merTools 1.0.0

After a decade on CRAN, merTools 1.0.0 is the 1.0 long-term-support release.
It resolves the remaining open issues, fixes a correctness bug in
predictInterval() for nested random effects, repairs and extends the
shinyMer() explorer, adds a new plotREimpact() visualization, refreshes the
documentation, and tidies the test suite for low-maintenance, long-term use. The
public API is unchanged and existing results are preserved.

📖 Full changelog: NEWS.md · 📦 CRAN · 🌐 Package site

Highlights

• Validated against brms. A new
  Validating predictInterval() against brms
  vignette shows predictInterval() reproducing full-Bayesian posterior
  prediction intervals almost exactly (point estimates correlate 0.9998,
  matching 90% intervals and out-of-sample coverage) at a small fraction of the
  computational cost — the lme4 + predictInterval() path runs in about a second
  versus several minutes for a single brms fit.
• Correctness fix for nested random effects (#124). predictInterval() no
  longer returns seed-dependent point estimates when a prediction frame mixes
  observed and unobserved levels of an interaction grouping factor (e.g.
  (1 | a/b)). Observed-level predictions are bit-for-bit unchanged.
• New new.levels = "draw". For a group the model never saw, "draw"
  samples its effect from the estimated random-effect covariance — the analogue
  of brms::posterior_predict(allow_new_levels = TRUE). The default
  ("zero") is unchanged.
• New plotREimpact(), and plotFEsim() now highlights significant terms.
  Pass a named list of REimpact() results to overlay grouping factors on
  shared axes.
• shinyMer() revived and extended — a Model Summary tab and subset-based
  Random/Average draws.
• Refreshed docs — a new Exploring Contextual Effects vignette, a corrected
  Prediction Intervals vignette, and a proper citation("merTools").

New Features

• predictInterval() gains a new.levels argument ("zero" default, "draw")
  for unobserved groups, reflecting between-group uncertainty from VarCorr.
• plotREimpact() visualizes REimpact() output; named lists overlay grouping
  factors for direct comparison.
• plotFEsim() highlights fixed effects whose interval excludes the null.
• shinyMer() Model Summary tab (#78) and subset-restricted draws (#32).

Bug Fixes

• Reproducible predictInterval() for partially-observed nested / interaction
  random effects (#124).
• Restored a single RNG stream in predictInterval() (helper refactor had reset
  the seed mid-call); verified bit-for-bit against prior CRAN output.
• Correct handling of multiple random-effect term blocks per grouping factor —
  (x + y || g), (1|g) + (0 + x|g) (#118).
• averageObs() / findFormFuns() on matrix-LHS (cbind binomial) models (#83).
• Several shinyMer() "Substantive Effect" tab defects; migrated to
  DT::renderDT().
• Replaced bare subbars() with reformulas::subbars() for the lme4
  deprecation.

Documentation, tests & CI

• New Contextual Effects vignette (#136); rebuilt Prediction Intervals
  vignette and fixed its figure (#116); improved inst/CITATION (#137).
• Cross-version numeric regression harness
  (tests/comparisons/predictInterval-regression.R); pinned RNG version;
  disabled testthat parallelism for deterministic seeds.
• Bumped actions/checkout@v5; addressed the GitHub Actions Node.js 20
  deprecation.

---

1.0 marks the start of a long-term-support phase: stable API, with future
1.0.x releases focused on bug fixes, CRAN/dependency compatibility, and docs.
Thanks to Carl Frederick, Alex Whitworth, Ben Bolker, Davis Vaughan, and the
issue reporters — including @dotPiano, whose report led to the #124 fix.