Stephen,
Thank you for your explanations. My understanding now is:
1. STATA's ginides command assumes that the all your obs are in 1 population. So if I wanted to treat each county as a separate population and calculate gini indices for each tract within a county, I would have to qualify the command with an "if county" or "by county" statement. (I tried this and STATA is giving me an error). Is there another way to do this besides using subsets of your data and then appending them?
2. Your code treats each tract as a population. Thus, the income distribution of one tract will not influence the gini of another tract.
3. your ineqdeco command would give the same results as the ginidesc command
Is this right? Or am I confused.
Thanks,
phoenix
> -----Original Message-----
> From: [email protected]
> [mailto:[email protected]]On Behalf Of Jenkins S P
> Sent: Sunday, March 07, 2004 5:56 AM
> To: [email protected]
> Subject: Re: st: gini coeff
>
>
> On Sat, 6 Mar 2004, Do, Phoenix wrote:
>
> > I am calculating a gini coeff for each tract in my census
> dataset. The
> > observations are at the blockgroup level and I have 3 counties.
> >
> > This is my command:
> >
> > ginidesc percapi, bygroup(tract)
> >
> >
> > However, when I use only a subset of my data, some of the ginicoeff
> > change. I thought the ginicoeffs were not influenced by the
> composition
> > of other tracts. In other words, the ginicoeff of tractA
> depends only on
> > the income distribution of the blockgroups within tractA and is
> > independent of the income distributions of other tracts. I
> am finding
> > this to not be the case.
> >
>
> The Gini coefficient depends on the (normalised) ranking by income of
> each observation. An observation's ranking in the population (which
> depends on everyone from all the groups) differs from an individual's
> ranking within a given group. This is also why the Gini
> coefficient does
> not neatly decompose into a sum of Within-Group +
> Between-Group Inequality
> (cf. the generalised entropy indices) -- there is an additional term
> summarising overlapping between groups. (Put another way, the
> issue you
> draw attention to, is precisely why the majority of analysts
> do not rely
> on the Gini coefficient for inequality decomposition by population
> subgroups.)
>
> If, however, you are simply interested in simply getting a
> Gini for each
> subgroup (tract) and storing it, then it is straightforward to do with
> several programs, including my -ineqdeco- (or -ineqdec0-):
>
> levels tract, local(reg)
> ge gini_reg = .
>
> foreach r of local reg {
> ineqdeco percapi if tract == `r'
> replace gini_reg = $S_gini if tract == `r'
> }
>
> -ineqdeco income, by(region)- would produce output showing inequality
> indices for each region, but the results by region would not be saved.
> It would however, show the Within/Between decomposition for a range of
> indices, though not for the Gini (for the reasons given
> above). -ginidesc-
> provides the special Within/Between/Overlapping decomposition for the
> Gini. (Note too that -ginidesc- stores its results in
> matrices rather than
> in global macros.)
>
> The same principles (-foreach- and so on) could be used to
> get and save
> poverty index estimates using -povdeco-.
>
> PS ensure you have the latest version of -povdeco- (version
> 1.3 Feb 2001)
> and -ineqdeco- (version 1.6 April 2001). Get them using the
> -ssc- command.
>
>
> Stephen
> =============================================
> Professor Stephen P. Jenkins <[email protected]>
> Institute for Social and Economic Research (ISER)
> University of Essex, Colchester CO4 3SQ, UK
> Phone: +44 1206 873374. Fax: +44 1206 873151.
> http://www.iser.essex.ac.uk
> *
> * 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/
>
*
* 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/
