Package {vasicekreg}

Overview of the vasicekreg package

Description

The vasicekreg package provides distribution functions and GAMLSS families for Vasicek-type distributions on the unit interval. Four base families are available:

• NVASIM: normal kernel with mean parameterization, where \mu=E(Y).

• NVASIQ: normal kernel with quantile parameterization, where \mu=Q_Y(\tau) for a fixed \tau\in(0,1).

• LVASIQ: logistic kernel with quantile parameterization, where \mu=Q_Y(\tau) for a fixed \tau\in(0,1).

• HVASIQ: hyperbolic-secant kernel with quantile parameterization, where \mu=Q_Y(\tau) for a fixed \tau\in(0,1).

For responses observed at the boundaries, the normal-kernel mean model is also available as:

• ZANVASIM: point mass at zero and a continuous component on (0,1).

• OANVASIM: point mass at one and a continuous component on (0,1).

• ZOANVASIM: point masses at zero and one and a continuous component on (0,1).

The documentation uses augmented for these boundary mixtures and Vasicek-type for kernel-based constructions. The established family names are retained for backward compatibility and follow familiar GAMLSS abbreviations in which ZA and OA historically denote zero- and one-adjusted families. For repeated responses in [0,1), zabr and zavr provide two-part random-intercept regressions with, respectively, beta and normal-kernel Vasicek positive components. These dedicated maximum-likelihood functions are separate from the GAMLSS families. The shape parameter \sigma\in(0,1) controls dispersion in the continuous Vasicek component. The corresponding d, p, q, and r functions provide density or probability mass values, cumulative probabilities, quantiles, and random observations, respectively.

Details

Included datasets:

• bodyfat: body-fat proportions in (0,1) and demographic covariates for 298 individuals.

• aep: hospital-stay data from 1,383 patients; noinap / los contains observations at zero and one.

• transport: bicycle-trip proportions for 60 respondents, including observations at zero.

• trees: two-year tree-survival proportions for 26 parks, including observations at one.

• please_microbiome: longitudinal genus-level relative abundances from the pediatric study of Lewis et al. (2015), in long format (3186 rows, 59 subjects, 18 genera, baseline included as a subject-level covariate).

zabr: Zero-augmented beta regression for longitudinal responses. A logistic component models presence and a beta component models positive abundance; both include independent subject-specific Gaussian random intercepts.

zavr: Zero-augmented Vasicek regression with the same two-part random-intercept structure, replacing the positive beta distribution by NVASIM.

NVASIM: Normal-kernel mean parameterization and GAMLSS family. In regression models, covariates describe the conditional mean through \mu.

NVASIQ: Normal-kernel quantile parameterization and GAMLSS family. For a fixed quantile level \tau, covariates describe the conditional \tau-th quantile through \mu.

LVASIQ: Logistic-kernel quantile parameterization and GAMLSS family. For a fixed quantile level \tau, covariates describe the conditional \tau-th quantile through \mu. A logistic-kernel mean-regression family is not provided because the mean has no closed-form expression and does not equal \mu under this parameterization.

HVASIQ: Hyperbolic-secant-kernel quantile parameterization and GAMLSS family. For a fixed quantile level \tau, covariates describe the conditional \tau-th quantile through \mu. Its conditional mean and variance are obtained by numerical quadrature and \mu must not be interpreted as the mean.

ZANVASIM: Zero-augmented normal-kernel mean family. Here \nu=P(Y=0), \mu=E(Y\mid Y>0), and the marginal mean is E(Y)=(1-\nu)\mu.

OANVASIM: One-augmented normal-kernel mean family. Here \nu=P(Y=1), \mu=E(Y\mid Y<1), and the marginal mean is E(Y)=\nu+(1-\nu)\mu. The parameters \mu and \nu therefore have the same interpretations as their counterparts in the one-inflated beta family BEOI. The shape parameter \sigma is distribution-specific and should not be compared directly between these families.

ZOANVASIM: Zero-and-one-augmented normal-kernel mean family. Here \nu=P(Y=0), \tau=P(Y=1\mid Y>0), and \mu=E(Y\mid 0<Y<1). Consequently, P(Y=1)=(1-\nu)\tau and E(Y)=(1-\nu)[\tau+(1-\tau)\mu].

The distribution functions dNVASIM, pNVASIM, qNVASIM, dNVASIQ, pNVASIQ, qNVASIQ, dLVASIQ, pLVASIQ, qLVASIQ, dHVASIQ, pHVASIQ, and qHVASIQ call compiled C++ routines through Rcpp. The boundary-augmented distribution functions are implemented in R and reuse the compiled NVASIM functions for their continuous component. Parameter validation, the GAMLSS family definitions, and all log-likelihood derivatives are implemented in R. The mean and variance components of the LVASIQ() and HVASIQ() family objects are obtained by numerical quadrature because these moments have no closed-form expressions. The longitudinal two-part models use non-adaptive Gauss–Hermite quadrature from statmod; Hessian-based covariance estimates are obtained with numDeriv.

For the distribution functions and GAMLSS constructors associated with NVASIQ, LVASIQ, and HVASIQ, the fixed quantile level is supplied through the quantile argument. It is stored in the family definition and embedded in the residual expression, so no global variable is required. This fixed quantile level is distinct from the parameter tau in ZOANVASIM(), which represents the conditional probability at one among nonzero observations.

Author(s)

Josmar Mazucheli jmazucheli@gmail.com

Bruna Alves pg402900@uem.br

References

Chen, E. Z. and Li, H. (2016). A two-part mixed-effects model for analyzing longitudinal microbiome compositional data. Bioinformatics, 32(17), 2611–2617. doi:10.1093/bioinformatics/btw308

Dunn, P. K. and Smyth, G. K. (1996). Randomized quantile residuals. Journal of Computational and Graphical Statistics, 5(3), 236–244. doi:10.1080/10618600.1996.10474708

Fischer, M. J., Hui, A. and Hösle, S. (2017). wHS-type distributions with application to finance. Journal of Statistics and Management Systems, 20(1), 67–89. doi:10.1080/09720510.2016.1190575

Lewis, J. D., Chen, E. Z., Baldassano, R. N., et al. (2015). Inflammation, antibiotics, and diet as environmental stressors of the gut microbiome in pediatric Crohn's disease. Cell Host & Microbe, 18(4), 489–500. doi:10.1016/j.chom.2015.09.008

Mazucheli, J., Alves, B., Korkmaz, M. Ç., and Leiva, V. (2022). Vasicek quantile and mean regression models for bounded data: New formulation, mathematical derivations, and numerical applications. Mathematics, 10, 1389. doi:10.3390/math10091389

Mazucheli, J. (2026). A zero-augmented Vasicek mixed-effects regression model for longitudinal microbiome relative abundance data. Under review.

Moral, R. A., Hinde, J. and Demetrio, C. G. B. (2017). Half-normal plots and overdispersed models in R: The hnp package. Journal of Statistical Software, 81(10), 1–23. doi:10.18637/jss.v081.i10

Ospina, R. and Ferrari, S. L. P. (2010). Inflated beta distributions. Statistical Papers, 51, 111–126. doi:10.1007/s00362-008-0125-4

Ospina, R. and Ferrari, S. L. P. (2012). A general class of zero-or-one inflated beta regression models. Computational Statistics & Data Analysis, 56(6), 1609–1623. doi:10.1016/j.csda.2011.10.005

Rigby, R. A. and Stasinopoulos, D. M. (2005). Generalized additive models for location, scale and shape. Applied Statistics, 54(3), 507–554. doi:10.1111/j.1467-9876.2005.00510.x

Segata, N., Waldron, L., Ballarini, A., et al. (2012). Metagenomic microbial community profiling using unique clade-specific marker genes. Nature Methods, 9(8), 811–814. doi:10.1038/nmeth.2066

Vasicek, O. A. (2002). The distribution of loan portfolio value. Risk, 15(12), 160–162.

Witzany, J. (2013). A note on the Vasicek's model with the logistic distribution. Ekonomický časopis (Journal of Economics), 61(10), 1053–1066.

