Generating variables based conditionally with log of zero-value variable

[Phuc Nguyen]

Hi experts,
Today I try to find a way to generate a variable called logTAT1.

Normally, logTAT1= ln(TAT)
When the value of TAT-0, I will use the min value of TAT instead

So, my code is:

      gen logTAT1=.
      replace logTAT1 = ln(TAT)
      replace logTAT1 = ln(min(TAT)) if TAT=0

However, it resutls in an error.

invalid syntax
r(198);

end of do-file

r(198);

Could you please help me to sort it out? Thank you.

[Carlo Lazzaro]

Phuc:

 
 replace logTAT1 = ln(min(TAT)) if TAT==0

will probably do the trick.

[Phuc Nguyen]

Phuc:

replace logTAT1 = ln(min(TAT)) if TAT==0

will probably do the trick.

Thanks Carlo Lazzaro , I followed your guide but there are problems happened

[Nick Cox]

The function min() does not work like that. It works row-wise over two or more arguments, not column-wise over the elements of a variable or vector or matrix.

But even if min() did work like that the result would still be zero whenever zeros are present. Your code does not ignore the zeros.

What you trying to do, I think is

su TAT if TAT > 0

gen wanted = cond(TAT == 0, log(r(min)), log(TAT))

but that is still a bad idea without substantive grounds. What some people do is use log(half smallest positive value) if a value is zero, but that is a fudge too.

[Phuc Nguyen]

Hi Nick Cox ,

Thank you for your help, the idea coming from a paper from a good journal (Kolias,2011). In his paper, he use the smallest value replacing zero when getting log

Best regards,
Phuc

[Nick Cox]

I have no cats. So, if your desired variable is log of number of cats possessed, you are going to assume that I have 1 cat.. No thanks!

More importantly, there are many better ideas. See e.g.. the concurrent thread https://www.statalist.org/forums/for...-be-dealt-with

[Phuc Nguyen]

I have no cats. So, if your desired variable is log of number of cats possessed, you are going to assume that I have 1 cat.. No thanks!

More importantly, there are many better ideas. See e.g.. the concurrent thread https://www.statalist.org/forums/for...-be-dealt-with

Thank you Mr.Cox. I understand your example and had a look at the discussion link. I also saw the approach using ln(1+x) has been done in various papers in finance major, especially when calculating non-negative variable e.g., firm age