# Logistic regression with ordinal variables

There are 5 variables (y and x1, x2, x3, x4). All of them are ordinal (take values 0,1,2,3). Sample consist of 200 observations. Is it possible construct multinominal logistic regression in this case. And how can I estimate weight of each factor (in percent).

• stats.stackexchange.com/q/195246/3277 is a similar question, only that there Y is binary. Consider further links found in that thread. Commented Jul 2, 2016 at 12:17

If y is ordinal, you should probably explore ordinal logistic regression first.

Ordinal independent variables are tricky; the usual choice is to treat them either as categorical or continuous and neither option is perfect. However, there are some other choices, although they are not widely known or used.

In SAS, you can use optimal scoring for ordinal variables using PROC TRANSREG. This is probably also available in R, but I don't know which package.

Essentially, this involves figuring out the optimal values to use fro 0, 1, 2 and 3 to make the regression as good as possible.

I am not sure what you mean by "estimate the weight of each factor in percent".

• Thanks a lot for your answer. In my research, independent variables are individual customer's satisfaction about service, product, personnel etc. Dependent variable is global customer satisfaction. I would like to know which factor has the most and the least influense on result (y). If it's only possible in percent. Commented Jul 2, 2016 at 12:37
• You can't really know this. Commented Jul 3, 2016 at 13:13

As @Peter said, ordinal logistic regression is the approach to use if y is ordinal.

There are several options in R. See https://stats.stackexchange.com/a/93482 for a good description of these options.

If your data collection involved repeated measures, then you should consider the Ordinal package, which implements mixed model ordinal regression.