R/explore_bivariate.R

In rgardiner90/researchr: Automating Basic Analyses for Social Science Researchers

#' Exploring biviate regression results of a dataframe
#'
#' This function returns either a graph or table of bivariate regression results for a dataset.
#' The results are helpful when trying to first gain an understanding of relationships between
#' a large number of variables.
#'
#' @param data A matrix, data frame, or tibble.
#' @param dependent A string character of the dependent variable
#' @param independent A string vector of the different independent variables to test
#' @param control A string character of one control variable that can be used
#' @param p_value The significance threshold for a table output (defaults to 0.05). Graph is
#' fixed at 95 percent confidence intervals
#' @param type Specifies the type of output request.  either 'graph' or 'table'
#' @param model_type Specificies whether you have an OLS or logit model
#'
#' @importFrom magrittr %>%
#'
#' @export
#'
#' @examples
#' explore_bivariate(data = mtcars, dependent = "mpg", independent = c("cyl", "disp", "hp", "drat"), control = "gear", p_value = 0.05, type = "table")
#' explore_bivariate(mtcars, names(mtcars)[1], names(mtcars)[-1])
explore_bivariate <- function(data, dependent, independent, control = FALSE,
                              p_value = 0.05, type = "graph",
                              model_type = "ols") {

  # getting the names of the independnet variables
  independent_text <- independent
  independent_variable <- data[, independent_text]

  # setting the dv
  dependent_text <- dependent
  dependent_variable <- data[, dependent_text]

  # setting the iv list
  iv_list <- independent_text

  # determining whether ols or logit
  modelType <- ifelse(tolower(model_type) == "ols", "gaussian",
                      ifelse(tolower(model_type) == "logit", "binomial",
                             "please select either 'ols' or 'logit'"))

  if(control == FALSE) {


    # running the models
    models <- lapply(iv_list, function(x) {
      broom::tidy(glm(substitute(dv ~ i, list(dv = as.name(dependent_text),
                                              i = as.name(x))),
                      data = data,
                      family = modelType))
    })


    # getting output
    if(type == "graph") {

      data <- models %>%
        tibble::tibble() %>%
        tidyr::unnest(cols = c(.)) %>%
        dplyr::filter(term != "(Intercept)") %>%
        dplyr::select(term, estimate, std.error) %>%
        dplyr::mutate(lower = (estimate - (std.error * 1.96)),
                      upper = (estimate + (std.error * 1.96)),
                      term = forcats::fct_reorder(term, estimate),
                      significance = ifelse(estimate > 0 & lower > 0, "positive",
                                            ifelse(estimate < 0 & upper < 0,
                                                   "negative", "not significant")))

      cols <- c("negative" = "red2", "not significant" = "gray",
                "positive" = "#7CAE00")

      results <- data %>%
        ggplot2::ggplot(ggplot2::aes(x = term, y = estimate,
                                     ymin = lower, ymax = upper, color = significance)) +
        ggplot2::theme_minimal() +
        ggplot2::geom_hline(yintercept = 0.0, color = "red", lty = 2) +
        ggplot2::geom_point() +
        ggplot2::geom_linerange() +
        ggplot2::labs(title = "Results are from bivariate tests, not a single model.",
                      caption = "Graph results show a 95% confidence interval",
                      x = "", y = "Coefficient") +
        ggplot2::coord_flip() +
        ggplot2::scale_color_manual(values = cols)

      return(results)

    } else if(type == "table") {

      results <- models %>%
        tibble::tibble() %>%
        tidyr::unnest(cols =c(.)) %>%
        dplyr::filter(term != "(Intercept)") %>%
        dplyr::mutate(significance = ifelse(p.value < p_value, 1, 0))

      return(results)
    } else {
      return("Please select either 'graph' or 'table' for type")
    }

  } else {


    control_text <- control
    control_variable <- data[, control_text]


    # running the models
    models <- lapply(iv_list, function(x) {
      broom::tidy(glm(substitute(dv ~ i + control_var,
                                 list(dv = as.name(dependent_text),
                                      control_var = as.name(control_text),
                                      i = as.name(x))),
                      data = data,
                      family = modelType))
    })

    # getting output
    if(type == "graph") {

      data <- models %>%
        tibble::tibble() %>%
        tidyr::unnest(cols = c(.)) %>%
        dplyr::filter(term != "(Intercept)",
                      term != control) %>%
        dplyr::select(term, estimate, std.error) %>%
        dplyr::mutate(lower = (estimate - (std.error * 1.96)),
                      upper = (estimate + (std.error * 1.96)),
                      term = forcats::fct_reorder(term, estimate),
                      significance = ifelse(estimate > 0 & lower > 0, "positive",
                                            ifelse(estimate < 0 & upper < 0,
                                                   "negative", "not significant")))

      cols <- c("negative" = "red2", "not significant" = "gray",
                "positive" = "#7CAE00")

      results <- data %>%
        ggplot2::ggplot(ggplot2::aes(x = term, y = estimate,
                                     ymin = lower, ymax = upper, color = significance)) +
        ggplot2::theme_minimal() +
        ggplot2::geom_hline(yintercept = 0.0, color = "red", lty = 2) +
        ggplot2::geom_point() +
        ggplot2::geom_linerange() +
        ggplot2::labs(title = "Results are from bivariate tests, not a single model.",
                      caption = "Graph results show a 95% confidence interval",
                      x = "", y = "Coefficient") +
        ggplot2::coord_flip() +
        ggplot2::scale_color_manual(values = cols)

      return(results)

    } else if(type == "table") {

      results <- models %>%
        tibble::tibble() %>%
        tidyr::unnest(cols =c(.)) %>%
        dplyr::filter(term != "(Intercept)",
                      term != control) %>%
        dplyr::mutate(significance = ifelse(p.value < p_value, 1, 0))

      return(results)
    } else {
      return("Please select either 'graph' or 'table' for type")
    }


  }
}