; there is genetic data attached to my analysis.



Variable A (2003): During the past year, how many hours per month did you play a musical instrument?

. = missing
-3 = refused
-1 = don't know
0-200


Variable B (1994): How often did you play a musical instrument about 10 years ago?

. = missing
-3 = refused
1 = Often
2 = Rarely
3 = Never


Variable C (1974): How often did you play a musical instrument when you were about 35?

. = missing
-3 = refused
1 = Often
2 = Rarely
3 = Never

I'm struggling with the most efficient way to get arrive at what I need. I was able to brute force it, but I'd like to understand a more elegant approach.

Thank you!

[Andrew Musau]

You can set up your conditions and use the -cond()- function. Do read https://journals.sagepub.com/doi/10....867X0500500310. Three conditions are limiting:

---Continuous Play (value==1) is instrument engagement at all follow-up time points;
---Never Play (value==3) is no play at any of the three timepoints

and

---Unknown (value==.)

You can thus concentrate on these.

gen wanted = ///
    cond(inrange(A, 1, 200) & B==1 & C==1, 1, ///
        cond(A==0 & B==3 & C==3, 3, ///
            cond(!inrange(A, 0, 200) & !inrange(B, 1, 3) & !inrange(C, 1, 3), .a, 2)))

label values wanted wanted
lab def  wanted 1 "Continuous Play" 2 "Intermittent Play" 3 "Never Play"  .a "Unknown", modify

The above assumes that there are no positive nonmissing values outside the range [0, 200] for variable "A". You need to verify this. Also, you can change the number of hours of play for "continuous" to a higher value. I start at 1 hour.

[Leigh Forgues]

Thank you so much! My numbers are off slightly and I'm trying to understand why that is, but the code is super helpful.

Continuous and Never Play are perfect. Interestingly, ~200 extra observations are categorized as Intermittent than I was expecting. I will continue to explore - it has to be in the values somewhere.

[Andrew Musau]

#2 defines conditions for "Continuous", "Never Play" and "Unknown". Any other cases are categorized as "Intermittent".

Interestingly, ~200 extra observations are categorized as Intermittent than I was expecting.

Let us see a few of these. Since you expect them to be categorized differently, you can explain your expectation and it may be possible to revise some of the conditions subsequently.