Zhao, Y., Lee, A. H., Yau, K. K. W. and McLachlan, G. J. (2011). Assessing the adequacy of Weibull survival models: A simulated envelope approach. Journal of Applied Statistics, 38(10), 2089–2097. doi:10.1080/02664763.2010.545115

See Also

Useful links:

• https://jmazucheli.github.io/vasicekreg/

Hyperbolic-secant-kernel Vasicek-type quantile distribution

Description

HVASIQ() defines the hyperbolic-secant-kernel Vasicek-type distribution as a gamlss.family object for conditional quantile regression. The functions dHVASIQ(), pHVASIQ(), qHVASIQ(), and rHVASIQ() provide the density, distribution function, quantile function, and random generation.

Usage

dHVASIQ(x, mu, sigma, quantile = 0.5, log = FALSE)

pHVASIQ(q, mu, sigma, quantile = 0.5, lower.tail = TRUE, log.p = FALSE)

qHVASIQ(p, mu, sigma, quantile = 0.5, lower.tail = TRUE, log.p = FALSE)

rHVASIQ(n, mu, sigma, quantile = 0.5)

HVASIQ(quantile = 0.5, mu.link = "logit", sigma.link = "logit")

Arguments

Details

Let

H(w)=\frac{2}{\pi}\arctan\{\exp(w)\},\qquad h(w)=\frac{1}{\pi\cosh(w)},

and

q(u)=H^{-1}(u)= \log\left\{\tan\left(\frac{\pi u}{2}\right)\right\}.

For fixed \tau\in(0,1), define

a=\sqrt{\frac{1-\sigma}{\sigma}},\qquad z=a\{q(x)-q(\mu)\}+q(\tau).

The cumulative distribution function and density are

F(x\mid\mu,\sigma,\tau)=H(z)

and

f(x\mid\mu,\sigma,\tau)=a\frac{h(z)}{h\{q(x)\}}.

The quantile function is

Q(p\mid\mu,\sigma,\tau)= H\left[q(\mu)+\sqrt{\frac{\sigma}{1-\sigma}} \{q(p)-q(\tau)\}\right].

Consequently, Q(\tau)=\mu; \mu is exactly the conditional \tau-th quantile and is not the conditional mean in general.

The GAMLSS family uses analytical derivatives. For one observation, let

d=q(y)-q(\mu),\quad z=ad+q(\tau),\quad T=\tanh(z),\quad S=\operatorname{sech}^2(z),

b=\frac{1}{2\sigma(1-\sigma)},\qquad g=\frac{\pi}{\sin(\pi\mu)}.

If \ell is the individual log-likelihood contribution, then

\frac{\partial\ell}{\partial\mu}=agT,

\frac{\partial\ell}{\partial\sigma}=b(adT-1),

\frac{\partial^2\ell}{\partial\mu^2} =g^2\{-a\cos(\pi\mu)T-a^2S\},

\frac{\partial^2\ell}{\partial\mu\,\partial\sigma} =-abg(T+adS),

and

\frac{\partial^2\ell}{\partial\sigma^2} =b^2\{2(1-2\sigma)-(3-4\sigma)adT-a^2d^2S\}.

The conditional mean and variance do not have elementary closed forms and are evaluated by numerical integration of the quantile function.

The fixed level is supplied through the quantile argument. It is stored in the family definition and embedded as a numeric literal in the family components used by GAMLSS; no global variable is required.

Value

dHVASIQ() returns the density, pHVASIQ() the distribution function, qHVASIQ() the quantile function, and rHVASIQ() random deviates. HVASIQ() returns a gamlss.family object.

Author(s)

Josmar Mazucheli jmazucheli@gmail.com

References

Dunn, P. K. and Smyth, G. K. (1996). Randomized quantile residuals. Journal of Computational and Graphical Statistics, 5(3), 236–244.

Fischer, M. J., Hui, A. and Hösle, S. (2017). wHS-type distributions with application to finance. Journal of Statistics and Management Systems, 20(1), 67–89. doi:10.1080/09720510.2016.1190575

Mazucheli, J., Alves, B., Korkmaz, M. C. and Leiva, V. (2022). Vasicek quantile and mean regression models for bounded data: New formulation, mathematical derivations, and numerical applications. Mathematics, 10, 1389. doi:10.3390/math10091389

Examples

set.seed(123)
y <- rHVASIQ(500, mu = 0.60, sigma = 0.30, quantile = 0.25)

fit <- gamlss::gamlss(
    y ~ 1,
    sigma.formula = ~ 1,
    family = HVASIQ(quantile = 0.25),
    control = gamlss::gamlss.control(trace = FALSE)
)
fitted(fit, what = "mu")[1]

Logistic-kernel Vasicek-type distribution with quantile parameterization

Description

The function LVASIQ() defines the logistic-kernel Vasicek-type distribution as a gamlss.family object for conditional quantile regression. The functions dLVASIQ, pLVASIQ, qLVASIQ, and rLVASIQ give the density, distribution function, quantile function, and random generation. The parameter \mu is the conditional \tau-th quantile (0<\mu<1), \sigma is a shape parameter (0<\sigma<1), and \tau\in(0,1) is fixed by the user. The fixed level is supplied through the quantile argument.

Usage

dLVASIQ(x, mu, sigma, quantile = 0.5, log = FALSE)

pLVASIQ(q, mu, sigma, quantile = 0.5, lower.tail = TRUE, log.p = FALSE)

qLVASIQ(p, mu, sigma, quantile = 0.5, lower.tail = TRUE, log.p = FALSE)

rLVASIQ(n, mu, sigma, quantile = 0.5)

LVASIQ(quantile = 0.5, mu.link = "logit", sigma.link = "logit")

Arguments

Details

Let \mathrm{logit}(u)= \log\left(\frac{u}{1-u}\right) and \Lambda(z)=\frac{1}{1+e^{-z}}, with \lambda(z)= \Lambda(z)\left[1-\Lambda(z)\right]. Define

z = \sqrt{\frac{1-\sigma}{\sigma}} \left[\mathrm{logit}(x)-\mathrm{logit}(\mu)\right] +\mathrm{logit}(\tau).

Cumulative distribution function

F(x\mid\mu,\sigma,\tau)=\Lambda(z).

Probability density function

f(x\mid\mu,\sigma,\tau)= \sqrt{\frac{1-\sigma}{\sigma}} \frac{\lambda(z)}{x(1-x)}.

Quantile function

Q(p\mid\mu,\sigma,\tau)= \Lambda\!\left\{ \mathrm{logit}(\mu) +\sqrt{\frac{\sigma}{1-\sigma}} \left[\mathrm{logit}(p)-\mathrm{logit}(\tau)\right] \right\}.

By construction Q(\tau)=\mu, i.e. \mu is the \tau-th quantile. Note that, unlike the normal-kernel Vasicek distribution, the logistic kernel does not yield a closed-form mean; in particular E(X)\neq\mu in general.

The GAMLSS family uses analytical derivatives. For one observation, let

a=\sqrt{\frac{1-\sigma}{\sigma}},\qquad d=\mathrm{logit}(y)-\mathrm{logit}(\mu),\qquad P=\Lambda\left\{ad+\mathrm{logit}(\tau)\right\},

and define

V=P(1-P),\qquad b=\frac{1}{2\sigma(1-\sigma)},\qquad g=\frac{1}{\mu(1-\mu)}.

If \ell denotes the individual log-likelihood contribution, the first derivatives are

\frac{\partial\ell}{\partial\mu} =-ag(1-2P)

and

\frac{\partial\ell}{\partial\sigma} =-b\left\{1+ad(1-2P)\right\}.

The second and cross derivatives are

\frac{\partial^2\ell}{\partial\mu^2} =g^2\left\{ a(1-2\mu)(1-2P)-2a^2V \right\},

\frac{\partial^2\ell}{\partial\mu\,\partial\sigma} =abg\left\{ (1-2P)-2adV \right\},

and

\frac{\partial^2\ell}{\partial\sigma^2} =b^2\left\{ 2(1-2\sigma) +ad(3-4\sigma)(1-2P) -2a^2d^2V \right\}.

These expressions are evaluated directly by LVASIQ(); numerical differentiation is not used. The mean and variance components of the family object use numerical quadrature because the corresponding moments do not have elementary closed forms.

Value

dLVASIQ gives the density, pLVASIQ the distribution function, qLVASIQ the quantile function, and rLVASIQ generates random deviates. LVASIQ() returns a gamlss.family object.

Note

The level supplied through quantile is stored in the family definition and embedded as a numeric literal in the family components used by GAMLSS; no global variable is required.

