Package 'glmmrBase'

Description

Specification, analysis, simulation, and fitting of generalised linear mixed models. Includes Markov Chain Monte Carlo Maximum likelihood and Laplace approximation model fitting for a range of models, non-linear fixed effect specifications, a wide range of flexible covariance functions that can be combined arbitrarily, robust and bias-corrected standard error estimation, power calculation, data simulation, and more. See <https://samuel-watson.github.io/glmmr-web/> for a detailed manual. glmmrBase provides functions for specifying, analysing, fitting, and simulating mixed models including linear, generalised linear, and models non-linear in fixed effects.

Differences between glmmrBase and lme4 and related packages.

glmmrBase is intended to be a broad package to support statistical work with generalised linear mixed models. While there are Laplace Approximation methods in the package, it does not intend to replace or supplant popular mixed model packages like lme4. Rather it provides broader functionality around simulation and analysis methods, and a range of model fitting algorithms not found in other mixed model packages. The key features are:

• Stochastic maximum likelihood methods. The most widely used methods for mixed model fitting are penalised quasi-likelihood, Laplace approximation, and Gaussian quadrature methods. These methods are widely available in other packages. We provide Markov Chain Monte Carlo (MCMC) Maximum Likelihood and Stochastic Approximation Expectation Maximisation algorithms for model fitting, with various features. These algorithms approximate the intractable GLMM likelihood using MCMC and so can provide an arbitrary level of precision. These methods may provide better maximum likelihood performance than other approximations in settings with high-dimensional or complex random effects, small sample sizes, or non-linear models.

• Flexible support for a wide range of covariance functions. The support for different covariance functions can be limited in other packages. For example, lme4 only provides exchangeable random effects structures. We include multiple different functions that can be combined arbitrarily.

• We similarly use efficient linear algebra methods with the Eigen package along with Stan to provide MCMC sampling.

• Gaussian Process approximations. We include Hibert Space and Nearest Neighbour Gaussian Process approximations for high dimensional random effects.

• The Model class includes methods for power estimation, data simulation, MCMC sampling, and calculation of a wide range of matrices and values associated with the models.

• We include natively a range of small sample corrections to information matrices, including Kenward-Roger, Box, Satterthwaite, and others, which typically require add-on packages for lme4.

• The package provides a flexible class system for specifying mixed models that can be incorporated into other packages and settings. The linked package glmmrOptim provides optimal experimental design algorithms for mixed models.

• (New in version 0.9.1) The package includes functions to replicate the functionality of lme4, mcml_lmer and mcml_glmer, which will also accept lme4 syntax.

• (New in version 0.10.1) The package also provides mixed quantile regression models estimated using the stochastic maximum likelihood algorithms described above. These models specify an asymmetric Laplace distribution for the likelihood and integrate with the other features of the package described above.

Package development

The package is still in development and there may still be bugs and errors. While we do not expect the general user interface to change there may be changes to the underlying library as well as new additions and functionality.

Author(s)

Sam Watson [aut, cre]

Maintainer: Sam Watson <[email protected]>

Description

Skeleton list to declare a Beta distribution in a 'Model' object

Usage

Beta(link = "logit")

Arguments

Value

A list with two elements naming the family and link function

Description

Extracts the fitted fixed effect coefficients from an 'mcml' object returned from a call of 'MCML' or 'LA' in the Model class.

Usage

## S3 method for class 'mcml'
coef(object, ...)

Arguments

Value

A named vector.

Description

Extracts the coefficients from a 'Model' object.

Usage

## S3 method for class 'Model'
coef(object, ...)

Arguments

Value

Fixed effect and covariance parameters extracted from the model object.

Description

Returns the computed confidence intervals for a 'mcml' object.

Usage

## S3 method for class 'mcml'
confint(object, ...)

Arguments

Value

A matrix (or vector) with columns giving lower and upper confidence limits for each parameter.

Description

R6 Class representing a covariance function and data

R6 Class representing a covariance function and data

Details

For the generalised linear mixed model

$Y \sim F(\mu,\sigma)$

$\mu = h^-1(X\beta + Z\gamma)$

$\gamma \sim MVN(0,D)$

where h is the link function, this class defines Z and D. The covariance is defined by a covariance function, data, and parameters. A new instance can be generated with $new(). The class will generate the relevant matrices Z and D automatically. See glmmrBase for a detailed guide on model specification.

**Intitialisation** A covariance function is specified as an additive formula made up of components with structure (1|f(j)). The left side of the vertical bar specifies the covariates in the model that have a random effects structure. The right side of the vertical bar specify the covariance function 'f' for that term using variable named in the data 'j'. Covariance functions on the right side of the vertical bar are multiplied together, i.e. (1|f(j)*g(t)).

