vcov.lmerMod(full = TRUE) and
estfun.lmerMod() returned the random effect (co)variance
results in an order that did not match their labels when a random effect
block was 3x3 or larger, so results depended on the order in which
random effects were entered in the model formula (Github issue #2,
reported by James Pustejovsky). The parameter index matrix is now built
from Lambdat, which Lind actually indexes,
instead of Lambda.
vcov.lmerMod(full = TRUE, information = "observed")
failed for REML fits because the internal Hessian entries were not
coerced to numeric.
New regression tests covering Github issues #2 and #3, and updated lavaan model syntax (Github issue #4, reported by Yves Rosseel).
Documentation fixes, including the bread.glmerMod()
man page (contributed by Ben Bolker).
Suggested packages (mirt, lmeInfo, nlme, smdata) are now used conditionally in tests, and the duplicated Suggests field in DESCRIPTION was merged.
estfun.glmerMod() and llcont.glmerMod()
failed with “subscript out of bounds” when the cluster id variable was
not a consecutive integer sequence starting at 1 (e.g., character ids or
arbitrary numeric ids; Github issue #3, reported by teindor). Cluster
rows are now looked up by name rather than by coerced numeric
index.
Internal class(x) == "..." comparisons were replaced
with inherits(), avoiding “condition has length > 1”
errors on R >= 4.2.0 for models whose family was specified as a
function.
estfun.glmerMod() support
for binomial and poisson families fit with glmer(), with
scores of cluster-wise log-likelihoods computed via adaptive
Gauss-Hermite quadrature.