Author(s)

Josmar Mazucheli jmazucheli@gmail.com

References

Mazucheli, J., Alves, B., Korkmaz, M. C. and Leiva, V. (2022). Vasicek quantile and mean regression models for bounded data: New formulation, mathematical derivations, and numerical applications. Mathematics, 10, 1389.

Vasicek, O. A. (2002). The distribution of loan portfolio value. Risk, 15(12), 160–162.

Witzany, J. (2013). A note on the Vasicek's model with the logistic distribution. Ekonomický časopis (Journal of Economics), 61(10), 1053–1066.

Examples

set.seed(123)
x <- rLVASIQ(n = 1000, mu = 0.50, sigma = 0.25, quantile = 0.5)
S <- seq(min(x), max(x), length.out = 1000)

hist(x, prob = TRUE, main = "Logistic-kernel Vasicek-type")
lines(S, dLVASIQ(x = S, mu = 0.50, sigma = 0.25, quantile = 0.5), col = 2)

plot(ecdf(x))
lines(S, pLVASIQ(q = S, mu = 0.50, sigma = 0.25, quantile = 0.5), col = 2)

data <- data.frame(
  y = rLVASIQ(n = 100, mu = 0.50, sigma = 0.25, quantile = 0.50)
)
fit <- gamlss::gamlss(
    y ~ 1,
    data = data,
    family = LVASIQ(
      quantile = 0.50, mu.link = "logit", sigma.link = "logit"
    )
)
fitted(fit, what = "mu")[1:5]

Normal-kernel Vasicek-type distribution with mean parameterization

Description

Defines the normal-kernel Vasicek distribution under a mean parameterization for use as a gamlss.family. The parameter \mu represents the mean of the distribution, with 0 < \mu < 1, and \sigma is a shape parameter.

The density, distribution function, quantile function and random number generation are provided by dNVASIM(), pNVASIM(), qNVASIM() and rNVASIM(), respectively.

Usage

dNVASIM(x, mu, sigma, log = FALSE)

pNVASIM(q, mu, sigma, lower.tail = TRUE, log.p = FALSE)

qNVASIM(p, mu, sigma, lower.tail = TRUE, log.p = FALSE)

rNVASIM(n, mu, sigma)

NVASIM(mu.link = "logit", sigma.link = "logit")

Arguments

Details

Probability density function

f(x\mid \mu ,\sigma )=\sqrt{\frac{1-\sigma }{\sigma }}\exp \left\{ \frac{1}{2}\left[ \Phi ^{-1}\left( x\right) ^{2}-\left( \frac{\Phi ^{-1}\left( x\right) \sqrt{1-\sigma }-\Phi ^{-1}\left( \mu \right) }{\sqrt{\sigma }}\right) ^{2}\right] \right\}

Cumulative distribution function

F(x\mid \mu ,\sigma )=\Phi \left( \frac{\Phi ^{-1}\left( x\right) \sqrt{1-\sigma }-\Phi ^{-1}\left( \mu \right) }{\sqrt{\sigma }}\right)

Quantile function

Q(p \mid \mu ,\sigma )=F^{-1}(p \mid \mu ,\sigma )=\Phi \left(\frac{\Phi ^{-1}\left(\mu\right) +\Phi ^{-1}\left( p \right) \sqrt{\sigma }}{\sqrt{1-\sigma }}\right)

Expected value

E(X) = \mu

Variance

Var(X) = \Phi_2\left ( \Phi^{-1}(\mu),\Phi^{-1}(\mu),\sigma \right )-\mu^2

where (x, \mu, \sigma, p) \in (0,1) and \Phi_2(\cdot) is the cumulative distribution function for the standard bivariate normal distribution with correlation \sigma.

Value

NVASIM() returns a gamlss.family object.

Note

In the NVASIM() parameterization, \mu corresponds to the mean of the distribution and \sigma is a shape parameter.

Author(s)

Josmar Mazucheli jmazucheli@gmail.com Bruna Alves pg402900@uem.br

References

Hastie, T. J. and Tibshirani, R. J. (1990). Generalized Additive Models. Chapman and Hall, London.

Mazucheli, J., Alves, B., Korkmaz, M. C., and Leiva, V. (2022). Vasicek quantile and mean regression models for bounded data: New formulation, mathematical derivations, and numerical applications. Mathematics, 10, 1389. doi:10.3390/math10091389

Rigby, R. A. and Stasinopoulos, D. M. (2005). Generalized additive models for location, scale and shape (with discussion). Applied Statistics, 54(3), 507–554.

Rigby, R. A., Stasinopoulos, D. M., Heller, G. Z., and De Bastiani, F. (2019). Distributions for Modeling Location, Scale, and Shape: Using GAMLSS in R. Chapman and Hall/CRC.

Stasinopoulos, D. M. and Rigby, R. A. (2007). Generalized additive models for location, scale and shape (GAMLSS) in R. Journal of Statistical Software, 23(7), 1–46.

Stasinopoulos, D. M., Rigby, R. A., Heller, G., Voudouris, V., and De Bastiani, F. (2017). Flexible Regression and Smoothing: Using GAMLSS in R. Chapman and Hall/CRC.

Vasicek, O. A. (1987). Probability of loss on loan portfolio. KMV Corporation.

Vasicek, O. A. (2002). The distribution of loan portfolio value. Risk, 15(12), 160–162.

See Also

NVASIQ, pmvnorm

Examples

set.seed(123)
x <- rNVASIM(n = 1000, mu = 0.50, sigma = 0.69)
R <- range(x)
S <- seq(from = R[1], to = R[2], length.out = 1000)

hist(x, prob = TRUE, main = 'Vasicek')
lines(S, dNVASIM(x = S, mu = 0.50, sigma = 0.69), col = 2)

plot(ecdf(x))
lines(S, pNVASIM(q = S, mu = 0.50, sigma = 0.69), col = 2)

plot(quantile(x, probs = S), type = "l")
lines(qNVASIM(p = S, mu = 0.50, sigma = 0.69), col = 2)

library(gamlss)
set.seed(123)
data <- data.frame(y =  rNVASIM(n = 100, mu = 0.5, sigma = 0.69))

fit <- gamlss(y ~ 1, data = data, mu.link = 'logit', sigma.link = 'logit', family = NVASIM)
1 /(1 + exp(-fit$mu.coefficients))
1 /(1 + exp(-fit$sigma.coefficients))

## Not run: 
library(gamlss)
set.seed(123)

n <- 1000
x <- rbinom(n, size = 1, prob = 0.5)
eta <- 0.5 + 1 * x;
mu <- 1 / (1 + exp(-eta));
sigma <- 0.5;
y <- rNVASIM(n, mu, sigma)
data <- data.frame(y, x)

fit <- gamlss(y ~ x, data = data, family = NVASIM, mu.link = 'logit', sigma.link = 'logit', 
control = gamlss.control(n.cyc = 200))
summary(fit)

## End(Not run)

Normal-kernel Vasicek-type distribution with quantile parameterization

Description

The function NVASIQ() defines the normal-kernel Vasicek-type distribution as a gamlss.family object. In this parameterization, \mu corresponds to the fixed \tau-th quantile and \sigma is a shape parameter. The fixed level is supplied through the quantile argument. The functions dNVASIQ, pNVASIQ, qNVASIQ, and rNVASIQ define the density, distribution function, quantile function, and random generation for the Vasicek distribution, respectively.

Usage

dNVASIQ(x, mu, sigma, quantile = 0.5, log = FALSE)

pNVASIQ(q, mu, sigma, quantile = 0.5, lower.tail = TRUE, log.p = FALSE)

qNVASIQ(p, mu, sigma, quantile = 0.5, lower.tail = TRUE, log.p = FALSE)

rNVASIQ(n, mu, sigma, quantile = 0.5)

NVASIQ(quantile = 0.5, mu.link = "logit", sigma.link = "logit")

Arguments

Details

Probability density function:

f\left(x \mid \mu, \sigma, \tau\right) = \sqrt{\frac{1-\sigma}{\sigma}} \exp\left\{\frac{1}{2}\left[\Phi^{-1}(x)^2 - \left(\frac{\sqrt{1-\sigma}\left(\Phi^{-1}(x)-\Phi^{-1}(\mu)\right) - \sqrt{\sigma}\,\Phi^{-1}(\tau)}{\sqrt{\sigma}}\right)^2\right]\right\}.

