Fix r2() error for glmmTMB negative-binomial models without random effects

[raffaelemancuso]

Problem

r2() (and therefore model_performance()) errors for a glmmTMB
negative-binomial model (nbinom1 / nbinom2) that has no random effects:

library(glmmTMB)
m <- glmmTMB(count ~ x1 + x2, data = df, family = nbinom2(link = "log"))
performance::r2(m)
#> Error: `r2()` does not support models of class `glmmTMB` without random
#>   effects and from nbinom2-family with log-link-function.

Cause

For non-mixed models, r2.glmmTMB() walks an if / else if chain with branches
for the linear, logit, beta-binomial, binomial, Poisson, zero-inflated,
ordered-beta and beta families. There is no branch for the negative-binomial
family, so a non-mixed negbin model falls through to the final else, which
raises the error above.

Fix

Add a negative-binomial branch that returns McFadden's R² (mirroring the
existing beta-binomial branch). Nagelkerke's R² — used by the Poisson branch —
is unstable here: the null-model refit lands on a different dispersion parameter
and can return nonsensical negative values.

The branch is guarded with !info$is_zero_inflated so zero-inflated negbin
models keep routing to r2_zeroinflated().

} else if (info$is_negbin && !info$is_zero_inflated) {
  out <- r2_mcfadden(model)
}

Tests

Added a regression test in test-r2_mcfadden.R covering nbinom2 (which
previously errored) and nbinom1, checking that r2() agrees with
r2_mcfadden() and pinning the expected values.

Existing behaviour is unchanged and verified:

• Poisson without random effects still uses Nagelkerke's R².
• Mixed negbin models still use Nakagawa's marginal/conditional R².
• test-r2.R and test-r2_mcfadden.R pass in full.

[gemini-code-assist]

Caution

The consumer version of Gemini Code Assist on GitHub has been sunset. All code review activity has officially ceased.

[Copilot]

Pull request overview

Fixes performance::r2() for glmmTMB negative-binomial (nbinom1/nbinom2) models without random effects, which previously errored due to falling through the non-mixed glmmTMB family dispatch. The fix adds an explicit negative-binomial branch that uses McFadden’s pseudo-R² (via r2_mcfadden()), keeping zero-inflated models routed to r2_zeroinflated().

Changes:

• Add a non-mixed glmmTMB negative-binomial branch in r2.glmmTMB() to return McFadden’s R².
• Add a regression test covering nbinom2 (previously failing) and nbinom1.
• Document the behavior change in NEWS.md.

Reviewed changes

Copilot reviewed 3 out of 3 changed files in this pull request and generated no comments.

File	Description
R/r2.R	Adds a negative-binomial (non-zero-inflated) handling branch for non-mixed glmmTMB models, returning r2_mcfadden().
tests/testthat/test-r2_mcfadden.R	Adds regression coverage ensuring r2() works for glmmTMB nbinom1/nbinom2 without random effects and matches expected outputs.
NEWS.md	Notes that r2()/model_performance() no longer errors for these models and returns McFadden’s R².

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.