From dff29ec45f416e15d59174ce306d553c81bbbd21 Mon Sep 17 00:00:00 2001
From: Jeffrey Girard <me@jmgirard.com>
Date: Thu, 15 Oct 2026 23:34:32 -0500
Subject: [PATCH] Fix estimate_contrasts() error for multivariate models with
 many observations

.split_contrast_response_levels() turned a data frame or matrix response
into one regex pattern per column of observed values, which fails to
compile when the response has many observations. It now uses the column
names of the response.

Fixes #682
---
 DESCRIPTION                |  2 +-
 NEWS.md                    |  5 +++++
 R/format.R                 |  6 +++++-
 tests/testthat/test-brms.R | 30 ++++++++++++++++++++++++++++++
 4 files changed, 41 insertions(+), 2 deletions(-)

diff --git a/DESCRIPTION b/DESCRIPTION
index 5f0fa75a8..9acbbdbbe 100644
--- a/DESCRIPTION
+++ b/DESCRIPTION
@@ -1,7 +1,7 @@
 Type: Package
 Package: modelbased
 Title: Estimation of Model-Based Predictions, Contrasts and Means
-Version: 0.17.0.15
+Version: 0.17.0.16
 Authors@R:
     c(person(given = "Dominique",
              family = "Makowski",
diff --git a/NEWS.md b/NEWS.md
index 5573cca43..2434879e2 100644
--- a/NEWS.md
+++ b/NEWS.md
@@ -116,6 +116,11 @@
   x-axis, e.g. for predictions from intercept-only models, or for
   `estimate_prediction()` without the `by` argument.
 
+* Fixed an "invalid regular expression" error in `estimate_contrasts()` for
+  multivariate models (for example from *brms*) with many observations.
+  `estimate_contrasts()` now uses the names of the response variables, not
+  their observed values, to find the response in the contrast labels.
+
 # modelbased 0.17.0
 
 ## Breaking Changes
diff --git a/R/format.R b/R/format.R
index 17a253616..6d130db5c 100644
--- a/R/format.R
+++ b/R/format.R
@@ -624,7 +624,11 @@ format.marginaleffects_contrasts <- function(
   # the response levels (e.g., factor level order), so that "Response1" and
   # "Response2" use a meaningful, consistent level order
   response <- .safe(insight::get_response(model, verbose = FALSE))
-  if (is.factor(response)) {
+  # for multivariate models, the response is a data frame or matrix, and the
+  # response levels are its column names, not the observed values
+  if (is.matrix(response) || is.data.frame(response)) {
+    response_levels <- colnames(response)
+  } else if (is.factor(response)) {
     response_levels <- levels(response)
   } else {
     response_levels <- unique(as.character(response))
diff --git a/tests/testthat/test-brms.R b/tests/testthat/test-brms.R
index 8fe9dca99..681bb014e 100644
--- a/tests/testthat/test-brms.R
+++ b/tests/testthat/test-brms.R
@@ -87,6 +87,36 @@ test_that("estimate_means - brms, multivariate", {
 })
 
 
+test_that("estimate_contrasts - brms, multivariate, many observations", {
+  # the response values must not be used as response levels, else the
+  # regular expressions built from them grow too long and fail
+  set.seed(123)
+  d <- data.frame(
+    y1 = rnorm(600),
+    y2 = rnorm(600),
+    g = factor(sample(c("a", "b", "c"), 600, replace = TRUE))
+  )
+  m <- suppressWarnings(brms::brm(
+    brms::mvbind(y1, y2) ~ g,
+    data = d,
+    refresh = 0,
+    iter = 500,
+    chains = 1,
+    seed = 123
+  ))
+
+  out <- estimate_contrasts(m, "g", resp = "y1")
+  expect_identical(nrow(out), 3L)
+  expect_identical(as.character(out$Level1), c("b", "c", "c"))
+  expect_identical(as.character(out$Level2), c("a", "a", "b"))
+
+  out <- estimate_contrasts(m, "g")
+  expect_identical(nrow(out), 15L)
+  expect_true(all(c("Response1", "Response2") %in% colnames(out)))
+  expect_identical(levels(out$Response1), c("y1", "y2"))
+})
+
+
 test_that("estimate_means - brms, Wiener", {
   skip_if_not_installed("curl")
   skip_if_offline()