There are several common functions included for a named variable in data x. A non-exhaustive list (see glmmrBase for a full list): * gr(x): Indicator function (1 parameter) * fexp(x): Exponential function (2 parameters) * ar(x): AR function (2 parameters) * sqexp(x): Squared exponential (1 parameter) * matern(x): Matern function (2 parameters) * bessel(x): Modified Bessel function of the 2nd kind (1 parameter) For many 2 parameter functions, such as 'ar' and 'fexp', alternative one parameter versions are also available as 'ar0' and 'fexp0'. These function omit the variance parameter and so can be used in combination with 'gr' functions such as 'gr(j)*ar0(t)'.

Parameters are provided to the covariance function as a vector. The parameters in the vector for each function should be provided in the order the covariance functions are written are written. For example, * Formula: '~(1|gr(j))+(1|gr(j*t))'; parameters: 'c(0.05,0.01)' * Formula: '~(1|gr(j)*fexp0(t))'; parameters: 'c(0.05,0.5)'

Updating of parameters is automatic if using the 'update_parameters()' member function.

Using 'update_parameters()' is the preferred way of updating the parameters of the mean or covariance objects as opposed to direct assignment, e.g. 'self$parameters <- c(...)'. The function calls check functions to automatically update linked matrices with the new parameters.

Public fields

Methods

Public methods

• Covariance$n()

• Covariance$new()

• Covariance$update_parameters()

• Covariance$print()

• Covariance$subset()

• Covariance$chol_D()

• Covariance$log_likelihood()

• Covariance$simulate_re()

• Covariance$sparse()

• Covariance$parameter_table()

• Covariance$nngp()

• Covariance$hsgp()

• Covariance$clone()

Method n()

Return the size of the design

Usage

Covariance$n()

Returns

Scalar

Method new()

Create a new Covariance object

Usage

Covariance$new(formula, data = NULL, parameters = NULL)

Arguments

Returns

A Covariance object

Examples

\dontshow{
setParallel(FALSE) # for the CRAN check
}
df <- nelder(~(cl(5)*t(5)) > ind(5))
cov <- Covariance$new(formula = ~(1|gr(cl)*ar0(t)),
                      parameters = c(0.05,0.7),
                      data= df)

Method update_parameters()

Updates the covariance parameters

Usage

Covariance$update_parameters(parameters)

Arguments

Method print()

Show details of Covariance object

Usage

Covariance$print()

Arguments

Examples

\dontshow{
setParallel(FALSE) # for the CRAN check
}
df <- nelder(~(cl(5)*t(5)) > ind(5))
Covariance$new(formula = ~(1|gr(cl)*ar0(t)),
                      parameters = c(0.05,0.8),
                      data= df)

Method subset()

Keep specified indices and removes the rest

Usage

Covariance$subset(index)

Arguments

Examples

\dontshow{
setParallel(FALSE) # for the CRAN check
}
df <- nelder(~(cl(10)*t(5)) > ind(10))
cov <- Covariance$new(formula = ~(1|gr(cl)*ar0(t)),
                      parameters = c(0.05,0.8),
                      data= df)
cov$subset(1:100)

Method chol_D()

Returns the Cholesky decomposition of the covariance matrix D

Usage

Covariance$chol_D()

Returns

A matrix

Method log_likelihood()

The function returns the values of the multivariate Gaussian log likelihood with mean zero and covariance D for a given vector of random effect terms.

Usage

Covariance$log_likelihood(u)

Arguments

Returns

Value of the log likelihood

Method simulate_re()

Simulates a set of random effects from the multivariate Gaussian distribution with mean zero and covariance D.

Usage

Covariance$simulate_re()

Returns

A vector of random effect values

Method sparse()

If this function is called then sparse matrix methods will be used for calculations involving D

Usage

Covariance$sparse(sparse = TRUE, amd = TRUE)

Arguments

Returns

None. Called for effects.

Method parameter_table()

Returns a table showing which parameters are members of which covariance function term.

Usage

Covariance$parameter_table()

Returns

A data frame

Method nngp()

Reports or sets the parameters for the nearest neighbour Gaussian process

Usage

Covariance$nngp(nn = NULL)

Arguments

Returns

If 'nn' is NULL then the function will either return FALSE if not using a Nearest neighbour approximation, or TRUE and the number of nearest neighbours, otherwise it will return nothing.

Method hsgp()

Reports or sets the parameters for the Hilbert Space Gaussian process

Usage

Covariance$hsgp(m = NULL, L = NULL)

Arguments

Returns

If 'm' and 'L' are NULL then the function will either return FALSE if not using a Hilbert space approximation, or TRUE and the number of bases functions and boundary value, otherwise it will return nothing.

Method clone()

The objects of this class are cloneable with this method.

