log (y + 1) is not quite so ad hoc as this implies.
If values are zero or positive counts, then ln(y + 1) is 0 up.
If values are positive counts, then ln y is 0 up.
Constants between 0 and 1 inclusive won't do this for
you.
On the whole, however, I agree with the thrust of Robert's
argument that there are better ways of coping with zeros.
Nick
[email protected]
Robert Duval
> Which raises then the issue... why adding "+1" to y?... just to be
> able to take logs?
> then why not adding 3/4 or 1/2 or 1/4 for that matter?
>
> The problem with this solution is that, by picking an arbitrary number
> to add, you alter the curvature on your transformed dependent variable
> (y+1) in a pretty ad hoc manner.
>
> A more sensible solution woud be to do the log transformation on y
> whenever y>0, and deal with the zeroes (now missing) directly in your
> econometric method, be it Hecman SS correction, Tobit, or whatever way
> you want to deal with censoring.
> On 12/13/05, Maarten buis <[email protected]> wrote:
> > at Tue 12/13/2005 1:59 PM Victoria Levin wrote:
> > > My original dependent variable was left-censored at zero, but
> > > since I had to take logs, I did a monotonic transformation of
> > > the dependent variable (y+1), and then took logs of that
> > > [ln(y+1)]. So, my current dependent variable has many zeroes;
> > > the rest are positive values.
> > >
> > > 1) Somehow, when I run the full MLE heckman model:
> > >
> > > heckman y x1 x2 x3, select (x1 x2 x3)
> >
> > <snip>
> >
> > > I get the following error:
> > >
> > > Dependent variable never censored due to selection:
> > > model would simplify to OLS regression
> >
> > -heckman- assumes that an observation is censored if y has
> a missing value. You ensured that there
> > are no missing values, but by doing so prevented -heckman-
> from identifying the censored
> > observations. So just taking the log of y would be enough
> in your case.
*
* 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/
