RE: st: Egen for logit regression

["Nick Cox" <n.j.cox@durham.ac.uk>]

It sounds from this that your rich and poor variables are percentiles of income. That being so, a -logit- model seems quite inappropriate to me. You should be better off using a quantile approach consistently. 

Nick 
n.j.cox@durham.ac.uk 

muhammed abdul khalid

the reason of choosing logit is because there is negative corr for
wealth and earnings for the poor (p.20) and rich (p95). Thus, logit
for regression for poor and rich, and qreq for median group. which
means i have 3 regressions:

logit rich  lgearnings i.ethnic  i.strata  i.occupation i.edu age
logit poor  lgearnings i.ethnic  i.strata  i.occupation i.edu age
qreg lgwealth lgEarningsPc  i.ethnic  i.strata  i.occupation i.edu age
, quantile(.5)

This is not correct model?

As for effect of inheritance, it is not possible to look at the change
in wealth as my data is only for year 2007; comparison is not
possible.

2010/4/7 Maarten buis <maartenbuis@yahoo.co.uk>:
> --- muhammed abdul khalid wrote:
>> I have a question regarding my logit regression:
>> logit wealth lgearnings i.ethnic  i.strata  i.occupation i.edu age
>
> -logit- is for binary data, while a variable like wealth is typically
> not binary. Why did you choose -logit-?
>
>> How do i take into account that the effect on wealth  because some ethnic
>> group has more PREVIOUS wealth vs the others? My data is panel data of
>> only one year, so i cant compare with the previous years. I would like
>> to check the the premium on ethnicity IF everyone starts with same
>> wealth. Currently in my data , of which the age starts at 20, there is
>> positive wealth for ome obs (which i suspect due to inheritance).
>
> Maybe you can look at change in wealth rather than wealth per se.

*
*   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/