Re: st: Computing Quintiles with frequency weight

[Nick Cox <njcoxstata@gmail.com>]

Usually the explanation for this variability is in ties in the data,  
so inspect your data carefully. Also, frequency weights imply ties  
unless all are 1.

Nick

On 5 Aug 2011, at 03:49, "Polloni Stefano"  
<stefano.polloni@umontreal.ca> wrote:

> Greetings,
>
> I am trying to observe comsumption patterns among the different  
> income quintiles of a population. My sample contains 10 801  
> observations, each with a corresponding frequency weight so as to be  
> representative of the whole population (about 30 million people).
>
> I am using this command to compute my quintiles:
>
> xtile quintile= hhinctot[fw=weight], n(5)
> (hhinctot as the income variable)
>
> After doublechecking, I realized that some of my quintiles contained  
> substantially less or more than 20% of the observations (while  
> effectively taking in consideration the frequency weight - i.e. as  
> if I duplicated myself the observations according to their frequency)
>
> this is the result I get:
>
> Quintile 1: 20,406%
> Quintile 2: 20,552%
> Quintile 3: 19,239%
> Quintile 4: 22,323%
> Quintile 5: 17,479%
>
> I can get more precise results by computing the quintile myself  
> "manually", any idea why stata gets this far from 20%
*
*   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/