Warn for non-Bernoulli binomial models in r2_tjur(), r2_somers(), r2_efron() and r2_mckelvey()

[jmgirard]

r2(), r2_nagelkerke(), r2_coxsnell() and r2_mcfadden() already warn and return NULL for binomial models that are not Bernoulli models (#275). Four other R2 functions did not, and gave wrong values or unclear errors for a proportion response with weights or a cbind() response:

Function	Bernoulli	Proportion + weights	cbind()
r2_tjur()	0.273	0.299	0.299
r2_efron()	0.273	1	NA (base R warning)
r2_mckelvey()	0.312	0.035	0.035
r2_somers()	0.571	NA	error: "The names you entered for x and y are not available"

(Same mtcars model, am ~ factor(cyl), fit three ways.)

This PR adds the same guard and warning text to these four functions. For r2_efron() and r2_mckelvey(), the guard applies only to binomial models, so other families are not affected.

• New tests in tests/testthat/test-r2_binomial_aggregated.R: each function warns and returns NULL for both aggregated forms, stays silent with the expected value for the Bernoulli model, and r2_efron() stays silent for a Poisson model. The tests fail without the fix.
• NEWS entry and version bump to 0.18.2.5.

A correct R2 for aggregated binomial models (the lchoose() log-likelihood idea in the issue) is not part of this PR.

Fixes #165

[Copilot]

🔵 Needs a closer look

Exclude beta-binomial models from the Efron and McKelvey guards.

0 open findings

What changed in this PR

This PR adds warnings and NULL returns for non-Bernoulli binomial models across four R² measures.

Changes:

• Added guards to Tjur, Somers, Efron, and McKelvey R² functions.
• Added regression tests for aggregated and Bernoulli models.
• Updated NEWS and bumped the development version.

Review findings:

• R/r2_efron.R and R/r2_mckelvey.R should exclude beta-binomial models from the new guards.

File	Reviewed change
tests/​testthat/​test-r2_binomial_aggregated.R	Tests warnings, NULL results, and unaffected models.
R/​r2_tjur.R	Guards non-Bernoulli binomial models.
R/​r2_somers.R	Guards non-Bernoulli binomial models.
R/​r2_mckelvey.R	Adds a binomial-specific guard.
R/​r2_efron.R	Adds a binomial-specific guard.
NEWS.md	Documents the behavior change.
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.