# Interpretation of Multinominal Logistic Regression coefficients

I am struggling to understand my Multinominal Logistic Regression. This is my first time ever tackling such a model. Note that I was following this recipe.

I am trying to predict the redemption rate (shareholders that wish to refund their money from their SPAC investment) with the predictors shown below. The dependent variable Y is redemption rate ranges from 0-1, with 1 being really bad and all refunded their money. It is split up in three groups; 1 low redemption rate, 2 middle and 3 high.

I am familiar with OLS and log interpretation but struggle to understand this model.

Could anyone shine in with guidance on the coefficients? Like, for one unit increase in Total Assets, how will that affect the group low redemption rate relative to the middle?

The model was executed in R.

Sector is a dummy for the Healthcare sector and profitability is a dummy for if they were profitable before their merger • Why has the redemption rate been converted to a categorical variable? This means loss of information about the outcome. May 11 at 10:29
• @dipetkov multinom function from nnet automatically convert them into categories, I don't know why. I was left with over 100 categories if not converting them into 3 categories myself. However it is not an issue, as I just want to predict them in such a way regardless. Despite the loss of information, it still managed to predict with 80.77% accuracy. May 11 at 10:36
• If your baseline category is "middle", the log odds of being in category "low" instead of "middle" increase by 0.176 with a unit increase in total assets May 11 at 10:38
• I would consider it an issue. Why are you using a model for categorical variable Y to analyze a numeric Y? It is as if the nnet::multinom model is giving you hints not to use it. May 11 at 10:39
• @TomHoel the reason it gave you so many categories is because logistic regression is a categorical model. Basic logistic regression gives you the probability that an event occurs or does not occur, and your outcome is binary. Multinomial regression is an extension of that to more than 2 categories. The idea is to model the probability of being in each category given a certain set of predictors. The function assumed each of your 100+ response measurements was a category, because they're all different May 11 at 10:47