Added ordering for dplyr locale issue.

R/variance-linear.R

@@ -12,9 +12,10 @@ vcov_sr_diag <- function(data, mod, residual=NULL){
   if(is.null(residual)){
     residual <- stats::residuals(mod)
   }
-
-  result <- mod$data %>%
-    dplyr::mutate(resid=residual) %>%
+  result <- dplyr::tibble(
+    resid=residual,
+    treat=data$treat
+  ) %>%
     dplyr::group_by(.data$treat) %>%
     dplyr::summarize(se=stats::sd(.data$resid), .groups="drop") %>%
     dplyr::mutate(se=.data$se*sqrt(1/data$pie))

tests/testthat/test-factor-ordering.R

@@ -0,0 +1,55 @@
+library(MASS)
+library(dplyr)
+
+DATA <- RobinCar:::data_sim
+DATA$A <- as.factor(DATA$A)
+DATA$y_bin <- ifelse(DATA$y > 2, 1, 0)
+
+# Ensure that the naming/capitalization
+# of the treatment variable does not impact results.
+
+DATA <- DATA %>% mutate(
+  A2=paste0("TREAT", A),
+  A3=ifelse(A2 == "TREAT1", "treat1", A2)
+)
+
+# Doing DATA %>% group_by(A3) %>% summarize(n()) will
+# order the levels by c("TREAT0", "TREAT2", "treat1")
+#
+# whereas factor(DATA$A3) will order the levels
+# by c("TREAT0", "TREAT1", "TREAT2").
+
+test_that("GLM full function -- linear (ANOVA)", {
+
+  # The locale is actually the issue -- testthat
+  # automatically sets locale C but the issue only
+  # appears under locale != C. So here we set it to
+  # en_US.UTF-8.
+
+  original_locale <- Sys.getlocale("LC_COLLATE")
+  Sys.setlocale("LC_COLLATE", "en_US.UTF-8")
+
+  # English language ordering
+  langENG <- robincar_glm(
+    df=DATA,
+    response_col="y",
+    treat_col="A2",
+    car_scheme="simple",
+    formula="y ~ A2 + x1")
+
+  # C language ordering
+  langC <- robincar_glm(
+    df=DATA,
+    response_col="y",
+    treat_col="A3",
+    car_scheme="simple",
+    formula="y ~ A3 + x1")
+
+  expect_equal(langENG$result$estimate, langC$result$estimate)
+  expect_equal(c(langENG$varcov), c(langC$varcov))
+
+  # Reset the locale
+  Sys.setlocale("LC_COLLATE", original_locale)
+
+})
+