cumulative logit model

Dear all,

I would be very grateful if someone could help me!

I used binary logistic model to draw a probability curve of areas having endangered plants based on vegetation biomass. But, it would be better to include the number of endangered species (which varies from 0 to 3) in the model as well. Is the use of an ordinal logistic model the most appropriate for it?

If yes, how can I use it?

Thank you very much!
cumulative logit or Poisson?

My dependent variable is 'number of alien species in a plot' and the independent variable is 'aboveground biomass in the plot'.

Let me explain better...

I have used logistic regression to draw a probability curve of plots having one or more alien species in relation to biomass (figure in attachment). But doing it, I lost the information about the number of endangered species (that can vary between 0 and 3), since logistic regression works with presence/absence. To include this information, I am in doubt about using the cumulative logit model or a Poisson regression.

It seems that the cumulative logit model works better when the number of observations in each state is about the same (which is not my case, as you can see in the figure in attachment). I could end up combining level 2 and 3 to a level 'more than one'. But still there are not so many observations in this combined group...

Do you think that it's better to use a Poisson regression for this data or to use the cumulative logit anyway?

Thank you very much for your help!!


TS Contributor
I 'd say combine the 2 and 3 in one category and go on with ordinal. The point with ordinal is that you nedd to have the assumption of proportionality satisfied and also in SPSS help, there are tips that help you choose the link function, but then the interpretation is more difficult. with poisson you would still have the problem of one observation at level 3. How about bootstrapping your data? But i assume you use SPSS
Dear Masteras, thank you for the advice! I am still a beginner in Statistics, I don't understand well what bootstrapping means. I will study about it.

I use SPSS 17. I ran Analyze>Regression>Ordinal, combining 2 and 3 (putting the only 3 observation as 2). Would this be the right way to do it? I am sorry if it's a very dumb problem, but I get an output that I don't know how to interpret and to draw the line in the graph, since it doesn't show me the coefficients of the formula to predict the logit transformation of the probabilities of having an alien plant in the different levels of biomass.

The output for logistic regression gave me the Z and p values and the a and b values for the equation, and I could draw the graph that you saw previously. Why doesn't the ordinal regression give me these values?

In attachment I send the output that I get from SPSS 17 by running the logistic regression and the ordinal regression. So you can know what I am talking about.

Your help is very precious, thank you from deep heart.


TS Contributor
Analyze>Regression>Ordinal, combining 2 and 3 (putting the only 3 observation as 2)

then the biomass in the covariates. Press output, select test of parrallel lines and the three first from the saved variables choices and continue and then ok