Richard--
[I apologize in advance for the slim Stata content]
The discussion in
http://www.stata.com/statalist/archive/2007-04/msg00549.html
is on point--note debt can be zero or negative (financial assets are
negative debts). If you only look at the size of debt for those who
have positive debts, you are looking at a weird subsample. Think of a
person who has a bank account with $6K and student loan debt of $5K
--their debt is $5000 even though their net worth is $1000 (they are
lending the bank $6K). They can freely choose any debt amount between
0 and $5K with almost no real consequence aside from liquidity--the
observed value may be no more than an accident of timing in
check-writing rather than an outcome of consequence to be modeled.
The bottom line is, there is no real point in looking at debt in
isolation unless you've got a specific theoretical model which
absolutely requires it, since it has not much meaning without looking
at the rest of the balance sheet. Even in the case where assets should
be ignored, I can't think of a model where debts of zero should be
dropped, rather than modeled with e.g. -poisson- (one complication is
that you often want a log-log specification, to model elasticities, so
instead of -reg lny lnx- you want -poisson y lnx- but if x also can be
zero, you've got trouble--Nick Cox mentions this in the -transint-
help file from SSC, I believe).
For some models of debt, you might want -zip- to model the "access to
credit markets" process with the -inflate- option. In an interesting
experiment on savings incentives
(http://www.brookings.edu/views/papers/gale/20060711.htm), a large
portion of the treatment group got denied accounts by the bank--these
were savings accounts! Yet there is some substantial fraction of the
population that has almost no access to lending/borrowing markets.
On 7/27/07, Richard Williams <[email protected]> wrote:
> At 10:34 AM 7/27/2007, Austin Nichols wrote:
> >Vitor <[email protected]>:
> >As I have pointed out before on this list, e.g.
> >http://www.stata.com/statalist/archive/2007-04/msg00549.html
> >http://www.stata.com/statalist/archive/2006-12/msg00466.html
> >-tobit- is inappropriate if the zeros are not censored values
> >(representing a negative y* that is observed as y=0). If the zeros
> >are simply a point mass in the distribution of a nonnegative dep var,
> >then -poisson- or -glm- are better options. In your case, the fact
>
> Let me follow up with a related question for Austin or anyone
> else: A student of mine did a paper where she examined debt. She
> dropped all cases where the person had no debt. I told her that was
> a bad idea, but wasn't sure what a good idea was. I was thinking
> tobit or Heckman, but based on what Austin is saying here, is
> -poisson- or -glm- the way to go?
*
* 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/
