1

My independent variable is Party Identification which I think will be best to create dummy variables for (Conservative, Labour) and then Lib Dem would be the intercept if I'm not mistaken.

The dependent variable is a Likert Scale and I would like to model the effect of party identification on the ordinal variable to look for significant data etc.

If I was to do this in R, can I use polr, do I have to run a separate model for each dummy variable or can I include them all in same model with the variable I haven't created a dummy variable for being the intercept and how will I interpret the results.

kjetil b halvorsen
  • 77,844
James
  • 11
  • 1
    Welcome to Cross Validated! How would you do it for a continuous $y$ variable (e.g., a standard OLS linear regression)? – Dave Apr 28 '22 at 14:05
  • @Dave So I am assuming I can create the model with n-1 dummy variables for my categorical independent variables. – James Apr 28 '22 at 14:26
  • It sounds like, if you had a continuous $y$, you would use a regular ANOVA. Is that accurate? – Dave Apr 28 '22 at 14:30
  • @Dave Yes, but due to the ordered nature of the dependent variable I believe polr is correct – James Apr 28 '22 at 14:58
  • 1
    General resources on ordinal models may be found here. There are two functions in the R rms package for ordinal regression that give you more graphics and interpretation tools. – Frank Harrell Feb 14 '24 at 13:09
  • In R, you don't need to create the dummy variables. You just use the factor variable on the right hand side of the formula, and R will handle it. However, you choose which contrast coding scheme R uses to do this, if that's of interest. – Sal Mangiafico Feb 14 '24 at 15:27
  • In R, I also recommend the ordinal package, with is supported by e.g. the emmeans package. – Sal Mangiafico Feb 14 '24 at 15:27

1 Answers1

1

You will fit one (ordinal) model with the dummy-coded predictor variable. If you have some other predictor variable, the situation is analogous to Separate Models vs Flags in the same model

If there is no other predictor variables, the situation is analogout to one-way anova. Fitting separate regression models for each factor level do not make much sense in that case, since there will be no variation in x ...

kjetil b halvorsen
  • 77,844