Usage

Covariance$clone(deep = FALSE)

Arguments

Examples

## ------------------------------------------------
## Method `Covariance$new`
## ------------------------------------------------


df <- nelder(~(cl(5)*t(5)) > ind(5))
cov <- Covariance$new(formula = ~(1|gr(cl)*ar0(t)),
                      parameters = c(0.05,0.7),
                      data= df)

## ------------------------------------------------
## Method `Covariance$print`
## ------------------------------------------------


df <- nelder(~(cl(5)*t(5)) > ind(5))
Covariance$new(formula = ~(1|gr(cl)*ar0(t)),
                      parameters = c(0.05,0.8),
                      data= df)

## ------------------------------------------------
## Method `Covariance$subset`
## ------------------------------------------------


df <- nelder(~(cl(10)*t(5)) > ind(10))
cov <- Covariance$new(formula = ~(1|gr(cl)*ar0(t)),
                      parameters = c(0.05,0.8),
                      data= df)
cov$subset(1:100)

Description

Generate a data frame with crossed rows from two other data frames

Usage

cross_df(df1, df2)

Arguments

Details

For two data frames 'df1' and 'df2', the function will return another data frame that crosses them, which has rows with every unique combination of the input data frames

Value

data frame

Examples

cross_df(data.frame(t=1:4),data.frame(cl=1:3))

Description

Given input a, returns a length(a)^2 vector by cycling through the values of a

Usage

cycles(a)

Arguments

Value

vector

Description

Extracts the family from a 'mcml' object.

Usage

## S3 method for class 'mcml'
family(object, ...)

Arguments

Value

A family object.

Description

Extracts the family from a 'Model' object.

Usage

## S3 method for class 'Model'
family(object, ...)

Arguments

Value

A family object.

Description

Fitted values should not be generated directly from an 'mcml' object, rather fitted values should be generated using the original 'Model'. A message is printed to the user.

Usage

## S3 method for class 'mcml'
fitted(object, ...)

Arguments

Value

Nothing, called for effects, unless 'override' is TRUE, when it will return a vector of fitted values.

Description

Return fitted values. Does not account for the random effects. This function is a wrapper for 'Model$fitted()', which also provides a variety of additional options for generating fitted values from mixed models. For simulated values based on resampling random effects, see also 'Model$sim_data()'. To predict the values including random effects at a new location see also 'Model$predict()'.

Usage

## S3 method for class 'Model'
fitted(object, ...)

Arguments

Value

Fitted values

Description

Extracts the fixed effect estimates from an mcml object returned from call of 'MCML' or 'LA' in the Model class.

Usage

fixed.effects(object)

Arguments

Value

A named, numeric vector of fixed-effects estimates.

Description

Extracts the formula from a 'mcml' object. Separate formulae are specified for the fixed and random effects in the model, either of which can be returned. The complete formula is available from the generating 'Model' object as 'Model$formula' or 'formula(Model)'

Usage

## S3 method for class 'mcml'
formula(x, ...)

Arguments

Value

A formula object.

Description

Extracts the formula from a 'Model' object. This information can also be accessed directly from the Model as 'Model$formula'

Usage

## S3 method for class 'Model'
formula(x, ...)

Arguments

Value

A formula object.

Description

Exposes the automatic differentiator. Allows for calculation of Jacobian and Hessian matrices of formulae in terms of specified parameters. Formula specification is as a string. Data items are automatically multiplied by a parameter unless enclosed in parentheses.

Usage

hessian_from_formula(form_, data_, colnames_, parameters_)

Arguments

Value

A list including the jacobian and hessian matrices.

Examples

# obtain the Jacobian and Hessian of the log-binomial model log-likelihood. 
# The model is of data from an intervention and control group
# with n1 and n0 participants, respectively, with y1 and y0 the number of events in each group. 
# The mean is exp(alpha) in the control 
# group and exp(alpha + beta) in the intervention group, so that beta is the log relative risk.
hessian_from_formula(
  form_ = "(y1)*(a+b)+((n1)-(y1))*log((1-exp(a+b)))+(y0)*a+((n0)-(y0))*log((1-exp(a)))",
  data_ = matrix(c(10,100,20,100), nrow = 1),
  colnames_ = c("y1","n1","y0","n0"),
  parameters_ = c(log(0.1),log(0.5)))

Description

Returns a formula that can be used for glmmrBase Models from an lme4 input.

Usage

lme4_to_glmmr(formula, cnames)

Arguments

Details

The package lme4 uses a syntax to specify random effects as '(1|x)' where 'x' is the grouping variable. This function will modify such a formula, including those with nesting and crossing operators '/' and ':' into the glmmrBase syntax using the 'gr()' function. Not typically required by the user as it is used internally in the 'mcml_lmer' and 'mcml_glmer' functions.