Cumulative distribution function:

F\left(x \mid \mu, \sigma, \tau\right) = \Phi\left(\frac{\sqrt{1-\sigma}\left(\Phi^{-1}(x)-\Phi^{-1}(\mu)\right) - \sqrt{\sigma}\,\Phi^{-1}(\tau)}{\sqrt{\sigma}}\right).

where 0<x<1, 0<\mu<1, 0<\sigma<1, and 0<\tau<1; \mu is the \tau-th quantile and \sigma is the shape parameter.

Value

NVASIQ() returns a gamlss.family object that can be used to fit a Vasicek-type distribution using the gamlss function.

Note

For NVASIQ(), \mu corresponds to the \tau-th quantile and \sigma is a shape parameter. The level supplied through quantile is stored in the family definition and embedded as a numeric literal in the family components used by GAMLSS; no global variable is required.

Author(s)

Josmar Mazucheli jmazucheli@gmail.com

Bruna Alves pg402900@uem.br

References

Hastie, T. J. and Tibshirani, R. J. (1990). Generalized Additive Models. Chapman and Hall, London.

Mazucheli, J., Alves, B., Korkmaz, M. Ç., and Leiva, V. (2022). Vasicek quantile and mean regression models for bounded data: New formulation, mathematical derivations, and numerical applications. Mathematics, 10, 1389.

Rigby, R. A. and Stasinopoulos, D. M. (2005). Generalized additive models for location, scale and shape (with discussion). Applied Statistics, 54(3), 507–554.

Rigby, R. A., Stasinopoulos, D. M., Heller, G. Z., and De Bastiani, F. (2019). Distributions for Modeling Location, Scale, and Shape: Using GAMLSS in R. Chapman and Hall/CRC.

Stasinopoulos, D. M. and Rigby, R. A. (2007). Generalized additive models for location, scale and shape (GAMLSS) in R. Journal of Statistical Software, 23(7), 1–46.

Stasinopoulos, D. M., Rigby, R. A., Heller, G., Voudouris, V., and De Bastiani, F. (2017). Flexible Regression and Smoothing: Using GAMLSS in R. Chapman and Hall/CRC.

Vasicek, O. A. (1987). Probability of loss on loan portfolio. KMV Corporation.

Vasicek, O. A. (2002). The distribution of loan portfolio value. Risk, 15(12), 160–162.

See Also

NVASIM

Examples

set.seed(123)
x <- rNVASIQ(n = 1000, mu = 0.50, sigma = 0.69, quantile = 0.50)
R <- range(x)
S <- seq(from = R[1], to = R[2], length.out = 1000)

hist(x, prob = TRUE, main = "Vasicek")
lines(S, dNVASIQ(x = S, mu = 0.50, sigma = 0.69, quantile = 0.50), col = 2)

plot(ecdf(x))
lines(S, pNVASIQ(q = S, mu = 0.50, sigma = 0.69, quantile = 0.50), col = 2)

plot(quantile(x, probs = S), type = "l")
lines(qNVASIQ(p = S, mu = 0.50, sigma = 0.69, quantile = 0.50), col = 2)

library(gamlss)
set.seed(123)
data <- data.frame(
  y = rNVASIQ(n = 100, mu = 0.50, sigma = 0.69, quantile = 0.50)
)

fit <- gamlss(y ~ 1, data = data,
              family = NVASIQ(quantile = 0.50,
                             mu.link = "logit",
                             sigma.link = "logit"))
1 / (1 + exp(-fit$mu.coefficients))
1 / (1 + exp(-fit$sigma.coefficients))

set.seed(123)
n <- 100
x <- rbinom(n, size = 1, prob = 0.5)
eta <- 0.5 + 1 * x
mu <- 1 / (1 + exp(-eta))
sigma <- 0.5
y <- rNVASIQ(n, mu, sigma, quantile = 0.5)
data <- data.frame(y, x)

fit <- gamlss(
  y ~ x, data = data, family = NVASIQ(quantile = 0.50)
)

fitquantiles <- lapply(c(0.10, 0.25, 0.50, 0.75, 0.90), function(level) {
  gamlss(y ~ x, data = data, family = NVASIQ(quantile = level))
})

lapply(fitquantiles, summary)

One-augmented normal-kernel Vasicek-type distribution

Description

Defines a one-augmented normal-kernel Vasicek-type distribution for responses in (0,1]. The parameter \nu is the probability at one. Conditional on an observation in (0,1), the distribution is NVASIM with mean \mu and shape parameter \sigma.

Usage

d1NVASIM(x, mu = 0.5, sigma = 0.5, nu = 0.1, log = FALSE)

p1NVASIM(q, mu = 0.5, sigma = 0.5, nu = 0.1, lower.tail = TRUE, log.p = FALSE)

q1NVASIM(p, mu = 0.5, sigma = 0.5, nu = 0.1, lower.tail = TRUE, log.p = FALSE)

r1NVASIM(n, mu = 0.5, sigma = 0.5, nu = 0.1)

dOANVASIM(x, mu = 0.5, sigma = 0.5, nu = 0.1, log = FALSE)

pOANVASIM(q, mu = 0.5, sigma = 0.5, nu = 0.1, lower.tail = TRUE, log.p = FALSE)

qOANVASIM(p, mu = 0.5, sigma = 0.5, nu = 0.1, lower.tail = TRUE, log.p = FALSE)

rOANVASIM(n, mu = 0.5, sigma = 0.5, nu = 0.1)

OANVASIM(mu.link = "logit", sigma.link = "logit", nu.link = "logit")

Arguments

Details

Let Y_c\sim\mathrm{NVASIM}(\mu,\sigma) and let 0<\nu<1. The BEOI-type one-augmented distribution is defined by

P(Y=1)=\nu

and

f_Y(y)=(1-\nu)f_{Y_c}(y\mid\mu,\sigma),\quad 0<y<1.

Consequently,

E(Y)=\nu+(1-\nu)\mu

and

\mathrm{Var}(Y)=(1-\nu)\mathrm{Var}(Y_c)+ \nu(1-\nu)(1-\mu)^2.

Thus, \mu=E(Y\mid 0<Y<1) is the mean of the continuous component, whereas \nu+(1-\nu)\mu is the marginal mean.

Value

OANVASIM() returns a gamlss.family object. The functions d1NVASIM(), p1NVASIM(), q1NVASIM(), and r1NVASIM() return probability mass or density values, cumulative probabilities, quantiles, and random observations, respectively. dOANVASIM(), pOANVASIM(), qOANVASIM(), and rOANVASIM() are equivalent names following the GAMLSS family-name convention.

References

Ospina, R. and Ferrari, S. L. P. (2010). Inflated beta distributions. Statistical Papers, 51, 111–126.

Rigby, R. A. and Stasinopoulos, D. M. (2005). Generalized additive models for location, scale and shape. Applied Statistics, 54(3), 507–554.

See Also

NVASIM, ZANVASIM, BEOI

Examples

set.seed(123)
y <- r1NVASIM(1000, mu = 0.60, sigma = 0.30, nu = 0.20)
mean(y == 1)
mean(y)
0.20 + (1 - 0.20) * 0.60

## Not run: 
library(gamlss)
fit <- gamlss(
  y ~ 1,
  sigma.formula = ~ 1,
  nu.formula = ~ 1,
  family = OANVASIM(),
  control = gamlss.control(trace = FALSE)
)

## End(Not run)

Zero-augmented normal-kernel Vasicek-type distribution

Description

Defines a zero-augmented normal-kernel Vasicek-type distribution for responses in [0,1). The parameter \nu is the probability of a structural zero. Conditional on a positive response, the distribution is NVASIM with mean \mu and shape parameter \sigma.

Usage

d0NVASIM(x, mu = 0.5, sigma = 0.5, nu = 0.1, log = FALSE)

p0NVASIM(q, mu = 0.5, sigma = 0.5, nu = 0.1, lower.tail = TRUE, log.p = FALSE)

q0NVASIM(p, mu = 0.5, sigma = 0.5, nu = 0.1, lower.tail = TRUE, log.p = FALSE)

r0NVASIM(n, mu = 0.5, sigma = 0.5, nu = 0.1)

dZANVASIM(x, mu = 0.5, sigma = 0.5, nu = 0.1, log = FALSE)

pZANVASIM(q, mu = 0.5, sigma = 0.5, nu = 0.1, lower.tail = TRUE, log.p = FALSE)

qZANVASIM(p, mu = 0.5, sigma = 0.5, nu = 0.1, lower.tail = TRUE, log.p = FALSE)

rZANVASIM(n, mu = 0.5, sigma = 0.5, nu = 0.1)

ZANVASIM(mu.link = "logit", sigma.link = "logit", nu.link = "logit")

Arguments

Details

Let Y_+\sim\mathrm{NVASIM}(\mu,\sigma) and let 0<\nu<1. The zero-augmented distribution is defined by

P(Y=0)=\nu

and

f_Y(y)=(1-\nu)f_{Y_+}(y\mid\mu,\sigma),\quad 0<y<1.

