Nick was being a little too polite. The regular lognormal
distribution is not defined by any truncation parameter because it
has a natural threshold at 0. -lognfit- will work well for fitting a
full lognormal distribution in which the parameters (mean and
variance) are functions of company type if that is what interests you.
If your data is truncated in some way, it will be very hard to
establish that it is best fitted by a truncated lognormal rather than
a truncated normal distribution. Given the nature of your problem
you might wish to consider whether frontier models using Stata
commands -frontier- and -xtfrontier- might be more suitable for what
you are trying to do, since these are inherently specified in terms
of truncated distributions.
