I am using a multilevel regression model with a multinomial outcome (i.e., three unordered categories) as dependent variable, assigned the numeric values 0, 1, and 2. 0 being the reference category.  The model can be seen below:  

[![enter image description here][1]][1]

My question is: Is it possible to compare the odds ratio effect size for Variable 1 on both outcomes when having an multinomial dependent variabe? I have seen people deriving an odds ratio by dividing them (e.g.,  61.79 / 9.51 = 6.49 OR, Final model), meaning that having Outcome 2 is 6.49 times the odds higher than having Outcome 1, when one is exposed to Variable 1, compared to the reference category. 

Is this legitimate? I simply can't find any literature backing this up. 


  [1]: https://i.sstatic.net/GaTzF.png