# Non linearity with non-interval DV

#### noetsi

##### Fortran must die
The non-linear methods I know (have read about) have DV that are interval. How do you deal with non-linearity when they are not interval.

HLsmith may have answered this but I am not sure since I rarely work with general additive models

Note I am trying to do this the simplest way possible which is accurate

Last edited:

#### noetsi

##### Fortran must die
I am running logistic regression and it appears at least one variable is non-linear. I can't use logistic regression odds ratios in that case. But I don't know what method to use. All the non-linear methods I know to deal with this have interval dependent variables.

#### Dason

##### Ambassador to the humans
And what do you interpret non-linear to mean? Because if you just think that the relationship between the log-odds and one of your predictors isn't a straight line you can still use logistic regression. Just like how if you had that in a normal regression you could include transformations of your predictor to allow the model to not look like a straight line.

#### noetsi

##### Fortran must die
Well I interpret it to mean the impact changes over levels of the predictor. I know with an interval variable you can do transformations, for instance logging or more generally box cox. I was not sure you could do this with logistic regression. Some of the non-linearity analysis I have looked at says simply you can not use this with a non-interval DL. But they aren't transformations.

I understand that logistic regression does not assume linearity between the predictor and the DV for the raw data. But it does assume it between the predictor and the logit. I tested it with box tidwel - because I am unable to interpret the residuals. I don't know what non-linear logistic regression residuals should look like