Re: st: What multiple regression model for extreme distributions

[Michael Norman Mitchell <Michael.Norman.Mitchell@gmail.com>]

Dear Muhammed

 I think that it is possible that this is more of a question of theory 
than a statistical question. The great answers that have been posted 
reflect, as I see it, different theoretical assumptions about the nature 
of the outcome and how the predictors are related to the outcome (of 
saving). I think that these different statistical suggestions each could 
be valid under different theoretical frameworks. Perhaps returning to 
the literature on the nature of "saving" to get a theoretical basis 
would help to inform the statistical model that should be selected. It 
also could be an opportunity to see what statistical models have been 
accepted in publications in the past.

 I know this is more work, but if the aim is publication, it may be 
worthwhile.

Best regards,

Michael N. Mitchell
See the Stata tidbit of the week at...
http://www.MichaelNormanMitchell.com
Visit me on Facebook at...
http://www.facebook.com/MichaelNormanMitchell


muhammed abdul khalid wrote:
> Hi,
> Thank you for the replies.
>
> The data is cross sectional, and saving is simply measured based on
> respondents answer on how much saving they have ( in dollars) with the
> minimum being zero. There is no negative saving. Yes, saving is my
> dependent variable.
>
> I tried logit, zip, zinb, nbreg  but their std error varies greatly.
> Still unsure to what model should be used. My objective is to predict
> the contribution of education, gender, location and  ethnicity to
> saving of the household.
>
> Thank you again for kind response.
>
> Muhammed
> SciencesPo Paris.
>
>
>
>
>
>
> 2010/2/2 Austin Nichols <austinnichols@gmail.com>:
> > You have had a number of good suggestions already, but as Nick Cox
> > points out, the distribution of the dependent variable is not all that
> > relevant to what model you choose; it is the distribution of the
> > dependent variable conditional on explanatory variables that is
> > important.  Before you estimate a two-part "hurdle" or zero-inflated
> > model, I urge you to consider that the right set of explanatory
> > variables might well capture the reason for a large number of zero
> > outcomes (e.g. using -poisson- instead of -zip- etc.).  When it comes
> > to household saving (I think that is your dependent variable, not
> > independent), you also want to consider debt.  It may be the case that
> > households you are coding as zeros actually have negative saving
> > during the period under study.  Do you have panel data, or
> > cross-sectional data?  How is saving measured?
> >
> > On Tue, Feb 2, 2010 at 10:09 AM, <muhammed.abdulkhalid@gmail.com> wrote:
> > > I have a household income survey data ( 38,000 observations), and my
> > > problem is doing a multiple regression on saving ( independent var) to
> > > ethnicity/strata/employment
> > > etc( dependent var).
> > >
> > > The problem is this : 70% of my observation for the value of saving is
> > > zero. I had recode it to 1 and log them, but the distribution is still
> > > extremely skewed ( mean 0.78, std dev is 2.4  min 0 max 14). The
> > > historgam still looks like the letter L , exteremly skewed to the
> > > right with  long tail.  Obviously, OLS is out, and I tried Poisson(
> > > glm nbinomial) but the distribution is still not distributed normally.
> > > The data are in order i.e no missing values etc etc. It is clean.For
> > > some reason, lobit would not run.
> > *
> > *   For searches and help try:
> > *   http://www.stata.com/help.cgi?search
> > *   http://www.stata.com/support/statalist/faq
> > *   http://www.ats.ucla.edu/stat/stata/
*
*   For searches and help try:
*   http://www.stata.com/help.cgi?search
*   http://www.stata.com/support/statalist/faq
*   http://www.ats.ucla.edu/stat/stata/