Notice: On March 31, it was announced that Statalist is moving from an email list to a forum. The old list will shut down on April 23, and its replacement, statalist.org is already up and running.

# st: Problem with excluded observations with -ginidesc-

 From Joe McCrary To Statalist , ieeracba@satlink.com, Amanda Hoffman Subject st: Problem with excluded observations with -ginidesc- Date Tue, 14 Sep 2010 10:19:26 -0400

```Hi all.

I've been using -ginidesc- to develop inequality indices of
school-level measures for several districts in the U.S. My goal is to
be able to chart them longitudinally to see if enrollment policies
impacted the mixes of students (e.g., those eligible for free and
reduced-price meals (FRL)).

In 2003, one of the states did not report FRL data, so I wrote in a
statement to exclude those observations for that year:

. ginidesc  Pfrl03 if DistID~=3700720, by(DistID)

Gini Coefficient by subgroups
of DistID

----------------------
K |     Gini_k
----------+-----------
625690 |      0.337
1201710 |      0.225
3700720 |      0.274
----------------------
Stored in matrix

As you can see, even though I was excluding District 3700720, ginidesc
still reported results for it. Also, the numbers for the other
districts were way out of line with the years before and after 2003.
So since -ginidesc- calls -ineqdeco-, I then ran that to compute the
gini indices, which produced a table that makes more sense:

. ineqdeco Pfrl03 if DistID~=3700720, by(DistID)

Subgroup indices: GE_k(a) and Gini_k

----------------------------------------------------------------------
DistID |     GE(-1)       GE(0)       GE(1)       GE(2)        Gini
----------+-----------------------------------------------------------
625690 |    0.82561     0.24858     0.15316     0.14355     0.27361
1201710 |    0.32468     0.21598     0.18285     0.18375     0.33745
2503270 |    0.09620     0.08651     0.08126     0.07955     0.22489
3700720 |
----------------------------------------------------------------------
The Gini coefficients are consistent with those in the other years,
and 3700720 is excluded.

Anyone using -ginidesc- with subgroups, especially when there are
cases that are being excluded, should carefully check their results
against -ineqdeco-.

--
Joe McCrary
Senior Research Associate
WestEd-Atlanta
260 Peachtree St., Suite 2200
Atlanta, GA 30303

ph: 541-782-8626
fax: 404-393-3797

Skype: joe.mccrary