r - Recreate raw probabilities produced by multinom function -

i wanting use coefficients multi-nomial model create related probabilities manually each class of interest. way can score data observation within database using sql. using nnet package , working through example found.

this example uses hsbdemo data set , outcome variable program type attempts classify 3 classes- general/academic/vocation.

library("foreign") library("nnet") ml <- read.dta("http://www.ats.ucla.edu/stat/data/hsbdemo.dta") test <- multinom(prog2 ~ ses + write, data = ml) summary(test) 

now fits model , produces following coefficients:

## call: ## multinom(formula = prog2 ~ ses + write, data = ml) ##  ## coefficients:  ##          (intercept)  sesmiddle    seshigh      write ## general     2.852198 -0.5332810 -1.1628226 -0.0579287 ## vocation    5.218260  0.2913859 -0.9826649 -0.1136037 ##  ## std. errors: ##          (intercept) sesmiddle   seshigh      write ## general     1.166441 0.4437323 0.5142196 0.02141097 ## vocation    1.163552 0.4763739 0.5955665 0.02221996 ##  ## residual deviance: 359.9635  ## aic: 375.9635 

now determine associated probability general class have produced following equation using fitted coeffcients:

obs.linkval<- 2.852198+               -0.5332810*ifelse(as.character(ml$ses)=="middle",1,0)+               -1.1628226*ifelse(as.character(ml$ses)=="high",1,0)+               -0.0579287*ml$write obs.prob<- exp(obs.linkval)/(1+exp(obs.linkval)) 

however, when @ probabilities fitted model not match have calculated manually. below code return predicted probabilities fitted model. has attempted same?

predict(test, ml, "probs")