[Date Prev][Date Next][Thread Prev][Thread Next][Date index][Thread index]

Re: st: two level model

From   Viktor Slavtchev <>
Subject   Re: st: two level model
Date   Sun, 19 Aug 2007 13:45:51 +0200

first of all thank you very much for your very quick reply.
However, I have an additional question.
Please correct me if I am wrong, but as far as I know, in the McFadden's choice model only one option is allowed to be chosen.
That is, everyone goes for only one particular option out of set of many options. In other words, if one decides to invest in the US, all the money goes to US and nothing can be invested elsewhere.
As to my understanding, it will be equivalent to -mlogit- model or better -clogit- if a subset of the covariates is attributed not to the particular choice but rather to some individual characteristics.
In my case multiple choice can be made. Is -clogit- in such a case an appropriate method?
What do you think of some kind of two-level approach?
And second, it seems to be that -asclogit- is available only in Stata 10.
Since I am working with Stata 9 my question is about the differences between the new -asclogit- and the old -clogit-.
According to the Stata manual McFadden's choice model can be estimated by means of -clogit-.
warm regards

Maarten buis wrote:

I have only a very partial answer to your question. The situation you
describe reminds me of what economists call McFadden's choice model,
-asclogit- in Stata. This has a very similar structure: multiple
options and the choice between them can be influenced by both
characteristics of the choice and the individual making the choice. The
problem is that this model deals with discrete outcomes, and you are
dealing with proportions. However, finding and understanding a very
similar model can sometimes be very helpful in solving such problems.

Hope this helps,

--- Viktor Slavtchev <> wrote:

I have followed model.
People can make (multiple) choices. For example, people can invest
their money in different countries, whereat multiple investment
options are possible at the same time.
Some people decide to invest all the money in (say) Germany, other decide to invest in both Germany and US, third group invests in
Germany, US and UK.
The decision depends on country specific characteristics and on individual characteristics.
The estimation of the impact of country specific characteristics on
the investment decision should be not problem since these
characteristics are unique for each country.
But how to estimate the impact of individual characteristics on the investment decision since these are the same for each individual (do
not differ with the different countries within one person)?

id country investment_share interest_rate gender
1 Germany .2 .02 0
1 US .5 .03 0
1 UK .1 .025 0
1 France .1 .022 0
1 Italy .1 .023 0
2 Germany .6 .02 1
2 US .1 .03 1
2 UK .1 .025 1
2 France .1 .022 1
2 Italy .1 .023 1
3 Germany .2 .02 0
3 US .5 .03 0
3 UK .1 .025 0
3 France .1 .022 0
3 Italy .1 .023 0

I've just tried followed:
xi: glm investment_share interest_rate gender, family(binomial) link(logit) scale(x2)

Is this an appropriate way to solve the problem?
Or should I use some kind of multilevel approach?
Could -xtmixed- be an appropriate method?
Or perhaps there is another way?
Tanks for any help.

* For searches and help try:

Maarten L. Buis
Department of Social Research Methodology
Vrije Universiteit Amsterdam
Boelelaan 1081
1081 HV Amsterdam
The Netherlands

visiting address:
Buitenveldertselaan 3 (Metropolitan), room Z434

+31 20 5986715

___________________________________________________________ Want ideas for reducing your carbon footprint? Visit Yahoo! For Good
* For searches and help try:
*   For searches and help try:

© Copyright 1996–2017 StataCorp LLC   |   Terms of use   |   Privacy   |   Contact us   |   What's new   |   Site index