Its cumulative distribution function is

F_Y(y)=\nu+(1-\nu)F_{Y_+}(y\mid\mu,\sigma),\quad 0<y<1.

Consequently,

E(Y)=(1-\nu)\mu

and

\mathrm{Var}(Y)=(1-\nu)\mathrm{Var}(Y_+)+ \nu(1-\nu)\mu^2.

Thus, \mu is the mean conditional on Y>0; it is not the marginal mean when \nu>0. The marginal mean is (1-\nu)\mu.

Value

ZANVASIM() returns a gamlss.family object. The functions d0NVASIM(), p0NVASIM(), q0NVASIM(), and r0NVASIM() return density or probability mass values, cumulative probabilities, quantiles, and random observations, respectively. dZANVASIM(), pZANVASIM(), qZANVASIM(), and rZANVASIM() are equivalent names following the GAMLSS family-name convention.

References

Mazucheli, J., Alves, B., Korkmaz, M. C., and Leiva, V. (2022). Vasicek quantile and mean regression models for bounded data: New formulation, mathematical derivations, and numerical applications. Mathematics, 10, 1389. doi:10.3390/math10091389

Ospina, R. and Ferrari, S. L. P. (2010). Inflated beta distributions. Statistical Papers, 51, 111–126.

Rigby, R. A. and Stasinopoulos, D. M. (2005). Generalized additive models for location, scale and shape. Applied Statistics, 54(3), 507–554.

See Also

NVASIM, BEZI

Examples

set.seed(123)
y <- r0NVASIM(1000, mu = 0.60, sigma = 0.30, nu = 0.20)
mean(y == 0)
mean(y)
(1 - 0.20) * 0.60

library(gamlss)
fit <- gamlss(
  y ~ 1,
  sigma.formula = ~ 1,
  nu.formula = ~ 1,
  family = ZANVASIM(),
  control = gamlss.control(trace = FALSE)
)
fitted(fit, what = "mu")[1]
fitted(fit, what = "sigma")[1]
fitted(fit, what = "nu")[1]

Zero-and-one-augmented normal-kernel Vasicek-type distribution

Description

Defines a normal-kernel Vasicek-type distribution augmented by point masses at zero and one. Conditional on an observation in (0,1), the continuous component is NVASIM with mean \mu and shape parameter \sigma.

Usage

d01NVASIM(x, mu = 0.5, sigma = 0.5, nu = 0.1, tau = 0.1, log = FALSE)

p01NVASIM(
  q,
  mu = 0.5,
  sigma = 0.5,
  nu = 0.1,
  tau = 0.1,
  lower.tail = TRUE,
  log.p = FALSE
)

q01NVASIM(
  p,
  mu = 0.5,
  sigma = 0.5,
  nu = 0.1,
  tau = 0.1,
  lower.tail = TRUE,
  log.p = FALSE
)

r01NVASIM(n, mu = 0.5, sigma = 0.5, nu = 0.1, tau = 0.1)

dZOANVASIM(x, mu = 0.5, sigma = 0.5, nu = 0.1, tau = 0.1, log = FALSE)

pZOANVASIM(
  q,
  mu = 0.5,
  sigma = 0.5,
  nu = 0.1,
  tau = 0.1,
  lower.tail = TRUE,
  log.p = FALSE
)

qZOANVASIM(
  p,
  mu = 0.5,
  sigma = 0.5,
  nu = 0.1,
  tau = 0.1,
  lower.tail = TRUE,
  log.p = FALSE
)

rZOANVASIM(n, mu = 0.5, sigma = 0.5, nu = 0.1, tau = 0.1)

ZOANVASIM(
  mu.link = "logit",
  sigma.link = "logit",
  nu.link = "logit",
  tau.link = "logit"
)

Arguments

Details

Let Y_c\sim\mathrm{NVASIM}(\mu,\sigma). Write p_0, p_1, and p_c for the probabilities of zero, one, and the continuous component. The sequential BEOI-type parameterization is

\nu=P(Y=0),\qquad \tau=P(Y=1\mid Y>0).

Hence,

p_0=\nu,\qquad p_1=(1-\nu)\tau,\qquad p_c=(1-\nu)(1-\tau).

The distribution is

P(Y=0)=p_0,\qquad P(Y=1)=p_1

and

f_Y(y)=p_c f_{Y_c}(y\mid\mu,\sigma),\quad 0<y<1.

Its marginal mean and variance are

E(Y)=(1-\nu)\left[\tau+(1-\tau)\mu\right]

and

\mathrm{Var}(Y)= (1-\nu)\left[(1-\tau)\left\{\mathrm{Var}(Y_c)+\mu^2\right\} +\tau\right] -\left\{(1-\nu)\left[\tau+(1-\tau)\mu\right]\right\}^2.

Logit links for \nu and \tau guarantee valid probabilities. In the closure of the parameter space, \nu=0 gives the BEOI-type one-augmented model and \tau=0 gives the zero-augmented model. The implemented distribution functions and GAMLSS family use the open parameter space 0<\nu<1 and 0<\tau<1; these nested models are therefore limiting cases rather than admissible interior parameter values.

Value

ZOANVASIM() returns a four-parameter gamlss.family object. The functions d01NVASIM(), p01NVASIM(), q01NVASIM(), and r01NVASIM() return probability mass or density values, cumulative probabilities, quantiles, and random observations, respectively. dZOANVASIM(), pZOANVASIM(), qZOANVASIM(), and rZOANVASIM() are equivalent names following the GAMLSS family-name convention.

References

Ospina, R. and Ferrari, S. L. P. (2010). Inflated beta distributions. Statistical Papers, 51, 111–126.

Rigby, R. A. and Stasinopoulos, D. M. (2005). Generalized additive models for location, scale and shape. Applied Statistics, 54(3), 507–554.

See Also

NVASIM, ZANVASIM, OANVASIM, BEOI

Examples

set.seed(123)
y <- r01NVASIM(
  1000, mu = 0.60, sigma = 0.30, nu = 0.20, tau = 0.25
)
c(zero = mean(y == 0), one = mean(y == 1))
mean(y)
(1 - 0.20) * (0.25 + (1 - 0.25) * 0.60)

## Not run: 
library(gamlss)
fit <- gamlss(
  y ~ 1,
  sigma.formula = ~ 1,
  nu.formula = ~ 1,
  tau.formula = ~ 1,
  family = ZOANVASIM(),
  control = gamlss.control(trace = FALSE)
)

## Real-data example with observations at both boundaries
data("aep", package = "vasicekreg")
aep$inappropriate <- with(aep, noinap / los)
fit_aep <- gamlss(
  inappropriate ~ sex + ward + year + age + loglos,
  sigma.formula = ~ loglos,
  nu.formula = ~ loglos,
  tau.formula = ~ loglos,
  family = ZOANVASIM(),
  data = aep,
  control = gamlss.control(n.cyc = 200, trace = FALSE)
)

## End(Not run)

Hospital-stay appropriateness data

Description

Data on 1,383 patients admitted to Hospital del Mar, Barcelona, during 1988 and 1990. The data record the total length of stay and the number of days classified as inappropriate.

Format

A data frame with 1,383 observations and 8 variables:

los

Total number of days spent in hospital.

noinap

Number of hospital-stay days classified as inappropriate.

loglos

Logarithm of length of stay divided by 10, log(los / 10).

sex

Patient's sex: factor with levels 1 (male) and 2 (female).

ward

Hospital ward: factor with levels 1 (medical), 2 (surgical), and 3 (other).

year

Admission year: factor with levels 88 and 90.

age

Patient's age minus 55 years.

y

Two-column matrix response whose first column is noinap and whose second column is los - noinap.

Details

Gange et al. (1996) modeled the number of inappropriate days conditional on total length of stay using binomial and beta-binomial models. For augmented continuous-response models, the patient-level proportion can be constructed as noinap / los. The object distributed here retains the structure and values supplied by the gamlss.data package.

