[Nick Cox]

ineqdeco does a lot, and I really mean that, and it's written for Stata 8.2, so that speed-ups with Mata (and other means) are possible but not exploited. Paradoxically, some of the speed-ups would require more code to remove the egen calls that are included for convenience.

It is difficult to know how to answer the last question without a sense of how much experience you have in Stata programming.

With a good grasp of Stata and some experience in Stata programming, you could clone ineqdeco and cut out what you don't want, but you're responsible for the result.

With a very good grasp of Stata and more experience in Stata programming, you could rewrite ineqdeco and speed it up: how much is hard to say until it's done. I would regard that as several days' work for such people. it's not just the coding; it's banging hard on the code to check that it handles problems as required.

Stephen Jenkins is the author and may wish to comment. I imagine that he would love to rewrite ineqdeco but is snowed under with more pressing commitments for some time to come.

[Jason Cruso]

Thanks Nick! This is a serious problem for me. As I badly need the within and between measures of inequality using ineqdeco! But maybe I try to look into the code of ineqdeco.

[Stephen Jenkins]

My view, without a lot of thought, is that the speed issue is likely to lie with the number of bygroups, and nothing really to do with -ineqdeco- in a fundamental sense. Tell us how many bygroups there are (unique combinations of id and year). Also seriously look into -runby- as recommended by Robert Picard. (It'll be faster than -statsby-.)