I am interested why predicted probabilities for ordinal and multinomial regressions differ? Can we explain this difference in terms of multinomial coefficients? And what we can conclude from this difference?

Thanks