Calculating inequality of educational

[Sofia Cruz]

Hello;

I am using data from survey to calculate inequality of educational participation across people in a country of Central America. I would like to compute the Gini coefficient on the variable: years of schooling. Also I want to compute this variable by groups such as; region or gender. Can I use the command "ineqdeco" assuming that variable is discrete.

Thanks for your attention,

Regards.

[Stephen Jenkins]

You could use ineqdeco and get numbers out. In fact, perhaps you could use instead use ineqdec0 (that is a "0" at the end) which is also available from SSC, because that calculates the Gini for data sets that include obs with values of zero. (ineqdeco drops zero values.) But a more fundamental question is whether you should use the Gini to measure inequality in "years of schooling" or use some other approach, given the nature of the distribution with which you are dealing. I suspect that, in addition to the discrete nauture of the variable, there are quite a few with zero, and few values at the top. I seem to recall the World Bank or some other similar sources have acknowledged these issues somewhere, and it'd be worth you searching for these papers. (Sorry to be so vague; I am away from office and don't have access to usual resources.)

[Konrad Zdeb]

Hi,

Those seem to be extremely useful. Just a minor side that the associated help file appears not to display correctly.

[Stephen Jenkins]

Konrad: I don't know why the programs are not displaying properly (or whether this occurs for other programs on SSC as well). To expand: I have both my programs installed on my PCs and laptop and if I (a) help ineqdeco or help ineqdec0, the programs display fine (as they have for years). On the other hand (b), if I findit ineqdeco, and click on the blue-coloured link to the program, I get a screen in which there are clickable blue links to the help files. If I click on them, I see what you show in your screenshots. I suspect you did (b). I don't know why this is occurring. If this is a general issue with other programs on SSC as well, perhaps you should start another, new, thread on the topic with more informative title?

[Stephen Jenkins]

Sofia: the World Bank publication I was thinking of was probably Vinod, Wang and Fan, (1999), "Measuring Education Inequality: Gini Coefficients of Education", downloadable from http://elibrary.worldbank.org/doi/pd...1813-9450-2525. They have some discussion of the prevalence of zeros (and acknowledge the upper bound issue). The software problems that they allude are not a problem nowadays: ineqdec0 on SSC will calculate Ginis for data including zeros (where you have unit record data on individuals -- as you do). You can use glcurve on SSC to draw Lorenz curves in the manner that Vinod et al. do too. Given my own query about whether a Gini coefficient is a good way to summarise inequality in "years of education", I would place high priority on also looking at (comparing) complete distributions of years of schooling, both using Lorenz curves and also histograms/bar charts (but not kernel density-estimated density functions given the intrinsic discreteness of the variable). Good luck.

[Nils Enevoldsen]

I expected the malformed help file to be due to some line-ending issue, but after reading this earlier thread, it seemed more likely to be a syntax error in the help file that confuses -rnethelp- but not -help-. I don't know the syntax of help files, but through process of elimination I discovered that eliminating the double tab character between the following sections of the help file will cause it to be rendered correctly. Since there's no double line break (just "line break, tab, tab, line break"), I think rnethelp is treating them as one paragraph.

{p 8 17 2} {cmd:ineqdeco} {it:varname} [{it:weights}]
    [{cmd:if} {it:exp}] [{cmd:in} {it:range}]
    [, {cmdab:by:group}{cmd:(}{it:groupvar}{cmd:)}
       {cmdab:w:elfare} {cmdab:s:ummarize}   ]
        
{p 8 17 2} {cmd:ineqdec0} {it:varname} [{it:weights}]
    [{cmd:if} {it:exp}] [{cmd:in} {it:range}]
    [, {cmdab:by:group}{cmd:(}{it:groupvar}{cmd:)}
       {cmdab:w:elfare} {cmdab:s:ummarize}   ]

Also note that to troubleshoot the issue, one can use -rnethelp- with a local file as well as a remote one:

rnethelp "http://fmwww.bc.edu/RePEc/bocode/i/ineqdeco.hlp"
rnethelp "~/Library/Application Support/Stata/ado/plus/i/ineqdeco.hlp"

[Stephen Jenkins]

Nils: great detective work -- thank you. I will look into making the relevant fix in the help file (and send revision to SSC) when I have time ... which is means 'maybe not for a while'.

[Nils Enevoldsen]

One final update. I reported the discrepancy between -rnethelp- and -help- to tech support. They reply: "Thank you for pointing out that to us. It has been reported to the development team and will be fixed in the future." They did not specify which behavior was correct.

Apologies to Sophia Cruz for sidetracking the thread.

[Stephen Jenkins]

I can see the unnecessary double tab in my file and have deleted it. However, checking that all is ok now (before sending the revision to SSC) is frustrated by my inability to access rnethelp. Perhaps I am doing something wrong but, imitating Nils, I get what is shown below when I issue the following lines from the command line:

. rnethelp "http://fmwww.bc.edu/RePEc/bocode/i/ineqdeco.hlp"
unrecognized command:  rnethelp
r(199); 
. which rnethelp
command rnethelp not found as either built-in or ado-file
r(111);

However, If I findit ineqdeco, and hover my mouse over the blue-coloured link to the help file that appears on the relevant SSC page, I can see that Stata will issue that very same command were I to click on the link. This particular issue may be a "me" thing, or perhaps a MS Windows thing? (I am using Win 7/64; I guess Nils is using some flavour of *nix.)

[Nils Enevoldsen]

Enter

rnethelp "http://fmwww.bc.edu/RePEc/bocode/i/ineqdeco.hlp"

in the Viewer, rather than in the Command window. I don't understand what -rnethelp- actually is. It's kind of like a command, but it only works in the Viewer?

[Stephen Jenkins]

Aha! -rnethelp- does indeed work now, via the Viewer. I've revised the help file, and it looks fine using -rnethelp-. So, I'll send the file to Kit Baum to replace the current version on SSC. Thanks again, Nils. Back now to Sofia's substantive topic!

[Sofia Cruz]

Hi there!!, Sorry if I replied until but I was sick a terrible virus very common in Central America. Just now I back I have read your messages and you are right Stephen Wang and Fan, (1999) this paper show about Gini coefficient and thanks for link and I will be reading the paper, and thanks to all for your help. Regards