Goodness knows. There are many recipes, all of them problematic. "Go out
and measure it again" is the best advice! That divides sciences. In some
it is regarded as self-evident, in others as possibly a strange joke.
I just want to point out that far being conservative, using a very small
positive value rather than zero creates outliers in terms of log of the
variable in question. Using half the smallest non-zero value observed is
another fudge that is not quite so objectionable.
For example, consider 0, 1, 2 fudged to
(1) 2^-10, 1, 2
(2) 0.5, 1, 2.
Using log base 2 to convenience -- clearly the same principle applies to
any other base -- you can see that logged these come to
(1) -10, 0, 1
(2) -1, 0, 1
So far from being minute, any smidgen is a big deal on a log scale.
Also in the "yes, I thought of that already" category is advice to try
different solutions and see how much difference they make.
Viktor Slavtchev
I want to estimate production function in logs in order to interpret the
coefficients in terms of elasticities. one of the RHS variables has a
lot of zero value. in such cases, some people usually add a very small
value in order to avoid the problem of 'undefined' log of zero. a
similar issue was discussed some time ago among other things and in a
quite different context but an interesting alternative suggestion was
made (which I can't remember anymore). unfortunately, I was not able to
find this particular thread.
in that sence, any suggestions how to deal with the problem would be
greatly appreciated
*
* 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/
