What's the rationale behind egen_inequality excluding zeros?

[Simon Heß]

I'm using the user-written egen_inequal to compute Gini coefficients. Like so:

 egen land_gini = gini(owned_hectares), by(village_2003)

It automatically excludes obs with 0 hectares and computes the Gini within the group of non-zero observations.

I fail to see why the this would *ever* be desirable, but in particular I am convinced in the case of land and hectares it should be different.

Why is this like that and what can I do to change this with egen_inequal?

PS: If Zurab Sajaia or @Michael Lokshin don't want to work on the code anymore I am also happy to dive into the ado file.

[Maarten Buis]

You can see the .ado file: viewsource _ggini.ado

[FernandoRios]

Hi Simon,
I think the main reason is that most inequality measures are not designed for handling zero or negative values, in particular when there is a large number of Zeros in the data.
If you decide to go over the Gini ado, one option you have, that can handle Negatives and Zeros, is to use the Covariance Formula of Gini.
Another option you have is to use "rifvar". This is an extension to egen that you can install using "ssc install oaxaca_rif".
you could do something like:

 
 egen land_gini_rif = rifvar(owned_hectares), by(village_2003) gini egen land_gini=mean(land_gini_rif), by(village_2003)

HTH

[Stephen Jenkins]

-ineqdec0- (SSC) estimates Gini coefficients and allows for zero or negative values. [Cf. -ineqdeco- (SSC) which also calculates Gini coefficients and other indices, but works with positive values.] Combine -ineqdec0- with -statsby-. Something like:

statsby gini = r(gini), by(village_2003) saving(my_ginis.dta, replace):   ineqdec0 owned_hectares

and then merge the data in my_ginis.dta back onto your original data file using village_2003 as the merge key

[Simon Heß]

Thanks for all you useful answers

[Leah Bevis]

Hi Stephen, I just found the same problem as Simon, and so I was glad to know that ineqdec0 can calculate the gini including zeros. Your command works perfectly for me if I run it separately from statsby (i.e., I can see the gini indices in the stata output), but when I try to export the r(gini) values by using your command above, it creates a matrix with all missing gini values. Any idea how I need to tweak your command? I looked at the statsby help, and it looks right to me.

[Stephen Jenkins]

Leah: sorry, but I don't understand the problem in your case. (Your reference to "matrix" confuses me.) The following code does what I expect:

sysuse auto, clear
statsby gini = r(gini), by(foreign) saving(junk.dta, replace):   ineqdec0 mpg
use junk.dta
describe
list

resulting in:

. sysuse auto, clear
(1978 automobile data)

. statsby gini = r(gini), by(foreign) saving(junk.dta, replace):   ineqdec0 mpg
(running ineqdec0 on estimation sample)

      Command: ineqdec0 mpg
         gini: r(gini)
           By: foreign

Statsby groups
----+--- 1 ---+--- 2 ---+--- 3 ---+--- 4 ---+--- 5 
..

. use junk.dta
(statsby: ineqdec0)

. describe

Contains data from junk.dta
 Observations:             2                  statsby: ineqdec0
    Variables:             2                  14 Sep 2022 09:18
-------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
Variable      Storage   Display    Value
    name         type    format    label      Variable label
-------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
foreign         byte    %8.0g      origin     Car origin
gini            float   %9.0g                 r(gini)
-------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
Sorted by: foreign

. list

     +---------------------+
     |  foreign       gini |
     |---------------------|
  1. | Domestic   .1301388 |
  2. |  Foreign   .1433695 |
     +---------------------+

. 
end of do-file