interpreting ineqdec0 bygroup output

[laura leigh]

I am generating wealth inequality metrics by metro area for the entire nation (U.S.) The wealth data include negative numbers, so I use ineqdec0 to calculate GE(2) and Gini by CBSA - like this

ineqdec0 wealth, by(CBSA)

the figures i obtain are difficult for me to interpret. I find there are negative numbers in the list of gini coefficients. So they are not gini coefficients, they are decompositions, but how do I interpret them?

I may in actuality need gini coefficients for each CBSA, and so I see that I can use sgini for that. But I would like to better understand the output from ineqdec0 as it may be useful too.

Thanks in advance.

[Stephen Jenkins]

Laura; please read the forum FAQ, and note the recommendations to (a) tell us where you got your user-written program from; and (b) to show us exactly what you typed into Stata and exactly what you got back, and reporting this using CODE delimiters. The FAQ explains why.
What may be particularly relevant here is whether you are using the latest version of -ineqdec0- (or -ineqdeco-). You can get it from SSC. Also relevant for interpretation is the mean of your wealth variables (which might be negative). A GIni coefficient is only guaranteed to be between 0 and 1 if the outcome variable is non-negative. So check the mean values for each CBSA.
I don't understand the question about needing the Ginis for each CBSA. The values are both displayed, and also saved in r(.) macros

[laura leigh]

reformatted this answer below

[laura leigh]

Dr. Jenkins,

Thanks so much for your response and my apologies for not following the FAQ.

I installed ineqdec0 in the last month by typing

ssc install indeqdec0

, so it is from ssc. I just tried to reinstall and it confirmed I have the latest version.

The command I use to obtain results is as follows. I have also included the warnings and notes that show up in the output following the command.

ineqdec0 wealth, by (CBSAnum)
Warning: wealth has 261397 values < 0. Used in calculations
Warning: wealth has 7 values = 0. Used in calculations
Note: p10 (and smaller percentiles) <= 0

For many CBSAs, the gini coefficients take on negative values, and now that I check, mean wealth is also negative in those CBSAs. So perhaps that is it! I thought perhaps I was misunderstanding the output and that these were not in fact gini coefficents.

I have no idea how to interpret the negative results, and I suppose I will just use GE2 instead. Perhaps my situation is odd in that I am interested in wealth inequality but taking into account highly indebted populations.

Regards,

Laura

[Stephen Jenkins]

Laura: they really are Gini coefficients. They are well-defined. Interpretation: what you have is a Lorenz curve that starts below the horizontal axis (cumulating over those negative values), before rising above it in the way that you expected. As usual, the Gini can be defined geometrically with reference to the Lorenz curve. Moreover, as usual, larger (less negative; more positive) values correspond to more inequality. "Just using GE2" (half squared coefficient of variation) doesn't solve the issue. When you're working with distributions that contain negative values or zeros, things aren't the same as "usual". (For an example of a variable with zero and positive values, think of "years of schooling". In many developing countries, there may be a significant fraction of the population with zero years of schooling.) For some further reading about the issues that arise when summarising distributions of wealth, you could look at https://www.iser.essex.ac.uk/publica...s/iser/2005-05 and references therein.

[laura leigh]

Thank you for these comments. I thought I had posted a thank you in response and see that I have not - they were very helpful.

I am looking at this https://www.iser.essex.ac.uk/publica...s/iser/2005-05 article and seeing that one method to handle data with negative wealth that generate Gini coefficients above 1 is to use a normalization provided in Chen, C., Tsaur, T., & Rhai, T. (1982). The Gini Coefficient and Negative Income. Oxford Economic Papers, 34(3), new series, 473-478. Retrieved from http://www.jstor.org/stable/2662589.

I don't suppose I could be this lucky, but any chance there is a command in indec0 or another user-written program to perform these normalization? Another scholar has suggested that I simply shift the entire distribution by adding the most negative value to each item so that every item is positive, but without going into it at length, I don't think that's right.

[gokhan ekonomi]

Hi Dr. Jenkins; Stephen Jenkins i have a question about this subject.

I m studying on ealry modern probate records. these records contains some honorific titles. So i want to see overal gini of population and theil index with intragroup-between group contributions. So i will run ineqdeco or ineqdecO command. I have some negative values because of debts. i saw in the article "The Gini coefficient and the case of negative values" Francesca De Battistia , Francesco Porro∗b , and Achille Vernizzia 2019. "When the negative values represent less than the 1% of the total observations of the dataset, we suggest to use Ga, since the loss of information due to the discarding of negative values is balanced out by the ease of calculation". In my sample I have 1000 samples and negative values does not exceed 100. so i neglected negative values and i transformed wealth values to log. form and use "inecdeco". so is it right approach?

actually my question is that; my datas contains negative wealth values but mean is positive and gini is in between 0 and 1. do i have to use ineqdec0 with negative values or upper method is more fit to this sitouation?

i will be so glad if anyone halp me;
kind regards

[Stephen Jenkins]

It seems odd to me to ignore the negative values. Why now 'show' the whole distribution as well. Use multiple perspectives, with the choice depending on your research goals. Some worked examples in: Markus Jäntti and Stephen P. Jenkins, ‘Methods for summarizing and comparing wealth distributions’,ISER Working Paper 2005-05, https://www.iser.essex.ac.uk/publica...s/iser/2005-05

PS I see no great need to normalise the Gini to constrain it to the [0,1] interval.

[gokhan ekonomi]

thx for ur return; if ı use "ineqdecO" i will obtain only GE(2) total, Within-group inequality, GE_W(a) and Between-group inequality, GE_B(a).

so can i interpret these results just like in theil (GE(1))?
when i run ineqdec0;

GE(2)= 1.19796

Within-group inequality, GE_W(a)
GE(2)= 1.15187

Between-group inequality, GE_B(a)
GE(2)= 0.04609

(In summary, the big part of inequlatiy stemmed from within group inequality. )
i m really glad for the help.

[gokhan ekonomi]

I also run "ineqdecgini". Residual Gini > within group GİNİ > Between group gini in terms of actual and % values.
so what does it mean?

do u have any comment or article to understand the difference between "ineqdec0" & "ineqdecgini"?