Ordered Probit/Logit models

[chris...@virgin.net]

I'm using SPSS to analyse some categorial data (so am using some loglinear
model.) I've singled out one of the factor variables as being dependent on two
others and so am using the logit model. However the dependent categorial data
is clearly ordered and I don't think the standard SPSS logit regression has an
option for this. Does anyone know whether ordered logit/probit regression (I
have heard these other models described as adjacent categories and/or
continuation ratio logit models)
is possible using SPSS?

Thanks

Chris Bojke

-----------== Posted via Deja News, The Discussion Network ==----------
http://www.dejanews.com/ Search, Read, Discuss, or Start Your Own

[J.Hendrickx]

In article <7561k5$cra$1...@nnrp1.dejanews.com>, chris...@virgin.net
says...

> I'm using SPSS to analyse some categorial data (so am using some loglinear
> model.) I've singled out one of the factor variables as being dependent on two
> others and so am using the logit model. However the dependent categorial data
> is clearly ordered and I don't think the standard SPSS logit regression has an
> option for this. Does anyone know whether ordered logit/probit regression (I
> have heard these other models described as adjacent categories and/or
> continuation ratio logit models)
> is possible using SPSS?
>

There's a macro for ordered logistic regression, aka cumulative logistic
regression, at http://baserv.uci.kun.nl/~johnh/mlogist/.

Adjacent logits is basically a multinomial logistic model using the
repeated contrast for the dependent variable, rather than the indicator
contrast. The log likelihood is unchanged since no restrictions are
imposed, and the parameters of one model can be transformed into the
other in a straightforward fashion. See Agresti (1990: 318), "Categorical
Data Analysis", for details. You can estimate this type of model with the
mlogist macro by specifying the repeated contrast for your dependent
variable.

Continuation ratio models can be estimated as separate binomial logistic
models for 1 versus >1, 2 versus >2, etc. See Agresti, pp. 319-320.

Hope this helps,

John Hendrickx