Source

Gange, S. J., Munoz, A., Saez, M. and Alonso, J. (1996). Use of the beta-binomial distribution to model the effect of policy changes on appropriateness of hospital stays. Applied Statistics, 45(3), 371–382.

References

Stasinopoulos, M. and Rigby, R. (2025). gamlss.data: Data for Generalized Additive Models for Location Scale and Shape. R package version 6.0-7. https://CRAN.R-project.org/package=gamlss.data

Examples

data("aep", package = "vasicekreg")
str(aep)
inappropriate <- with(aep, noinap / los)
table(inappropriate == 0, inappropriate == 1)

Body Fat Proportions Dataset

Description

Body fat proportions from individuals assisted in a public hospital in Curitiba, Paraná, Brazil.

Usage

bodyfat

Format

A data frame with 298 observations and 10 variables. The five body-fat responses are proportions in (0,1) (for example, 0.163 represents 16.3 percent):

• ID: individual identifier.

• ARMS: arms fat proportion.

• LEGS: legs fat proportion.

• BODY: body fat proportion.

• ANDROID: android fat proportion.

• GYNECOID: gynoid fat proportion.

• AGE: age of individuals.

• BMI: body mass index.

• SEX: 1 for female and 2 for male.

• IPAQ: physical activity level according to IPAQ (0 = sedentary, 1 = insufficiently active, 2 = active).

Author(s)

Josmar Mazucheli jmazucheli@gmail.com

Bruna Alves pg402900@uem.br

References

Mazucheli, J., Alves, B., Korkmaz, M. Ç., and Leiva, V. (2022). Vasicek quantile and mean regression models for bounded data: New formulation, mathematical derivations, and numerical applications. Mathematics, 10, 1389.

Mazucheli, J., Leiva, V., Alves, B., and Menezes, A. F. B. (2021). A new quantile regression for modeling bounded data under a unit Birnbaum-Saunders distribution with applications in medicine and politics. Symmetry, 13(4), 1–21.

Petterle, R. R., Bonat, W. H., Scarpin, C. T., Jonasson, T., and Borba, V. Z. C. (2020). Multivariate quasi-beta regression models for continuous bounded data. The International Journal of Biostatistics, 17(1), 39–53.

Examples

data(bodyfat, package = "vasicekreg")

bodyfat$AGE <- bodyfat$AGE - 46.00
bodyfat$BMI <- bodyfat$BMI - 24.72
bodyfat$SEX <- as.factor(bodyfat$SEX)
bodyfat$IPAQ<- as.factor(bodyfat$IPAQ)

library(gamlss)

## Mean regression model
fitmean <- gamlss(
  ARMS ~ AGE + BMI + SEX + IPAQ,
  data = bodyfat,
  family = NVASIM(mu.link = "logit", sigma.link = "logit")
)

## Not run: 
quantile_levels <- c(0.25, 0.50, 0.75)

## Quantile regression models with the normal kernel
fit_normal <- lapply(quantile_levels, function(level) {
  gamlss(
    ARMS ~ AGE + BMI + SEX + IPAQ,
    data = bodyfat,
    family = NVASIQ(
      quantile = level,
      mu.link = "logit",
      sigma.link = "logit"
    )
  )
})

## Quantile regression models with the logistic kernel
fit_logistic <- lapply(quantile_levels, function(level) {
  gamlss(
    ARMS ~ AGE + BMI + SEX + IPAQ,
    data = bodyfat,
    family = LVASIQ(
      quantile = level,
      mu.link = "logit",
      sigma.link = "logit"
    )
  )
})

## Quantile regression models with the Hyperbolic-secant-kernel
fit_hsk <- lapply(quantile_levels, function(level) {
  gamlss(
    ARMS ~ AGE + BMI + SEX + IPAQ,
    data = bodyfat,
    family = HVASIQ(
      quantile = level,
      mu.link = "logit",
      sigma.link = "logit"
    )
  )
})

lapply(fit_normal, summary)
lapply(fit_logistic, summary)
lapply(fit_hsk, summary)

## End(Not run)

Longitudinal microbiome abundances from the PLEASE study

Description

A long-format data frame of genus-level relative abundances and associated clinical covariates from the pediatric study of Lewis et al. (2015). The analytic dataset was reconstructed from the public chvlyl/PLEASE repository and processed following the filtering and recoding conventions of the ZIBR package (Chen and Li, 2016). Genus-level relative abundances were originally quantified from shotgun metagenomic sequencing using MetaPhlAn 1.7.6 (Segata et al., 2012).

Usage

please_microbiome

Format

A data frame with 3186 rows and 7 variables. Each row is one post-baseline observation of a genus in a subject, so the 18 genera contribute 177 rows each (3 visits \times 59 subjects).

Genus

character. Genus name, e.g. "g__Bacteroides".

Sample

character. Sample identifier.

Subject

character. Subject identifier (e.g. "S5001").

Time

numeric. Follow-up week: 1, 4 or 8. The baseline (week 0) is stored in Baseline and does not appear here.

Treat

factor. Treatment arm with levels "antiTNF" (reference) and "EEN" (exclusive enteral nutrition).

Baseline

numeric. Subject's relative abundance of the same genus at week 0, on its original proportion scale in [0,1].

Y

numeric. Relative abundance of the genus at the given post-baseline visit, in [0,1]. May contain zeros.

Details

The dataset covers 59 subjects (47 anti-TNF, 12 EEN) with observations at all four scheduled visits (baseline and weeks 1, 4 and 8) and 18 bacterial genera, for a total of 236 samples in the wide format used by Chen and Li (2016). The long format provided here contains the 177 post-baseline observations per genus, together with the baseline abundance replicated within subject and genus.

Processing followed the ZIBR workflow: samples with fewer than 10{,}000 non-human reads were removed; genera were retained if present in more than 40\% of the remaining samples and if their 90th percentile of relative abundance, computed including zeros, exceeded 1\%; and the abundances of the retained genera were renormalized to sum to one within each sample, so that each modeled abundance is relative to the retained genera rather than to the whole community. These steps preceded the restriction to the anti-TNF and EEN arms (the partial enteral nutrition arm was excluded) and the selection of subjects with observations at all four scheduled visits.

In the analysis reported in the companion paper, baseline abundance was used as a subject-level covariate on its original proportion scale, without centering or standardization, and both components of each two-part model included baseline abundance, week, and treatment. One genus, Bacteroides, had only three zeros among the post-baseline observations and was excluded from the Beta–Vasicek model comparison, leaving 17 genera; its observations are nonetheless retained here for completeness.

Source

PLEASE repository: https://github.com/chvlyl/PLEASE

Original study: Lewis, J. D., Chen, E. Z., Baldassano, R. N., et al. (2015). Inflammation, antibiotics, and diet as environmental stressors of the gut microbiome in pediatric Crohn's disease. Cell Host & Microbe, 18(4), 489–500. doi:10.1016/j.chom.2015.09.008

References

Chen, E. Z. and Li, H. (2016). A two-part mixed-effects model for analyzing longitudinal microbiome compositional data. Bioinformatics, 32(17), 2611–2617. doi:10.1093/bioinformatics/btw308

Segata, N., Waldron, L., Ballarini, A., et al. (2012). Metagenomic microbial community profiling using unique clade-specific marker genes. Nature Methods, 9(8), 811–814. doi:10.1038/nmeth.2066

Examples

data(please_microbiome)
head(please_microbiome)

# Dimensions and structure
dim(please_microbiome)                     # 3186 x 7
length(unique(please_microbiome$Genus))    # 18 genera
length(unique(please_microbiome$Subject))  # 59 subjects
table(please_microbiome$Treat)             # 47 antiTNF, 12 EEN

# Balanced within genus: 177 rows each
table(please_microbiome$Genus)

Transportation to campus

Description

Data from a stratified sample of 60 respondents concerning their mode of transportation to campus. The sampling design oversampled respondents who sometimes traveled to campus by bicycle.

Format

A data frame with 60 observations and 7 variables:

ntrips

Number of trips to campus during the preceding four weeks.

nbiked

Number of those trips made by bicycle.

status

Respondent's institutional status: faculty, staff, or student.

gender

Respondent's gender, recorded as "F" or "M".

parking

Duration of the parking permit, in months.

distance

Distance to campus. The source documentation does not specify the measurement unit.

propbiked

Proportion of trips to campus made by bicycle, calculated as nbiked / ntrips.

