R/predictAt.R

In jon-mellon/mellonMisc: Lots of small things

#' @export predictAt
predictAt <- function(mod, at, multiply.out = T, se = T) {
  vars <- names(at)
  if(multiply.out) {
    at <- expand.grid(at)
  }
  at$prob <- NA
  if(se) {
    at$se <- NA
  }
  for(ii in 1:nrow(at)) {
    data <- model.frame(mod)
    if(any(!vars %in% names(data))) {
      stop("Variable not found in data")
    }
    for(jj in vars) {
      data[, jj] <- at[ii, jj]
    }
    pred.prob <- predict(mod, newdata = data, type = "response", se.fit = se)
    
    if("svyglm" %in%  class(mod)) {
      pred.prob <- dtf(fit = as.vector(pred.prob), 
                       se.fit = sqrt(attributes(pred.prob)$var))
    }
    
    if(!se) {
      pred.prob <- dtf(fit = pred.prob)
    }
    if(any(names(attributes(pred.prob))=="var") ){
      pred.prob <- dtf(fit = as.vector(pred.prob), se.fit = sqrt(attributes(pred.prob)$var))
    }
    
    
    at[ii, "prob"] <- Hmisc::wtd.mean(pred.prob$fit, weights = data$`(weights)`, na.rm = T)
    if(se) {
      at[ii, "se"]   <- Hmisc::wtd.mean(pred.prob$se.fit, weights = data$`(weights)`, na.rm = T)  
    }
    
  }
  if(se) {
    at$lci <- at$prob - 1.96 * at$se
    at$uci <- at$prob + 1.96 * at$se
  }
  return(at)
}