R/alr_results.R

In nstrayer/nviz:

# Takes an alternating logistic regression fit from the ordgee function
# and generates a nice table with plots of the coefficient values and their confidence intervals.
#' Visualize Results of Alternating Logistic Regressions.
#'
#' Takes output from the `ordgee` function in the package `geepack` and returns a table with visual results along with numerical values imposed on the graph
#'
#' @param model_fit an object returned by `ordgee`
#' @param title Text for the title portion of your results
#' @param included_models which model you want to show in results. Defaults to both mean and correlation. For mean enter "mean", correlation: "correlation".
#' @export
#' @examples
#' alr_results(my_fit, "Results from fitting alternating logistic regression", included_models = "both")
alr_results <- function(model_fit, title, included_models = "both"){
  model_summary <- summary(model_fit)

  #Grab the mean model estimates and standard errors
  mean_model_params <- model_summary$mean[,c(1,2)] %>%
    mutate(model_type = "mean model", variable = row.names(.))
  #Grab the correlation model's as well
  corr_model_params <- model_summary$correlation[,c(1,2)] %>%
    mutate(model_type = "correlation model", variable = row.names(.))

  if(included_models == "both"){
    #combine the two
    model_params <- rbind(mean_model_params, corr_model_params) %>%
      mutate(model_type = fct_inorder(model_type))

    our_subtitle = "The 'mean model' panel contains estimates of the effect of a given variable's one unit increase\n(holding all other variables constant) on the log-odds of event occuring at a given measurement.\n\nThe 'correlation model' panel shows a variable's effect on the log-odds of an event occuring given\nan event occured at another measurement.\n"
  } else if(included_models == "mean"){
    #combine the two
    model_params <- mean_model_params %>%
      mutate(model_type = fct_inorder(model_type))
    our_subtitle = "The 'mean model' panel contains estimates of the effect of a given variable's one unit increase\n(holding all other variables constant) on the log-odds of event occuring at a given measurement.\n"
  } else {
    model_params <- corr_model_params %>%
      mutate(model_type = fct_inorder(model_type))

    our_subtitle = "The 'correlation model' panel shows a variable's effect on the log-odds of an event occuring\ngiven an event occured at another measurement.\n"
  }


  ggplot(model_params, aes(y = estimate,x = variable, color = variable)) +
    geom_hline(yintercept = 0, alpha = 0.2)+
    geom_pointrange(aes(ymin = estimate - 1.96*san.se, ymax = estimate +1.96*san.se)) +
    geom_text(aes(label = sprintf("Est: %3.3f", estimate)),
              vjust = -1, size = 3)+
    geom_text(aes(label = sprintf("SE: %3.3f", san.se)),
              vjust = 1.8, size = 3)+
    facet_wrap(~model_type, scales = "free_y") +
    coord_flip() +
    scale_colour_brewer(palette="Paired") +
    labs(y = "coefficient estimate", title = title, subtitle = our_subtitle)
}