[jmgirard]

This follows #677 and #678. With the marginaleffects backend, some variable names cannot work, because marginaleffects uses them for its own columns. marginaleffects::predictions() renames columns named Predicted, SE, CI_low and CI_high (reported as vincentarelbundock/marginaleffects#1771). marginaleffects also writes or uses s.value, df, predicted* and rowid itself. Before, these names gave wrong estimates, SEs or CIs, missing columns, or unclear errors such as "missing value where TRUE/FALSE needed". estimate_means(), estimate_contrasts() and estimate_slopes() now stop with an error that names the variable and asks users to rename it:

• by variables of estimate_means(), and of estimate_contrasts() with a factor contrast: Predicted, SE, CI_low, CI_high, df, rowid, s.value.
• a factor contrast variable: SE, CI_low, CI_high, df, rowid.
• other predictors of estimate_means() and estimate_contrasts(): SE, CI_low, CI_high, df.
• by variables of estimate_slopes(): df, rowid, s.value, predicted, predicted_hi, predicted_lo.

The check is .check_marginaleffects_names() in R/checks.R. It runs in get_marginalmeans() and get_marginaltrends() before the marginaleffects call, and it skips Bayesian models. estimate_slopes() with a by variable named SE, CI_low or CI_high keeps working with the .1 suffix from #678.

Some clashes come from modelbased itself, and these now work:

• estimate_contrasts(by = "Level1") (or "Level2") gave two columns named Level1. The variable is now Level1.1.
• estimate_contrasts(by = "hypothesis") did not split the contrast labels into Level1 and Level2, because marginaleffects names its own column hypothesis.1.
• estimate_contrasts(m, "Parameter") failed, because the contrast labels were converted to the type of the Parameter variable and became NA.
• If <name>.1 is already a variable, the suffix is now the next free number (for example p.2). Before, by = c("p", "p.1") gave a column p.1 with the values of p.

Not fixed here: Bayesian models, inequality comparisons, contrasts of slopes (numeric contrast variable), and the trend and covariate roles of estimate_slopes(). Some of these also give wrong output on main.

Tests in test-focal-name-clashes.R:

• a loop over the 33 error cases that checks the message,
• a grid over 14 names and 6 roles that compares each other combination with the same call for a variable named safe_var,
• tests for Level1/Level2, hypothesis, Parameter and the free suffix.

Level1 and Level2 are removed from the "worked before" list for estimate_contrasts(by =). That test passed only because the duplicate column names matched the reference.

devtools::test() gives 0 failures and 0 errors locally. devtools::check() gives 0 errors and 0 warnings. The version is bumped to 0.17.0.14.

[Copilot]

🟢 Approval recommended

Targeted regression tests cover the changed behavior, and no unresolved issues were identified.

0 open findings

What changed in this PR

This PR makes the marginaleffects backend report variable-name collisions clearly instead of returning incorrect results or unclear errors. It also fixes several contrast-output name collisions.

Changes:

• Add checks for variable names that marginaleffects reserves or overwrites.
• Resolve contrast-column collisions and choose the next available suffix.
• Add regression tests, update NEWS, and bump the development version.

File	Description
tests/​testthat/​test-focal-name-clashes.R	Tests errors, valid names, and collision handling.
R/​get_marginaltrends.R	Checks names before calculating trends.
R/​get_marginalmeans.R	Checks names before calculating means and contrasts.
R/​format.R	Resolves contrast labels and focal-column suffixes.
R/​checks.R	Defines the name-collision check and error.
NEWS.md	Documents the behavior and remaining limitations.
DESCRIPTION	Bumps the development version.

🧠 Review effort: Lite

---

💡 Add a code-review agent skill or configure MCP servers for context-aware, tailored reviews. Learn more in the docs.