tests/testthat/test-fitlm.R

In metR: Tools for Easier Analysis of Meteorological Fields

context("Fitlm")


n <- 100
x <- rnorm(n)
y <- rnorm(n)
z <- rnorm(n)
good_fit <- lm(x ~ y + z)


test_that("regression works", {
  sim_fit <- coef(good_fit)
  sim_fit <- list(term = names(sim_fit), estimate = unname(sim_fit))

  expect_equal(FitLm(x, y, z), sim_fit)

  summ <- summary(good_fit)

  sim_fit <- c(
    sim_fit,
    list(
      std.error = unname(summ$coefficients[, 2]),
      df = rep(summ$df[2], 3),
      r.squared = rep(summ$r.squared, 3),
      adj.r.squared = rep(summ$adj.r.squared, 3)
    )
  )

  expect_equal(FitLm(x, y, z, se = TRUE), sim_fit)

  expect_equal(ResidLm(x, y, z), unname(residuals(good_fit)))
})


test_that("special cases work", {
  expect_equal(FitLm(x, rnorm(n), z = rnorm(n))$term[-1], c("V1", "z"))

  expect_equal(FitLm(rnorm(1), rnorm(1))$estimate, c(NA_real_, NA_real_))

  x[-1] <- NA
  expect_equal(FitLm(x, y)$estimate, c(NA_real_, NA_real_))
})


test_that("weighted, regression works", {
  weights <- runif(n, 0.1, 1)
  good_fit <- lm(x ~ y + z, weights = weights)

  sim_fit <- coef(good_fit)
  sim_fit <- list(term = names(sim_fit), estimate = unname(sim_fit))

  expect_equal(FitLm(x, y, z, weights = weights), sim_fit)

  summ <- summary(good_fit)

  sim_fit <- c(
    sim_fit,
    list(
      std.error = unname(summ$coefficients[, 2]),
      df = rep(summ$df[2], 3),
      r.squared = rep(summ$r.squared, 3),
      adj.r.squared = rep(summ$adj.r.squared, 3)
    )
  )

  expect_equal(FitLm(x, y, z, se = TRUE, weights = weights), sim_fit)

  expect_equal(ResidLm(x, y, z, weights = weights), unname(residuals(good_fit)))
})


test_that("Supports NAs", {
  y[10] <- NA
  good_fit <- lm(x ~ y + z)

  sim_fit <- coef(good_fit)
  sim_fit <- list(term = names(sim_fit), estimate = unname(sim_fit))

  expect_equal(FitLm(x, y, z), sim_fit)

  summ <- summary(good_fit)

  sim_fit <- c(
    sim_fit,
    list(
      std.error = unname(summ$coefficients[, 2]),
      df = rep(summ$df[2], 3),
      r.squared = rep(summ$r.squared, 3),
      adj.r.squared = rep(summ$adj.r.squared, 3)
    )
  )

  expect_equal(FitLm(x, y, z, se = TRUE), sim_fit)

  expect_length(ResidLm(x, y, z), length(y))
  expect_equal(
    as.vector(na.omit(ResidLm(x, y, z))),
    unname(residuals(good_fit))
  )
})