Details

The data originate from a consulting study reported by Korosteleva (2019). Menezes, Mazucheli, and Bourguignon (2021) analyzed them using an inflated unit-Weibull quantile regression model. The object distributed here retains the variable names and values supplied by the uwquantreg package.

Source

Korosteleva, O. (2019). Advanced Regression Models with SAS and R. Boca Raton, FL: CRC Press.

https://github.com/AndrMenezes/uwquantreg

References

Menezes, A. F. B., Mazucheli, J. and Bourguignon, M. (2021). A parametric quantile regression approach for modeling zero- or one-inflated double bounded data. Biometrical Journal, 63(4), 841–858. doi:10.1002/bimj.202000126

Menezes, A. F. B. (2026). uwquantreg: unit-Weibull quantile regression. R package version 0.1.0. https://github.com/AndrMenezes/uwquantreg

Examples

data("transport", package = "vasicekreg")
str(transport)
with(transport, all.equal(propbiked, nbiked / ntrips))

Survival of young trees

Description

Data from a study conducted by a parks and recreation department on the two-year survival of young trees planted in 26 parks.

Format

A data frame with 26 observations and 7 variables:

planted

Number of trees planted.

survived

Number of trees that survived for two years.

pest

Frequency of pest-control treatment.

fertilization

Frequency of soil fertilization.

precip

Average annual precipitation, in inches.

wind

Average annual wind speed, in miles per hour.

prop

Proportion of planted trees that survived for two years, calculated as survived / planted.

Details

The data originate from a consulting study reported by Korosteleva (2019). Menezes, Mazucheli, and Bourguignon (2021) analyzed them using an inflated unit-Weibull quantile regression model. The object distributed here retains the variable names and values supplied by the uwquantreg package.

Source

Korosteleva, O. (2019). Advanced Regression Models with SAS and R. Boca Raton, FL: CRC Press.

https://github.com/AndrMenezes/uwquantreg

References

Menezes, A. F. B., Mazucheli, J. and Bourguignon, M. (2021). A parametric quantile regression approach for modeling zero- or one-inflated double bounded data. Biometrical Journal, 63(4), 841–858. doi:10.1002/bimj.202000126

Menezes, A. F. B. (2026). uwquantreg: unit-Weibull quantile regression. R package version 0.1.0. https://github.com/AndrMenezes/uwquantreg

Examples

data("trees", package = "vasicekreg")
str(trees)
with(trees, all.equal(prop, survived / planted))

Simulated envelopes for residuals from Vasicek GAMLSS models

Description

Constructs pointwise simulated envelopes for normalized randomized quantile residuals and generalized Cox–Snell residuals from a Vasicek model fitted by gamlss(). Each simulated response is generated under the fitted model, the model is re-estimated, and the residuals from the re-estimated model are ordered before the envelope is computed.

Usage

vasicek_envelope(
  object,
  residual = c("quantile", "cox-snell"),
  nsim = 200L,
  level = 0.95,
  envelope = c("quantile", "minmax"),
  seed = NULL,
  data = NULL,
  max.attempts = 5L * nsim,
  refit = NULL,
  simulate = NULL,
  verbose = interactive()
)

## S3 method for class 'vasicek_envelope'
print(x, ...)

## S3 method for class 'vasicek_envelope'
plot(
  x,
  which = x$residual[1L],
  show.mean = TRUE,
  xlab = NULL,
  ylab = NULL,
  main = "",
  envelope.col = "grey85",
  mean.col = "blue",
  reference.col = "red",
  point.col = "black",
  ...
)

Arguments

Details

Let U_i denote the probability integral transform used by the normalized randomized quantile residual. The two residuals are

r_i^q=\Phi^{-1}(U_i)

and

r_i^{CS}=-\log(1-U_i).

For augmented families, U_i is randomized over the relevant jump of the fitted distribution at zero or one, so the reported residuals inherit whatever randomization residuals() applies for those families; the Cox–Snell residual is then also randomized at boundary observations and has an \operatorname{Exp}(1) reference distribution under a correctly specified model. The Cox–Snell residual is obtained from the quantile residual on the log survival scale, -\log\{\Pr(Z>r_i^q)\}, which is numerically stable in the upper tail and preserves the ordering of the quantile residuals.

Each residual is displayed as a full quantile–quantile plot of the ordered residuals against the corresponding theoretical quantiles (normal or exponential), not as a half-normal plot. The envelope is pointwise, not simultaneous: even under a correctly specified model a fraction of points is expected to fall outside the band. With envelope = "quantile", the two tail probabilities are equal and sum to 1-\mathrm{level}. With envelope = "minmax", the limits are the minimum and maximum at each order position, following the construction used by Zhao et al.

The pointwise mean of the ordered simulated residuals is the reference calibrated to the fitted model and to the sample size. The identity line drawn by the plot method is a theoretical idealization and, for Cox–Snell residuals, may separate from the mean curve in the upper tail at finite n; in that region the mean curve is the more reliable reference.

Failed or nonconverged re-fits are discarded and replaced until nsim successful samples are obtained or max.attempts is reached. The simulation uses fitted values for every distribution parameter, so covariate-dependent parameters are retained. Automatic simulation and refitting of NVASIQ, LVASIQ, and HVASIQ recover the fixed level embedded in the fitted object; no global quantile-level variable is consulted. Consequently, fitted models at different quantile levels can be used in the same R session safely.

Automatic refitting assumes that data contains exactly the observations used by the original fit; if rows were dropped for missing values or via subset, supply a matching data or a custom refit.

Value

An object of class "vasicek_envelope". Its results component contains, for each requested residual, the theoretical order statistics, ordered observed residuals, pointwise lower, mean and upper curves, and the matrix of ordered simulated residuals. The object also records the number of successful simulations, attempts and failures.

References

Moral, R. A., Hinde, J., and Demetrio, C. G. B. (2017). Half-normal plots and overdispersed models in R: The hnp package. Journal of Statistical Software, 81(10), 1–23. doi:10.18637/jss.v081.i10

Zhao, Y., Lee, A. H., Yau, K. K. W., and McLachlan, G. J. (2011). Assessing the adequacy of Weibull survival models: A simulated envelope approach. Journal of Applied Statistics, 38, 2089–2097. doi:10.1080/02664763.2010.545115

Examples

## Not run: 
library(gamlss)
set.seed(123)
dat <- data.frame(y = rNVASIM(100, mu = 0.55, sigma = 0.30))
fit <- gamlss(y ~ 1, sigma.formula = ~ 1, family = NVASIM(),
              data = dat, trace = FALSE)

env <- vasicek_envelope(fit, nsim = 200, seed = 123)
plot(env, which = "quantile")
plot(env, which = "cox-snell")

## End(Not run)

Zero-Augmented Beta random-intercept model

Description

Fits a two-component model for responses in [0,1) with a point mass at zero (discrete component, parameterized by \gamma) and a Beta distribution for positive values (continuous component, parameterized by \beta).

Usage

zabr(
  data,
  y,
  formula_bin = NULL,
  formula_cont = NULL,
  random = NULL,
  logistic_cov = NULL,
  beta_cov = NULL,
  subject_ind = NULL,
  time_ind,
  component_wise_test = TRUE,
  quad_n = 30,
  verbose = FALSE,
  joint_test = NULL,
  sd_lower = 1e-05,
  start = NULL,
  control = list(),
  hessian = TRUE
)

Arguments

Details

Formula and legacy arguments are mutually exclusive for each model component: use either formula_bin or logistic_cov, either formula_cont or beta_cov, and either random or subject_ind. Supplying both arguments in any pair is an error.

For a response Y_{it} \in [0,1) observed on subject i (i = 1, \ldots, N) at time t (t = 1, \ldots, T), the model places a point mass at zero and a continuous component on (0,1):

Y_{it} = 0 \quad \mbox{with probability } 1 - p_{it},

Y_{it} \sim \mathrm{Beta}\left(\mu_{it}\phi,\; (1-\mu_{it})\phi\right) \quad \mbox{with probability } p_{it},

where 0 < p_{it} < 1, 0 < \mu_{it} < 1, and \phi > 0. Let X_{it} and Z_{it} be the covariate vectors entering the discrete and continuous components, respectively; they may share columns or be disjoint. Both components are modeled on the logit scale:

\mathrm{logit}(p_{it}) = \log\left(\frac{p_{it}}{1 - p_{it}}\right) = a_i + \gamma_0 + X_{it}^\top \gamma,

\mathrm{logit}(\mu_{it}) = \log\left(\frac{\mu_{it}}{1 - \mu_{it}}\right) = b_i + \beta_0 + Z_{it}^\top \beta,

where a_i and b_i are subject-specific random intercepts that induce correlation across repeated measurements on the same subject,

a_i \sim N(0, \sigma_1^2), \qquad b_i \sim N(0, \sigma_2^2).

