method = "blmm" now defaults to test = "t.satterthwaite.adaptive". Each
Monte Carlo draw refines its own covariance fit from a shared starting
point (at most eight steps) and gets its own Satterthwaite degrees of
freedom. Draws that fail the numerical checks are refit with lmerTest.
BLMM calls that omit test will give different estimates, standard
errors, df and p-values than in earlier versions.test = "t.satterthwaite"
(the shared-anchor approximation; cheaper than the adaptive default).test = "t.residual", which reproduces the BLMM default of 1.4.0 and
earlier, is deprecated and warns once per aldex() call.lmerTest.vignette("ALDEx3-mixed-effects")) now
describes the adaptive method and its Satterthwaite df, with the
oral_mouthwash_data example, contrasts and measured runtimes.Versions 1.5.0–1.6.2 were not released on CRAN. Compared with 1.4.0:
contrasts argument to aldex() for testing linear combinations of
fixed effects; results are stored in fit$contrasts.summary() includes requested contrasts with the coefficients and adds a
type column to tell them apart. Use component = "coefficients" or
"contrasts" to select one.aldex.p.adjust() adjusts several coefficients or contrasts as one
multiple-testing family without refitting.df (per-draw degrees of freedom) is now included in return.pars by
default.test = NULL resolves per engine, and incompatible method/test
combinations now fail before sampling.Matrix (>= 1.6-0) is now a runtime dependency.test="t.satterthwaite.adaptive": per-draw centering,
at most eight safeguarded covariance iterations, local Satterthwaite
inference, and exact rescue when numerical checks fail. The shared-anchor
default remains unchanged.NA p-values. Missing inference
propagates through Monte Carlo aggregation; fitting errors still stop.aldex.p.adjust(fit, groups = c("age", "genotype")) to adjust jointly
across all features of selected coefficients or stored contrasts, without
refitting. A list of character vectors specifies several disjoint families;
unlisted rows retain separate families. Coefficients and contrasts can share
a family; qualified names distinguish rows with the same name.p.adjust.method and onesided in new fits. For older fits lacking
these settings, the helper defaults to BH and two-sided inference; pass
nondefault settings explicitly. The returned object records its adjustment
families, and summary() uses the updated results.lmerTest are unchanged.Matrix is now an explicit runtime dependency (>= 1.6-0).ALDEx3.blmm.backend ("auto",
"dense", or "sparse") and ALDEx3.blmm.memory (estimated working bytes)
options support reproducible validation. An optional
ALDEx3.blmm.diagnostics callback receives the selected backend and its
rationale once per BLMM invocation or streaming chunk. Memory estimates
include worker and automatic-differentiation (AD) costs; they are not a hard
process-memory limit.streamsize documentation to describe the existing heuristic of
ten times the raw array size (the default corresponds to about 800 MB of response
draws). Streaming behavior is unchanged; its threshold is not a hard memory
limit, and changing BLMM chunk boundaries can change approximate results.summary(fit) now includes coefficients and any requested contrasts, with a
type column when combining them. Explicit component = "coefficients" and
component = "contrasts" filters remain available. Fits without requested
contrasts retain the previous column schema. Intercept filtering applies only
to coefficients; stored results and multiple-testing adjustments are unchanged.method = "blmm" now defaults to test = "t.satterthwaite", the
shared-anchor Satterthwaite approximation. Coefficient-specific degrees of
freedom can change p-values even when estimates and standard errors barely
change. test = "t.residual" retains the previous BLMM convention, including
Satterthwaite inference for draws refitted with lmerTest.test = NULL resolves by engine: "t.HC3" for ordinary linear models,
"t.satterthwaite" for BLMM, and native "t" inference for lme4 and
nlme. Explicit incompatible choices now fail before sampling. The two
new selectors are available only with BLMM. Results record method and test.contrasts accepts a named numeric matrix of scalar linear hypotheses at
fitting time. Contrasts use the full within-fit coefficient covariance and
the original composition and scale draws. Results are stored separately in
fit$contrasts; use summary(fit, component = "contrasts") to summarize them.
The return.pars fields control coefficient and contrast results together;
degrees of freedom are now included by default.lmerTest Satterthwaite inference used by the lme4 engine and its
fallback refits.lmerTest fits rather than returning
pathological uncertainty.sample.sm() now accepts s.sd, and coefficient.sm() now accepts c.sd.s.cov and c.cov arguments for covariance matrices.
The legacy s.var, s.cor, and c.cor arguments retain their original
semantics with deprecation warnings until ALDEx3 2.0.0.gamma is a standard deviation in clr.sm() and tss.sm(),
and documented that sample-level and coefficient-level priors have different
covariance structures even when their SD inputs are numerically equal.method = "blmm": an ALDEx3-specific approximate mixed-effects engine using a batched profiled REML anchor fit per feature, draw-specific local covariance updates, and exact conditional GLS fixed-effect solves. Falls back to lme4 for features where the approximation cannot be evaluated cleanly.n.cores) for method = "blmm" and method = "lme4".lme4.aldex.plot() for pairwise ALDEx3 model contrasts, with volcano,
effect, MA, and waterfall plot modes.aldex.effect() for ALDEx2-inspired effect diagnostics on a single
binary ALDEx3 contrast. The function reports the mean contrast estimate,
mean pooled within-group standard deviation, mean Cohen's d, and an
ALDEx2-style directional overlap diagnostic.cohensd() documentation and preserved its legacy return value: a
feature by Monte Carlo sample matrix of Cohen's d draws.logComp + logScale, and clearly report that they require stored
Monte Carlo arrays.gut_crohns_data is available during vignette rebuilds.