I am trying (without so much luck) to explain to a reviewer how I am setting
up my logistic regression model.
Briefly, the dependent variable, first_trop, is a continuous variable, and
about half the values are zero.
What I want to do is replace the univariate model with a bivariate one which
uses one term for values of first_trop>0 and a dummy variable for
first_trop=0. The tricky part is, the best transformation for the
continuous part is a logarithmic transformation. I had done it a slightly
different way, but -mfp- does the same thing:
. mfp logistic is_dead first_trop , catzero(first_trop)
and -mfp- does the following transformation:
-> gen double Ifirs__1 = ln(X)+3.42714503 if e(sample)
(where: X = first_trop/10)
-> gen byte Ifirs__0 = first_trop<=0 if e(sample)
The question is: If I do the same thing manually, ln(X) is missing if
first_trop=0, and a missing number plus a constant is still missing. But
-mfp- manages to do it without dropping the observations where first_trop==0
Can anyone help explain how this is happens, or how I can justify it?
Many thanks.
*
* For searches and help try:
* http://www.stata.com/support/faqs/res/findit.html
* http://www.stata.com/support/statalist/faq
* http://www.ats.ucla.edu/stat/stata/