Multinomial Regression - predicting probabilities
I have responses to a questionnaire on patient experience. My aim is to add a component of this questionnaire (on communication) to an already established questionnaire (so adding a domain).
The responses to the original questionnaire I have are in the following format:
I had a good experience with the consultant = agree
I felt I was listened to = agree completely
The communication experience was positive = agree
I felt I was taken care ok = so-so
The possible responses for each of the 4 questions within the communication domain are agree completely, agree, so-so, disagree, disagree completely. I therefore have for each patient a combination of the 4 responses. For a sub set of patients I also showed them a health state (as described by the 4 statements above) and asked them, to rate this as very poor, poor, average, good or very good.
I therefore have a subset of the health states as described by the 4 statements, with an overall rating for this health state. I ran multinomial regression on the data in order to determine whether the responses from the questionnaire could be used to predict an overall rating for the health state.
I need to be able to predict a rating for every combination of responses (54). I ran the analysis in SPSS. I have been trying to work out the equations behind the multinomial regression and co-efficients from the multinomial regression. However, the equation’s seem very complicated.
Can anyone advise, or does anyone know of some existing SPSS coding which can be applied?
Re: Multinomial Regression - predicting probabilities
Are you aware of regression models for categorical variables? Basically you use a tonne of dummy variables that take on a 0 or 1 value and if you only have categorical variables, then you do what is known as a logistic regression.
If you want to know more details then do a search in google for logistic regression, its derivation, and the estimation of parameters and what they mean.