R/interpret.logbin.smooth.r

In logbin: Relative Risk Regression Using the Log-Binomial Model

interpret.logbin.smooth <- function(formula) {
  p.env <- environment(formula)
  tf <- terms.formula(formula, specials = c("B","Iso"))
  
  if (attr(tf, "intercept") != 1) stop("models without intercept are not supported by logbin.smooth")
  if (any(attr(tf,"order") > 1)) stop("models with interactions are not supported by logbin.smooth")
  if (attr(tf, "response") == 0) stop("missing response")
  
  terms <- attr(tf, "term.labels")
  vars <- attr(tf, "variables")
  nt <- length(terms)
  
  respvar <- attr(tf, "response")
  response <- as.character(vars[respvar+1L])
  full.formula <- fake.formula <- paste0(response, "~")
  
  Bp <- attr(tf, "specials")$B
  Isop <- attr(tf, "specials")$Iso
  off <- attr(tf, "offset")
  vtab <- attr(tf, "factors")
  
  ns <- length(Bp) + length(Isop)
  if(ns == 0) stop("formula does not include any semi-parametric terms. Use 'logbin' instead.")
  
  kp <- 1
  smooth.ind <- NULL
  smooth.spec <- list()
  for (i in seq_len(nt)) {
    varind <- which(as.logical(vtab[,i]))
    if ((varind %in% Bp) || (varind %in% Isop)) {
      st <- eval(parse(text = terms[i]), envir = p.env)
      full.newterm <- st$termlabel
      fake.newterm <- st$term
      smooth.ind <- c(smooth.ind, i)
      smooth.spec[[st$term]] <- st
    } else full.newterm <- fake.newterm <- as.character(vars[varind+1L])
    if (kp > 1) {
      full.formula <- paste0(full.formula, "+", full.newterm)
      fake.formula <- paste0(fake.formula, "+", fake.newterm)
    } else {
      full.formula <- paste0(full.formula, full.newterm)
      fake.formula <- paste0(fake.formula, fake.newterm)
    }
    kp <- kp + 1
  }
  
  if (!is.null(off)) {
    if (kp > 1)
      fake.formula <- paste0(fake.formula, "+")
    fake.formula <- paste0(fake.formula, paste(as.character(vars[off+1L]),collapse="+"))
    kp <- kp + 1
  }
    
  full.formula <- as.formula(full.formula, p.env)
  fake.formula <- as.formula(fake.formula, p.env)
  
  list(full.formula = full.formula, fake.formula = fake.formula, 
       smooth.spec = smooth.spec, smooth.ind = smooth.ind,
       terms = tf)    
}