INEQDECO: bygroup

[hyunjoo eom]

Dear members,

I have a question about the ineqdeco, especially the results of the bygroup() option.

(I am interested in the GE)

The result inequality value of the subgroup indices seems like the population is treated as a separate population, rather than showing inequality value within each subgroup.

By this, I mean GE= (within from group1 + within from group2 + within from group 3) + between

I am only seeing the sum of within-group inequality and between, but not each group's within contribution to inequality.

Is there any way to measure this?

Thank you in advance!

[Stephen Jenkins]

You can calculate the required expression(s) using the saved results. After running ineqdeco, type return list and you will see that all the ingredients you might need are there.

PS you should tell us where you got community-contributed programs from -- please read the Forum FAQ. (ineqdeco and its sibling ineqdec0 are on SSC.)

[hyunjoo eom]

Dear Dr. Jenkins,

Thank you for your reply.
ps. I am using a program downloaded from ssc ineqdeco.

Best regards,

[hyunjoo eom]

Dear Dr. Jenkins,

Thank you for your reply the other day.
I tried to see if I can see the within components for each groups of GE1, but the value I found was the summed value.

I went back to the ineqdeco code from http://fmwww.bc.edu/repec/bocode/i/ineqdeco.ado

and tried to see if I can get the value before they are summed.

I think the code line 341,
egen double `with1' = sum(`fi' * `i1k' * `lambdak') if `touse' Is where it does the summation of within components of k groups.

What I would like is to get the within values of each k, which would look something like

if there are 3 k groups:

'with1_k1' = `fi' * `i1k1' * `lambdak1'
'with1_k2'= `fi' * `i1k2' * `lambdak2'
'with1_k3'= `fi' * `i1k3' * `lambdak3'


I am fairly new to writing coding.
I think the code is there, but I just need to generate new variables from the codes.

If you could give me any advice on doing this, it would be greatly helpful.

Thank you,

[Stephen Jenkins]

Read the literature and help file again, in order to properly understand the inequality decomposition formulae. Don't look at the ado-file code; match up the components of the decomposition formulae with the various returned results.

Study carefully the results of running the code that follows (for the case of GE(0)), and then apply the logic to your case

sysuse auto
ineqdeco mpg, by(foreign)
return list
di r(ge0) // GE(0)
// decomposition elements
di r(v_0)*r(ge0_0) + r(v_1)*r(ge0_1) + r(between_ge0) 
di r(within_ge0)
di r(v_0)*r(ge0_0) + r(v_1)*r(ge0_1)

di 100*r(v_0)*r(ge0_0) / r(ge0)
di 100*r(v_1)*r(ge0_1) / r(ge0)
di 100*r(within_ge0) / r(ge0)
di 100*r(between_ge0) / r(ge0)

[hyunjoo eom]

Dear Dr. Jenkins,

Thank you for your reply. Your comments are very helpful.

Warm regards,

[hyunjoo eom]

Dear Dr. Jenkins,

I have a question regarding the theil's index, I hope you could give any guidance.

Income is one of the most widely used for measuring inequality of individuals (and regions).

I am interested in the distribution of two values, say the share of resource A and share of population B.

In measuring inequality of "resource available to pop B", I created a variable that is the absolute difference between resource A and population B in the geographical areas (rather than individual scale).

I am treating this "magnitude of difference in resources by population" in place of income, in the inequality index.

The reason for trying to use this theil's index is for its decomposability.

I do not have clear idea whether or not this can be used to measure the inequality of distribution of resources per population in the area.

More simply, I would like to know if variable of interest other than the single value can be used.

I have seen the use of GDP to calculate inequality between regions, which is a somewhat modified version of some value, but I am not sure if values that are calculated rather than the "natural" value can be used.

Thank you in advance,
Hyunjoo

[Stephen Jenkins]

Your constructed variable -- if I understand it correctly -- appears likely to have zero and negative values.

Inequality indices such as Theil are typically defined for variables that take on positive values only. Theil is calculated using logs of values; the log of a negative or zero value is not defined. Any obs with such values would be dropped from the calculation. (The Gini coefficient is unusual in being meaningfully defined for zero and negative values.)

Maybe you could make some head way by looking at ratios rather than absolute differences? What is meaningful is up to you, the researcher with specialist knowledge of the field of application.

[hyunjoo eom]

Dear Dr. Jenkins,

Thank you for your comments.
The main question I had was whether or not I can construct variables to measure inequality of distribution, and it seems I can use constructed variables as long as I construct variables with non-negative values that are not zero.

Thank you as always!
Warm regards,
Hyunjoo