I ran a multinomial logistic regression on some likart data (1-5 survey response questions). For some background, the survey was sent to undergraduate students to rate their TA on various attributes. Three of the variables are whether the TA was trained, whether the TA is international, AND an interaction term trained*international. For a particular comparison, I have odds ratio point estimates of 2.8, 6.9, and .5 respectively. I'm fairly certain the 2.8 indicates that being trained means you are 2.8 times more likely to not be in the reference group. Likewise, 6.9 times more likely to not be in the reference group if international. But how do I interpret the last coefficient?