get_variance() Triggers Model Recompilation and Returns NULL for …

…`brms` Mixed Models in v0.20.3

Fixes #915

R/compute_variances.R

@@ -78,15 +78,20 @@
 
   # we also need necessary model information, like fixed and random effects,
   # variance-covariance matrix etc. for the null model
-  if (is.null(model_null)) {
-    model_null <- .safe(null_model(model, verbose = FALSE))
+  if (faminfo$is_linear && !faminfo$is_tweedie) {
+    # we don't need these for linear models
+    me_info_null <- NULL
+  } else {
+    if (is.null(model_null)) {
+      model_null <- .safe(null_model(model, verbose = FALSE))
+    }
+    me_info_null <- .get_variance_information(
+      model_null,
+      faminfo = faminfo,
+      name_fun = name_fun,
+      verbose = verbose
+    )
   }
-  me_info_null <- .get_variance_information(
-    model_null,
-    faminfo = faminfo,
-    name_fun = name_fun,
-    verbose = verbose
-  )
 
   # Test for non-zero random effects ((near) singularity)
   no_random_variance <- FALSE

R/get_modelmatrix.R

@@ -158,15 +158,24 @@ get_modelmatrix.svyglm <- function(x, ...) {
 #' @export
 get_modelmatrix.brmsfit <- function(x, ...) {
   formula_rhs <- safe_deparse(find_formula(x)$conditional[[3]])
-  formula_rhs <- stats::as.formula(paste0("~", formula_rhs))
-  # the formula used in model.matrix() is not allowed to have special functions,
-  # like brms::mo() and similar. Thus, we reformulate after using "all.vars()",
-  # which will only keep the variable names.
-  .data_in_dots(
-    ...,
-    object = stats::reformulate(all.vars(formula_rhs)),
-    default_data = get_data(x, verbose = FALSE)
-  )
+  # exception: for null-models, we need different handling, else `reformulate()`
+  # will not work.
+  if (identical(formula_rhs, "1")) {
+    mm <- get_data(x, verbose = FALSE)
+    mm[[1]] <- 1
+    colnames(mm)[1] <- "(Intercept)"
+    mm[1]
+  } else {
+    formula_rhs <- stats::as.formula(paste0("~", formula_rhs))
+    # the formula used in model.matrix() is not allowed to have special functions,
+    # like brms::mo() and similar. Thus, we reformulate after using "all.vars()",
+    # which will only keep the variable names.
+    .data_in_dots(
+      ...,
+      object = stats::reformulate(all.vars(formula_rhs)),
+      default_data = get_data(x, verbose = FALSE)
+    )
+  }
 }
 
 #' @export

tests/testthat/test-brms.R

@@ -900,3 +900,22 @@ test_that("get_modelmatrix", {
     )
   )
 })
+
+# get variance
+test_that("get_variance works", {
+  mdl <- brms::brm(mpg ~ hp + (1 | cyl), data = mtcars)
+  out <- get_variance(mdl)
+  expect_equal(
+    out,
+    list(
+      var.fixed = 5.0734440813087,
+      var.random = 22.887162197839,
+      var.residual = 14.9024565369,
+      var.distribution = 14.9024565369,
+      var.dispersion = 0,
+      var.intercept = c(cyl = 22.8871621978389)
+    ),
+    tolerance = 1e-4,
+    ignore_attr = TRUE
+  )
+})