R/lmodel2.R

In tidymodels/broom: Convert Statistical Objects into Tidy Tibbles

#' @templateVar class lmodel2
#' @template title_desc_tidy
#'
#' @param x A `lmodel2` object returned by [lmodel2::lmodel2()].
#' @template param_unused_dots
#'
#' @evalRd return_tidy(
#'   "term",
#'   "estimate",
#'   "p.value",
#'   "conf.low",
#'   "conf.high",
#'   method = "Either OLS/MA/SMA/RMA"
#' )
#'
#' @details There are always only two terms in an `lmodel2`: `"Intercept"`
#'   and `"Slope"`. These are computed by four methods: OLS
#'   (ordinary least squares), MA (major axis), SMA (standard major
#'   axis), and RMA (ranged major axis).
#'
#'   The returned p-value is one-tailed and calculated via a permutation test.
#'   A permutational test is used because distributional assumptions may not
#'   be valid. More information can be found in
#'   `vignette("mod2user", package = "lmodel2")`.
#'
#' @examplesIf rlang::is_installed(c("lmodel2", "ggplot2"))
#'
#' # load libraries for models and data
#' library(lmodel2)
#'
#' data(mod2ex2)
#' Ex2.res <- lmodel2(Prey ~ Predators, data = mod2ex2, "relative", "relative", 99)
#' Ex2.res
#'
#' # summarize model fit with tidiers + visualization
#' tidy(Ex2.res)
#' glance(Ex2.res)
#'
#' # this allows coefficient plots with ggplot2
#' library(ggplot2)
#'
#' ggplot(tidy(Ex2.res), aes(estimate, term, color = method)) +
#'   geom_point() +
#'   geom_errorbarh(aes(xmin = conf.low, xmax = conf.high)) +
#'   geom_errorbarh(aes(xmin = conf.low, xmax = conf.high))
#'
#' @export
#' @seealso [tidy()], [lmodel2::lmodel2()]
#' @aliases lmodel2_tidiers
#' @family lmodel2 tidiers
tidy.lmodel2 <- function(x, ...) {
  ret <- x$regression.results[c(1:3, 5)] %>%
    select(
      method = Method,
      Intercept,
      Slope,
      p.value = `P-perm (1-tailed)`
    ) %>%
    pivot_longer(
      cols = c(dplyr::everything(), -method, -p.value),
      names_to = "term",
      values_to = "estimate"
    ) %>%
    arrange(method, term)

  # add confidence intervals
  confints <- x$confidence.intervals %>%
    pivot_longer(
      cols = c(dplyr::everything(), -Method),
      names_to = "key",
      values_to = "value"
    ) %>%
    tidyr::separate(key, c("level", "term"), "-") %>%
    mutate(level = ifelse(level == "2.5%", "conf.low", "conf.high")) %>%
    tidyr::pivot_wider(
      id_cols = c(Method, term),
      names_from = level,
      values_from = value
    ) %>%
    dplyr::arrange(Method) %>%
    as.data.frame() %>%
    select(method = Method, term, conf.low, conf.high)

  ret %>%
    inner_join(confints, by = c("method", "term")) %>%
    # change column order so `p.value` is at the end
    select(-p.value, dplyr::everything()) %>%
    as_tibble()
}


#' @templateVar class lmodel2
#' @template title_desc_glance
#'
#' @inherit tidy.lmodel2 params examples
#'
#' @evalRd return_glance(
#'   "r.squared",
#'   "p.value",
#'   theta = "Angle between OLS lines `lm(y ~ x)` and `lm(x ~ y)`",
#'   H = "H statistic for computing confidence interval of major axis slope",
#'   "nobs"
#' )
#'
#' @export
#' @seealso [glance()], [lmodel2::lmodel2()]
#' @family lmodel2 tidiers
#'
glance.lmodel2 <- function(x, ...) {
  as_glance_tibble(
    r.squared = x$rsquare,
    theta = x$theta,
    p.value = x$P.param,
    H = x$H,
    nobs = stats::nobs(x),
    na_types = "rrrri"
  )
}