Ordinal logistic regression with continuous and categorical independent variable (both ordinal and nominal) We want to perform ordinal logistic regression with the following properties:
Dependent variable: Three ordinal variables (low, medium, and high)
Independent variables:

*

*Age : Continuous variable

*Region: Categories (Nominal)

*Income : Categorical (Ordinal) - Data collected in income range; for instance,
from 20000USD-50000USD, 50000USD-1000000USD, >1000000USD

Transformations

*

*Ordinal variable levels to low, medium, and high to 1, 2, and 3, respectively.

*Income: Convert 20000USD-50000USD, 50000USD-1000000USD, >1000000USD to 1, 2, and 3. (Treating ordinal categorical to continuous)

Now, I have only converted region to factors while performing analysis in R. Do I need to make any transformation for nominal categorical variable "Region" before performing ordinal logistic regression? For instance, dummy variable conversion?
 A: Yes you always need to transform nominal categorical variables into dummy variable before including them in a regression model of any kind (including ordered logit).
The coefficients you get from any regression model always tell you how you expect the dependent variable to change (in some way or another) when you increase the associated variable "by one unit" (holding other variables constant). For a continuous variable like age or income a "one unit increase" makes sense - you get one year older or earn one more dollar. For a dummy variable like "female" or "recievedtreatment" that also makes sense  - it means going from 0 (male/control) to 1 (female/treatment). But if your region variable is coded like 1=north 2=south 3=east 4=west then "increase by one unit" doesn't mean anything. Of course the model doesn't know that, so it will still give you coefficient telling you what will happen when you "increase region by 1" but the result will be garbage with no actual interpretation. So to deal with a variable like this you need to transform it into a set of dummy variables (leaving one out as the reference category).
And to reiterate - everything I just said is true of any kind of model: OLS, logit, ologit, poisson, tobit, whatever. In general, for most question about how you treat independent variables in a model (using dummy variables, calculating interactions, transformations, etc.) the actual model type is irrelevant. The same basic dynamics apply for all models (although the details of how they impact the DV will differ)
