Ordinal logistic regression divided by level


New Member
I'm working with a dataset and should do univariate regression on it, to see if there are a relationship between the response variable, nominal with three categories, and the explanatory variable, values between 0-5 (would say it is nominal).
I would like to get a result where you can see the B-value (and s.d. etc) for two of the response variable and have the first level as reference category.

Any ideas how to do this?



No cake for spunky
Well first off if the DV is nominal with 3 levels you have to use multinomial logistic regression. You say at one point above that you have one response variable with 3 levels and later that you have "two of the response variable." Do you have one or two response variables?

If you have a single DV and a nominal response variable you can make the nominal response a dummy. But you can only make one of its levels the reference category.


New Member
Sorry for the confusion.
I have one ordered nominal responce variable with 3 levels, the first level I want to use as the response variable, and one explanatory variable, also nominal with 5 levels.
To make it more clear, the responce is "no depression", "risk for depression" and "depression", want to use "no depression" as the reperence category. My explanatory variable is number of misscariage before this pregnancy; 0,1,2,3,4 and 5 or more.

If I use:
polr(depression ~ misscariage, data = Test, Hess = TRUE, method = "logistic")
(from MASS package)

polr(depression ~ misscariage, data = Test, Hess = TRUE, method = "logistic")


No depression|Probable depression
Probable depression|Depression

Residual Deviance: 2583.891
AIC: 2589.891

How I would like the result to look like

Depression____________________B_________Std. Error
Probable Depression Intercept
_______Depression Intercept

Or something simmilar to this, so that I can compare the different levels of "Depression" with each other..


No cake for spunky
I wanted to appologize for addressing this, I missed the category this was in so I did not realize you were asking a coding question (I use SAS not R). I was curious why you considered your DV nominal rather than ordinal - it seems to go from no depression to depression (that is to have ordered states).


New Member
No worries, always good to reflect on what you are doing, can somethimes find better ways of doing it, though it is a coding questiong :)

I want to consider my DV as ordinal, and as I understood it the polr() should do this, and using an logistic as the link function, but I can be wrong about this. I will also do further analysis of the dataset, this is only the first step, but can not get the output and the result that I "want".. Got an example from SPSS and trying to get simmilar output as that example...


No cake for spunky
I thought when you said ordered nominal you meant nominal not ordinal. I assume R works like SAS in that if you use the logistic link function and the data has more than 2 levels the default will automatically be considered it ordered (but the documentation is the only way to be sure). One thing I wanted to caution about - because it is easy to miss - is that ordered logistic regression makes the assumption central to the model that you can dichotomize your DV and that the way you do so does not matter (that is which way you would have chosen to dichotomize it does not matter).

If this assumption is not correct you have to use multinominal logistic regression even if your DV is ordered. SAS calls the test of this assumption the Score Test and I am sure R has an equivalent test.


New Member
Okey, I usually know what I mean, not I just need to learn how to present it in a way so that other people understand what I mean too ;)

Thanks for the heads up, have to check that too then :)