Vector of conditional probabilities from a multinomial logistic regression given particular covariate values

probs.logit.covs

probs.logit_covs <- function(groups, cov, data){
  ##
  ## given a (set of) covariate values, calculates the class membership probabilities
  ##
  ## groups: response field name of interest in NATSAL e.g. het1yr (string)
  ## cov: explanatory covariate name e.g. age_shift (string)
  ## data: e.g. NATSAL.dat (array)
  
  ## ordered sequence of covariate levels
  ## copy those covered in data but could include intermediate values too
  cov.seq <- sort(unique(data[,cov]))
  
  ## vector of group (class) labels for groups
  ## e.g. number of partners
  grps.seq <- sort(unique(data[,groups]))
  
  ## include only columns of interest
  ## and create an mlogit input format array
  head(data0 <- data[,c(groups, cov)])
  head(data.long <- mlogit.data(data0, shape="wide", choice=groups), 50)
  
  formula.cov <- as.formula(paste(groups, " ~ 0 | ", cov, sep=""))
  fit.mlogit <- mlogit(formula.cov, data=data.long)
  
  coeff <- fit.mlogit$coefficients
  
  ## calculate individual group probabilities
  ## for a given (set of) covariates
  ## p(X=k)
  probs.logit <- matrix(NA, ncol=length(grps.seq), nrow=length(cov.seq))
  colnames(probs.logit) <- grps.seq
  rownames(probs.logit) <- cov.seq
  
  j <- 0
  for (x in cov.seq){                       
    j <- j+1
    probs.logit[j,] <- probs.logit(grps.seq, cov, coeff, x)
  }
  
  probs.logit
}
## END FUNCTION ##