Define a categorical variable.

[River Huang]

Dear All, According to the following data (this example is taken from http://bbs.pinggu.org/thread-6277610-1-1.html), I'd like to create a categorical variable d, which is equal to 1 if b2a=rmax, 2 if b2b=rmax, 3 if b2c=rmax, and 4 if b2d=rmax.

* Example generated by -dataex-. To install: ssc install dataex
clear
input byte(b2a b2b b2c b2d) float rmax
100  0  0  0 100
  0 80 20  0  80
  0  0 60 40  60
 90  0 10  0  90
  0 10 20 70  70
end
label values b2a B2A
label values b2b B2B
label values b2c B2C
label values b2d B2D

The following is my code:

gen d = (b2a == rmax)
replace d = 2 if b2b == rmax
replace d = 3 if b2c == rmax
replace d = 4 if b2d == rmax

Any other suggestion?

[Nick Cox]

What happens if two or more conditions are satisfied?

[River Huang]

Hi, Nick, I know your concern, but there is no such situation in the data set.

[Maarten Buis]

Experience tells me that in a real dataset anything can happen, also things that should not be possible. So robust code should at least check and return an error message when the impossible happens.

[River Huang]

Dear Maarten, I can not agree more. My suggested procedure has been tested in a real data set (the percentage of shares held by different ownerships) by the person who asked this question. But I share with your concern in other (more general) situations.

[Romalpa Akzo]

What make you still unsatisfied with your code? If it is about 1-line coding, the following might have some small sense.

1. If it could be sure that there is only one rmax among b2* in each observation:

gen d1=1*(b2a==rmax)+ 2*(b2b==rmax)+ 3*(b2c==rmax)+4*(b2d==rmax)

2. If not:

gen d2=1*(b2a==rmax)*(b2b<rmax)*(b2c<rmax)*(b2d<rmax)+2*(b2b==rmax)*(b2c<rmax)*(b2d<rmax)+3*(b2c==rmax)*(b2d<rmax)+4*(b2d==rmax)

[River Huang]

Hi Romalpa, Yes, I'm looking for a more concise code, if any, and many thanks for your suggestions.