The fixed-effect coefficients are denoted by \gamma (discrete component) and \beta (continuous component); \phi is the Beta precision parameter. The two random intercepts are independent, and the components have no shared parameters. Consequently, the marginal likelihood factorizes into discrete and continuous components. The two components are optimized separately; this is equivalent to maximizing their joint likelihood. Random effects are integrated out using non-adaptive Gauss–Hermite quadrature.

Value

An object of class "zabr" with components including logistic_est_table, beta_est_table, precision_table, random_effects, loglikelihood, joint_p, fit_statistics (joint model), fit_statistics_bin and fit_statistics_cont (per component; see section Fit statistics), and vcov. Fixed-effect names in coef() and vcov() use the prefix gamma_ for the discrete component and beta_ for the continuous component, including their intercepts.

Fit statistics

fit_statistics refers to the joint model and follows the "Fit Statistics" table of SAS PROC NLMIXED: AIC = -2\ell + 2k and BIC = -2\ell + k\log(s), where \ell is the maximized log-likelihood, k the total number of parameters and s the number of subjects. BIC uses subjects, not observations. AIC() and BIC() return the same values as fit_statistics.

fit_statistics_bin and fit_statistics_cont are informational. The continuous component counts only observations with Y > 0 and subjects with at least one such observation, as if it were fitted alone to the positive responses. Consequently, AIC is additive across components, whereas BIC is additive only when every subject has at least one positive response. To compare models, use fit_statistics or BIC(); do not sum component-wise values.

logLik() sets the nobs attribute to the number of subjects, so that BIC() follows PROC NLMIXED.

The values coincide numerically with PROC NLMIXED only when the log-likelihood coincides: the same model with independent random intercepts and the same number of parameters, non-adaptive quadrature with the same number of points (NOAD and QPOINTS= equal to quad_n), and data sorted by subject.

References

Chen, E. Z. and Li, H. (2016). A two-part mixed-effects model for analyzing longitudinal microbiome compositional data. Bioinformatics, 32(17), 2611–2617. doi:10.1093/bioinformatics/btw308

See Also

zavr

Examples

data(please_microbiome)

d <- subset(please_microbiome, Genus == "g__Bifidobacterium")
d$Subject <- factor(d$Subject)

fit <- zabr(
  data         = d,
  y            = "Y",
  formula_bin  = ~ Baseline + Time + Treat,
  formula_cont = ~ Baseline + Time + Treat,
  random       = ~ 1 | Subject,
  time_ind     = "Time"
)
fit

Zero-Augmented Vasicek random-intercept model

Description

Fits a two-component model for responses in [0,1) with a point mass at zero (discrete component, parameterized by \gamma) and a Vasicek distribution for positive values (continuous component, parameterized by \beta).

Usage

zavr(
  data,
  y,
  formula_bin = NULL,
  formula_cont = NULL,
  random = NULL,
  logistic_cov = NULL,
  vasicek_cov = NULL,
  subject_ind = NULL,
  time_ind,
  component_wise_test = TRUE,
  quad_n = 30,
  verbose = FALSE,
  joint_test = NULL,
  sd_lower = 1e-05,
  start = NULL,
  control = list(),
  hessian = TRUE
)

Arguments

Details

Formula and legacy arguments are mutually exclusive for each model component: use either formula_bin or logistic_cov, either formula_cont or vasicek_cov, and either random or subject_ind. Supplying both arguments in any pair is an error.

For a response Y_{it} \in [0,1) observed on subject i (i = 1, \ldots, N) at time t (t = 1, \ldots, T), the model places a point mass at zero and a continuous component on (0,1):

Y_{it} = 0 \quad \mbox{with probability } 1 - p_{it},

Y_{it} \sim \mathrm{NVASIM}\left(\mu_{it}, \sigma\right) \quad \mbox{with probability } p_{it},

where 0 < p_{it} < 1, 0 < \mu_{it} < 1, and \sigma \in (0,1). The continuous component is the normal-kernel Vasicek (mean-parameterized) distribution NVASIM, for which \mu_{it} = E(Y_{it} \mid Y_{it} > 0). Let X_{it} and Z_{it} be the covariate vectors entering the discrete and continuous components, respectively; they may share columns or be disjoint. Both components are modeled on the logit scale:

\mathrm{logit}(p_{it}) = \log\left(\frac{p_{it}}{1 - p_{it}}\right) = a_i + \gamma_0 + X_{it}^\top \gamma,

\mathrm{logit}(\mu_{it}) = \log\left(\frac{\mu_{it}}{1 - \mu_{it}}\right) = b_i + \beta_0 + Z_{it}^\top \beta,

where a_i and b_i are subject-specific random intercepts that induce correlation across repeated measurements on the same subject,

a_i \sim N(0, \sigma_1^2), \qquad b_i \sim N(0, \sigma_2^2).

The fixed-effect coefficients are denoted by \gamma (discrete component) and \beta (continuous component); \sigma is the shape parameter of the Vasicek component. The two random intercepts are independent, and the components have no shared parameters. Consequently, the marginal likelihood factorizes into discrete and continuous components. The two components are optimized separately; this is equivalent to maximizing their joint likelihood. Random effects are integrated out using non-adaptive Gauss–Hermite quadrature.

This function differs from zabr only in the continuous component: zabr uses a Beta distribution with precision \phi, whereas zavr uses the Vasicek NVASIM distribution with shape \sigma. The two models share the same discrete component and the same random-intercept structure.

Value

An object of class "zavr" with components including logistic_est_table, vasicek_est_table, shape_table, random_effects, loglikelihood, joint_p, fit_statistics (joint model), fit_statistics_bin and fit_statistics_cont (per component; see section Fit statistics), and vcov. Fixed-effect names in coef() and vcov() use the prefix gamma_ for the discrete component and beta_ for the continuous component, including their intercepts.

Fit statistics

fit_statistics refers to the joint model and follows the "Fit Statistics" table of SAS PROC NLMIXED: AIC = -2\ell + 2k and BIC = -2\ell + k\log(s), where \ell is the maximized log-likelihood, k the total number of parameters and s the number of subjects. BIC uses subjects, not observations. AIC() and BIC() return the same values as fit_statistics.

fit_statistics_bin and fit_statistics_cont are informational. The continuous component counts only observations with Y > 0 and subjects with at least one such observation, as if it were fitted alone to the positive responses. Consequently, AIC is additive across components, whereas BIC is additive only when every subject has at least one positive response. To compare models, use fit_statistics or BIC(); do not sum component-wise values.

logLik() sets the nobs attribute to the number of subjects, so that BIC() follows PROC NLMIXED.

The values coincide numerically with PROC NLMIXED only when the log-likelihood coincides: the same model with independent random intercepts and the same number of parameters, non-adaptive quadrature with the same number of points (NOAD and QPOINTS= equal to quad_n), and data sorted by subject.

References

Chen, E. Z. and Li, H. (2016). A two-part mixed-effects model for analyzing longitudinal microbiome compositional data. Bioinformatics, 32(17), 2611–2617. doi:10.1093/bioinformatics/btw308

Mazucheli, J., Alves, B., Korkmaz, M. Ç., and Leiva, V. (2022). Vasicek quantile and mean regression models for bounded data: New formulation, mathematical derivations, and numerical applications. Mathematics, 10, 1389. doi:10.3390/math10091389

See Also

zabr, NVASIM

Examples

data(please_microbiome)

d <- subset(please_microbiome, Genus == "g__Bifidobacterium")
d$Subject <- factor(d$Subject)

fit <- zavr(
  data         = d,
  y            = "Y",
  formula_bin  = ~ Baseline + Time + Treat,
  formula_cont = ~ Baseline + Time + Treat,
  random       = ~ 1 | Subject,
  time_ind     = "Time"
)
fit