Thanking you for all your suggestions. Since my dependent variable is
certainly a proportion between 0 and 1 (and not a binary with only 0,1
values) now I am faced with three options.
1) fractional logit model as in glm command (Where I am still not sure
which family link combination to be used)
2) mlbeta as suggested by clive
3) betafit as suggested by Nick
Nick, could you please explain the portion "alphavar(varlist1)
betavar(varlist2)" in betafit command? Is there any standard way of
deciding which are the variables to be included in the varlist1 and
varlist2 for alphavar and betavar respectively?
Thanks,
Rijo John.
On Sun, 30 Oct 2005, Nick Cox wrote:
<If that is a possibility, so also is the -betafit-
<of Stephen Jenkins et alius, downloadable from SSC.
<
<Nick
<[email protected]
<
<Clive Nicholas
<
<> Rijo John replied to Nick Cox:
<>
<> > Thanks Nick. I am trying to estimate the impact of spending on a
<> > particular commodity (say X) on the purchase of other goods
<> and services.
<> > So I have the share of, say commodity Y (Its share in
<> household budget)
<> > as the dependent variable and the expenditure on commodity X along
<> with
<> > various control variables for household as explanatory
<> variables. This
<> > can be a standrad OLS estimation. But since the depended
<> variable is a
<> > fraction and is bound between 0 and 1 we go for fractional
<> logit model.
<> >
<> > I hope this clarifies the problem. Given this I would like
<> to know what
<> > would be the right choice of family and link in GLM.
<> >
<> > "family(gaussian) link(logit)" or "family(binomial)
<> link(logit)" or some
<> > other combination?
<> >
<> > in "family(binomial) link(logit)" is it tru that stata
<> takes all values
<> > other than 0 in the dependent variable as 1. If thats the
<> case it does
<> > not take the budget share (absolute amount) information
<> into account. It
<> > will only see if the budget share is absent or not.
<>
<> Since your dependent variable is a fractional, or 'compositional',
<> variable ranging continuously from 0 to 1, another
<> alternative is to use
<> Sean 'Jack' Buckley's -mlbeta-. This allows maximum
<> likelihood estimation
<> with Beta-distributed dependent variables, as well as
<> modelling separate
<> equations for both a mean and dispersion effect. This routine is _not_
<> downloadable from SSC, so type:
<>
<> . net from http://www2.bc.edu/~bucklesj
<>
<> and then click on -mlbeta-. Hope that helps.
***************************************************
Rijo.M.John,Research Scholar
Indira Gandhi Institute of Development Research,
Film City Road, Goregaon East,
Mumbai, India-400065.
*
* 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/
