R/meas_effect_uncond.R

In FrankLef/fciR: Data and functions used with Fundamentals of Causal Inference by Babette Brumback

#' Compute estimates of the unconditional association measures
#'
#' Compute estimates of the unconditional association measures.
#'
#' Compute estimates of the unconditional association measures,
#' hence the u in \code{bootu()}, and their confidence intervals.
#' 1- We use Gaussian, Poisson and Binomial glm to solve the
#'    estimating equations of the measures, not their distributions
#' 2- The association measures' distributions (ci) are estimated by
#'    bootstrapping
#' Assumptions: We assume that (3.2) holds but not (3.1)
#' See p. 45 and 46 for more details.
#' @inheritParams meas_effect_cond
#'
#' @importFrom stats glm coef plogis
#'
#' @source Section 3.3
#'
#' @return Dataframe in a useable format for \code{rsample::bootstraps}.
#' @export
meas_effect_uncond <- function(data, formula = Y ~ `T`) {

  x0 <- "(Intercept)"  # name of intercept used by lm, glm, etc.
  outcome.name <- all.vars(formula[[2]])  # the outcome variable
  input.names <- all.vars(formula[[3]])  # the independent variables


  # estimate the conditional probabilities
  coefs <- coef(glm(formula = formula, family = "gaussian", data = data))

  p0 <- coefs[x0]
  p1 <- sum(coefs)
  # estimate the risk difference
  RD <- p1 - p0

  # use loglinear model to estimate the log relative risk
  coefs <- coef(glm(formula = formula, family = "poisson", data = data))
  logRR <- coefs[input.names]

  # prepare data to estimate the log other relative risk
  ystar <- 1 - data[, outcome.name]
  xstar <- 1 - data[, input.names]

  # use loglinear model to estimate the log other relative risk
  coefs <- coef(glm(ystar ~ xstar, family = "poisson"))
  logRRstar <- coefs[2]

  # use logistic model to estimate the log of other risk
  coefs <- coef(glm(formula = formula, family = "binomial", data = data))
  logOR <- coefs[input.names]

  # return the results
  out <- c(p0, p1, RD, logRR, logRRstar, logOR)
  data.frame(
    term = c("P0", "P1", "RD", "logRR", "logRR*", "logOR"),
    estimate = unname(out),
    std.err = NA_real_
  )
}


#' @rdname meas_effect_uncond
#' @export
bootu <- meas_effect_uncond