get_variance(): include uncorrelated random-slope variances in var.random for lme4 models

[jmgirard]

get_variance() left the random-slope variance out of var.random for lme4 models with uncorrelated random effects.

library(lme4)
m <- lmer(Reaction ~ Days + (Days || Subject), data = sleepstudy)
insight::get_variance(m)$var.random
# before: 627.57 (random-intercept variance only)
# after: 1649.53 = vc[1] + vc[2] * mean(Days^2)

The correlated model (Days | Subject) gives 1698.08 for comparison.

Cause: lme4 splits (x || g) into several VarCorr() elements ("g", "g.1") and names nested terms with make.names() ("subgrp.grp"), but ranef() returns one element, "g" or "subgrp:grp". var.random selected the VarCorr() elements by the ranef() names. So it dropped the slope elements, and for nested terms such as (1 + Days || grp / subgrp) it gave an error inside .safe() and returned NULL. The new helper .vc_random_terms() maps each VarCorr() element to its ranef() term.

The fix for #304 (2021) corrected var.intercept and var.slope but not var.random. The test get_variance-7 recorded the old value 627.57, so this PR changes that expectation. A new test compares var.random with values computed by hand from VarCorr() for the plain, centered and nested cases.

This changes performance::r2_nakagawa() and performance::icc() for these models (easystats/performance#428).

[Copilot]

🟢 Approval recommended

No unresolved review issues were identified.

0 open findings

What changed in this PR

Fixes get_variance() for lme4 models so uncorrelated random-slope variances are included in var.random, including nested terms.

Changes:

• Maps VarCorr() components to ranef() terms.
• Adds regression coverage.
• Updates NEWS and the development version.

File	Description
tests/​testthat/​test-get_variance.R	Adds regression coverage and updates expectations.
R/​compute_variances.R	Maps split and nested variance components correctly.
NEWS.md	Documents the behavioral 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.