Let's say I perform an ordinal regression analysis and I use 15 predictor variables of which 5 have turned out to be significant but each predictor has different levels as I have used ordinal variables as predictor variables (please see the image attached). How can someone interpret this?
In multiple regression analysis each variable (predictor) is shown to have a certain percentage of influence on the dependent variable but in ordinal regression a single variable has different levels (1-5) and only one or two of the levels are significant. For example, variable 1 has 2 levels that are significant. How can one interpret such a result? Thanks in advance.
1 Answer
Categorical predictors with multiple levels result in multiple parameters being estimated, and with just main effects of the predictor included in the model, each parameter estimate represents a contrast among a given level and the last level (in the parameterization used in the Ordinal Regression). Some of these might be significant while the omnibus test for the predictor might not be. Omnibus tests tend to reflect sort of averages of these individual effects. This is true here as it is in any linear or generalized linear model. There's nothing special about ordinal regression in this regard.