R/append_values.R

In mcguinlu/triangulate: What the Package Does (One Line, Title Case)

#' Title
#'
#' @param dat Bias dataframe in long format
#' @param values Distributions of bias for different risk of bias levels
#' @param common Should a single set of distributions be used across all domains
#'   (default is TRUE)?
#'
#' @export

tri_append_bias <- function(dat, values, common = T) {

  if (!all(dat$d %in% c("left","right","unpredictable","none"))) {
    stop(paste0("Not all values in the direction column - d - are left/right/unpredictable/none. Did you forget to pass your data through tri_absolute_direction? Values are:",
                unique(dat$d)))
  }


  values <- values %>%
    janitor::clean_names() %>%
    dplyr::mutate(dplyr::across(dplyr::starts_with("bias_"), as.double)) %>%
    dplyr::mutate(dplyr::across(c(domain,j), stringr::str_to_lower))

  # Define criteria on which to join with values#
  # If common = T, means that the values of Serious/Moderate are consistent across domains
  if (common == T) {
    by = c("j")

    if ("domain" %in% colnames(values)) {
      values <- dplyr::select(values, -domain)
    }

  } else {
    by = c("domain","j")
  }

  dat %>%
    janitor::clean_names() %>%

    # Add basic values
    dplyr::left_join(values, by = by) %>%

    # Set values to 0 when judgement is NA
    # Most common when one tool has less domains than another
    dplyr::mutate(dplyr::across(dplyr::matches("bias_._add|bias_._prop"),
                                ~ dplyr::case_when(is.na(j) == T ~ 0,
                                                   j == "NA" ~ 0,
                                                   T ~ .))) %>%

    # Set additive biases mean/var to 0 when type is "prop"
    dplyr::mutate(dplyr::across(dplyr::matches("bias_._add"),
                                ~ dplyr::case_when(t == "prop" ~ 0,
                                                   T ~ .))) %>%

    # Set proportional biases mean/var to 0 when type is "add"
    dplyr::mutate(dplyr::across(dplyr::matches("bias_._prop"),
                                ~ dplyr::case_when(t == "add" ~ 0,
                                                   T ~ .))) %>%

    # Adjust signs of bias based on d (direction)
    # Where d is unpredictable, set mean to 0, but keep variance estimate
    dplyr::mutate(
      bias_m_add = dplyr::case_when(d == "left" ~ bias_m_add * -1,     #changed
                                    d == "unpredictable" ~ 0,
                                    T ~ bias_m_add),
      bias_m_prop = dplyr::case_when(d == "left" ~ bias_m_prop * -1,   #changed
                                     d == "unpredictable" ~ 0,
                                     T ~ bias_m_prop)
    ) %>%

    return()
}



#' Title
#'
#' @param dat Indirectness dataframe in long format
#' @param values Distributions of bias for different risk of indirectness levels
#' @param common Should a single set of distributions be used across all domains
#'   (default is TRUE)?
#'
#' @export

tri_append_indirect <- function(dat, values, common = T) {

# Define criteria on which to join with values#
# If common = T, means that the values of Serious/Moderate are consistent across domains
if (common == T) {
  by = c("j")

  if ("domain" %in% colnames(values)) {
    values <- dplyr::select(values, -domain)
  }

} else {
  by = c("domain","j")
}

dat %>%
  # Add basic values
  dplyr::left_join(values, by = by) %>%

  # Set values to 0 when judgement is NA
  # Most common when one tool has less domains than another
  dplyr::mutate(dplyr::across(dplyr::matches("ind_._add|ind_._prop"),
                              ~ dplyr::case_when(is.na(j) == T ~ 0,
                                                 j == "NA" ~ 0,
                                                 T ~ .))) %>%

  # Set additive biases mean/var to 0 when type is "prop"
  dplyr::mutate(dplyr::across(dplyr::matches("ind_._add"),
                              ~ dplyr::case_when(t == "prop" ~ 0,
                                                 T ~ .))) %>%

  # Set proportional biases mean/var to 0 when type is "add"
  dplyr::mutate(dplyr::across(dplyr::matches("ind_._prop"),
                              ~ dplyr::case_when(t == "add" ~ 0,
                                                 T ~ .))) %>%

  # Adjust signs of bias based on d (direction)
  # Where d is unpredictable, set mean to 0, but keep variance estimate
  dplyr::mutate(
    ind_m_add = dplyr::case_when(d == "left" ~ ind_m_add * -1,
                                 d == "unpredictable" ~ 0,
                                 T ~ ind_m_add),
    ind_m_prop = dplyr::case_when(d == "left" ~ ind_m_prop * -1,
                                  d == "unpredictable" ~ 0,
                                  T ~ ind_m_prop)
  ) %>%

  return()
}