please_microbiome, a long-format data frame of
genus-level relative abundances from the pediatric study of Lewis et
al. (2015), reconstructed from the public chvlyl/PLEASE
repository following the filtering and recoding conventions of the
ZIBR R package (Chen and Li, 2016, https://CRAN.R-project.org/package=ZIBR). The data cover
3186 post-baseline observations from 59 subjects (47 anti-TNF, 12 EEN)
and 18 bacterial genera, with the week-0 abundance of each genus stored
as a subject-level covariate. The dataset is documented in
R/please_microbiome.R and reproduced by
data-raw/please_microbiome.R.Reject simultaneous formula and legacy arguments for either component or for the subject grouping variable, with explicit error messages.
Removed the unused internal model_name argument and
the redundant NA in the random-effect boundary
check.
Clarified that separate component optimization maximizes the joint likelihood under independent random effects and no shared parameters.
Removed installation and export checks for the package itself
from zavr(). The positive density now calls
dNVASIM() directly, and dependency checks and version
reporting cover external packages only.
Standardized longitudinal fixed-effect notation as
gamma for the discrete component and beta for
the continuous component. Renamed the discrete coefficient prefix from
alpha_ to gamma_ in coef() and
vcov() and aligned printed headings and documentation. Code
that selects discrete coefficients by name must use
gamma_.
Added zavr(), a zero-augmented normal-kernel Vasicek
regression for longitudinal responses in [0, 1). A logistic
component models presence, a Vasicek component models the conditional
mean of positive responses, and each component has an independent
subject-specific Gaussian random intercept.
Added zabr(), a zero-augmented beta random-intercept
model following Chen and Li (2016), to provide a directly comparable
beta positive component under the same computational interface.
Added one-sided formula interfaces for the presence and positive
components and a random = ~ 1 | subject interface. The
earlier character-vector arguments remain temporarily available with
deprecation warnings.
Added non-adaptive Gauss–Hermite integration, multiple starting
values, component-wise and joint likelihood-ratio tests, Hessian-based
covariance estimates, optimization diagnostics, and
print(), coef(), vcov(),
logLik(), nobs(), and BIC()
methods.
Joint AIC and BIC are calculated from the complete two-part
likelihood. BIC uses the number of subjects in its penalty, following
the PROC NLMIXED convention. Component-wise criteria are
retained as diagnostics: AIC values are additive, whereas BIC values are
not generally additive because the positive component uses only subjects
with at least one positive response.
Added documentation, references, examples, dependency declarations, S3 registrations, and integration tests for the new longitudinal interfaces.
tau to
quantile in NVASIQ, LVASIQ, and
HVASIQ, including their d, p,
q, and r functions and compiled
interfaces.tau object. The fixed level is now passed
directly, for example as NVASIQ(quantile = 0.25), stored in
the family definition, and embedded as a numeric literal in the
derivative, deviance, initial-value, residual, and conditional-moment
components of the family object.vasicek_envelope() to recover the fixed level
from the fitted model and use it in bootstrap simulation and refitting.
Models fitted at different quantile levels can therefore coexist safely
in one session.tau unchanged as the estimated conditional
probability at one among nonzero observations in
ZOANVASIM.bodyfat responses, compares the three quantile kernels,
illustrates the zero-, one-, and zero-and-one-augmented families with
transport, trees, and aep, and
documents conditional and marginal fitted quantities and simulated
residual envelopes.HVASIQ family.bodyfat documentation: it contains 10
variables, and its five response variables are proportions in
(0, 1), not values on a 0–100 scale.tau by
p in the NVASIM quantile formula and corrected
the description of the bivariate-normal CDF.ZOANVASIM example based on the
boundary-valued inappropriate-stay proportion in aep.print.vasicek_envelope() S3 method
and tests for the print and plot interfaces.ZOANVASIM at
nu = 0 or tau = 0 are nested limiting cases in
the closure of its open parameter space.d,
p, and q functions with their implemented
behavior.LVASIQ, expanded the package-level references,
and documented all four included datasets in the package overview and
README.HVASIQ() GAMLSS family for conditional
quantile regression with a hyperbolic-secant kernel and a user-specified
fixed quantile level.dHVASIQ(), pHVASIQ(),
qHVASIQ(), and rHVASIQ() functions, including
stable log-probability and tail calculations.mu and sigma.vasicek_envelope() to fitted
HVASIQ models.d/p/q/rZANVASIM(), d/p/q/rOANVASIM(), and
d/p/q/rZOANVASIM(). The established numbered names remain
available. This correspondence allows likelihood-based methods such as
vcov.gamlss() to locate each family’s density
function.ZANVASIM(), a zero-augmented normal-kernel
Vasicek mean family, with nu modeling the probability at
zero.OANVASIM(), a one-augmented normal-kernel Vasicek
mean family, with nu modeling the probability at one.ZOANVASIM(), a zero-and-one-augmented
normal-kernel Vasicek mean family. In this family, nu is
the probability at zero and tau is the conditional
probability at one among nonzero observations.d0NVASIM(),
p0NVASIM(), q0NVASIM(),
r0NVASIM(), d1NVASIM(),
p1NVASIM(), q1NVASIM(),
r1NVASIM(), d01NVASIM(),
p01NVASIM(), q01NVASIM(), and
r01NVASIM() functions.vasicek_envelope() for parametric-bootstrap
simulated envelopes based on fitted gamlss objects.vasicek_envelope
objects. The documentation now distinguishes these full
quantile–quantile plots from half-normal plots and explains the
finite-sample simulated mean curve.subset must be supplied
explicitly or handled by a custom refit function.dpqr-0NvasicekM.R, dpqr-1NvasicekM.R, and
dpqr-01NvasicekM.R.ZANVASIM, OANVASIM, and
ZOANVASIM.x and q and
distinguished the boundary behavior of the zero-augmented,
one-augmented, and zero-and-one-augmented distributions.tau parameter in
ZOANVASIM from the fixed quantile level used by
NVASIQ and LVASIQ.OANVASIM and
BEOI using the one-inflated accuracy1 response
from the ReadingSkills data.VASIM to NVASIM.VASIQ to NVASIQ.LVASIQ for the logistic-kernel quantile
distribution functions.N or L) and parameterization
(M or Q) explicitly.NVASIQ() does not accept tau as an
argument. For GAMLSS fitting, tau must be defined as a
scalar variable in the global environment.dNVASIQ(),
pNVASIQ(), qNVASIQ(), and
rNVASIQ() continue to accept tau
explicitly.LVASIQ() GAMLSS family for conditional
quantile regression with the logistic-kernel Vasicek distribution.mu and sigma; model fitting does
not use numerical differentiation.qVASIM().log.p handling in all quantile
functions.variance components of VASIM() and
VASIQ().VASIQ() and removed dependence on a global tau
object.mu in
VASIQ().dpqr-vasicekmean.R and
dpqr-vasicekquant.R.variance = function(mu, sigma) in
the VASIM family.variance = function(mu, sigma) in
the VASIQ family.