Stata The Stata listserver
[Date Prev][Date Next][Thread Prev][Thread Next][Date index][Thread index]

Re : st: Multinomial logits with proportions for dependentvariables

From   Naji <>
To   <>
Subject   Re : st: Multinomial logits with proportions for dependentvariables
Date   Wed, 31 Aug 2005 10:12:54 +0200


It's better to ask respondent to distribute a fixed number purchases over
the J option. The fixed number can be the same for each respondent or the
number of purchases he makes in some period.

Just transform 
Probability of buying a product
Number of product purchases given overall purchases in the product category

I.e. .1 .5 .4 (for A, B & X) into (1,A 5,B 4,X given 10 purchases)
Once you fix respondent overall purchases (here 10), it's easy to get back
to MNL models. Keep in mind than your results depend on the choice of the
overall purchases number

Best regards

Le 31/08/05 6:57, « Clive Nicholas » <> a
écrit :

> Some time ago, Greg Lee wrote:
>> According to some econometrics texts (e.g. Greene's Econometric
>> Analysis), in multinomial logit models the dependent variable can be
>> specified as proportions adding up to one (as in the case where
>> consumers predict their probabilities of purchasing various products)
>> instead of using 0/1 inputs (when you know what they chose).
>> So, to clarify, I have J possible dependent options, and respondents are
>> attaching a possibility chance of choosing that option to each one. I
>> then wish to analyse this via multinomial or conditional logistic
>> regression.
>> Does anyone know if STATA can analyse this? From what I have seen looking
>> around it requires 0 or 1 inputs, but I stand open to correction
> I've never heard of a such model, I have to say. One (not entirely
> satisfactory) fix would be to use Jack Buckley's -mlbeta- routine, which
> models dependent variables-as-proportions as if they were
> beta-distributed, using maximum likelihood to generate the parameter
> estimates. This is _not_ available from SSC, so to download it within
> Stata:
> . net from
> and then click on -mlbeta-.
> The limitation here is, of course, that the probabilities of each outcome
> would have to be modelled separately, rather than together, as you want.
> Perhaps somebody else has a better solution.
> I hope this helps.
> CLIVE NICHOLAS        |t: 0(044)7903 397793
> Politics              |e:
> Newcastle University  |
> Whereever you go and whatever you do, just remember this. No matter how
> many like you, admire you, love you or adore you, the number of people
> turning up to your funeral will be largely determined by local weather
> conditions.
> *
> *   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