Value

A formula.

Examples

df <- data.frame(cl = 1:3, t = 4:6)
f1 <- lme4_to_glmmr(y ~ x + (1|cl/t),colnames(df))

Description

Extracts the final log-likelihood value from an mcml object returned from call of 'MCML' or 'LA' in the Model class. The fitting algorithm estimates the fixed effects, random effects, and covariance parameters all separately. The log-likelihood is separable in the fixed and covariance parameters, so one can return the log-likelihood for either component, or the overall log-likelihood.

Usage

## S3 method for class 'mcml'
logLik(object, fixed = TRUE, covariance = TRUE, ...)

Arguments

Value

An object of class 'logLik'. If both 'fixed' and 'covariance' are FALSE then it returns NA.

Description

Extracts the log-likelihood value from an 'Model' object. If no data 'y' are specified then it returns NA.

Usage

## S3 method for class 'Model'
logLik(object, ...)

Arguments

Value

An object of class 'logLik'. If both 'fixed' and 'covariance' are FALSE then it returns NA.

Description

For a data frames 'x' and 'target', the function will return a matrix mapping the rows of 'x' to those of 'target'.

Usage

match_rows(x, target, by)

Arguments

Details

'x' is a data frame with n rows and 'target' a data frame with m rows. This function will return a n times m matrix that maps the rows of 'x' to those of 'target' based on the values in the columns specified by the argument 'by'

Value

A matrix with nrow(x) rows and nrow(target) columns

Examples

df <- nelder(~(cl(10)*t(5)) > ind(10))
df_unique <- df[!duplicated(df[,c('cl','t')]),]
match_rows(df,df_unique,c('cl','t'))

Description

A wrapper for Model stochastic maximum likelihood model fitting replicating lme4's syntax

Usage

mcml_glmer(
  formula,
  data,
  family,
  start = NULL,
  offset = NULL,
  verbose = 1L,
  iter.warmup = 100,
  iter.sampling = 50,
  weights = NULL,
  ...
)

Arguments

Details

This function aims to replicate the syntax of lme4's 'lmer' command. The specified formula can be the standard lme4 syntax, or alternatively a glmmrBase style formula can also be used to allow for the wider range of covariance function specifications. For example both 'y~x+(1|cl/t)' and 'y~x+(1|gr(cl))+(1|gr(cl)*ar1(t))' would be valid formulae.

Value

A 'mcml' model fit object.

Examples

#create a data frame describing a cross-sectional parallel cluster
#randomised trial
df <- nelder(~(cl(10)*t(5)) > ind(10))
df$int <- 0
df[df$cl > 5, 'int'] <- 1
# simulate data using the Model class
df$y <- Model$new(
  formula = ~ factor(t) + int - 1 + (1|gr(cl)) + (1|gr(cl,t)),
  data = df,
  family = stats::binomial()
)$sim_data()
## Not run: 
fit <- mcml_glmer(y ~ factor(t) + int - 1 + (1|cl/t), data = df, family = binomial())

## End(Not run)

Description

A wrapper for Model stochastic maximum likelihood model fitting replicating lme4's syntax

Usage

mcml_lmer(
  formula,
  data,
  start = NULL,
  offset = NULL,
  verbose = 1L,
  iter.warmup = 100,
  iter.sampling = 50,
  weights = NULL,
  ...
)

Arguments

Details

This function aims to replicate the syntax of lme4's 'lmer' command. The specified formula can be the standard lme4 syntax, or alternatively a glmmrBase style formula can also be used to allow for the wider range of covariance function specifications. For example both 'y~x+(1|cl/t)' and 'y~x+(1|gr(cl))+(1|gr(cl)*ar1(t))' would be valid formulae.

Value

A 'mcml' model fit object.

Examples

#create a data frame describing a cross-sectional parallel cluster
#randomised trial
df <- nelder(~(cl(10)*t(5)) > ind(10))
df$int <- 0
df[df$cl > 5, 'int'] <- 1
# simulate data using the Model class
df$y <- Model$new(
  formula = ~ factor(t) + int - 1 + (1|gr(cl)) + (1|gr(cl,t)),
  data = df,
  family = stats::gaussian()
)$sim_data()
## Not run: 
fit <- mcml_lmer(y ~ factor(t) + int - 1 + (1|cl/t), data = df)

## End(Not run)

Description

Returns the file name and type for MCNR function

Usage

mcnr_family(family, cmdstan)

Arguments

Value

list with filename and type

Description

R6 Class representing a mean function/linear predictor

R6 Class representing a mean function/linear predictor

Details

For the generalised linear mixed model

$Y \sim F(\mu,\sigma)$

$\mu = h^-1(X\beta + Z\gamma)$

$\gamma \sim MVN(0,D)$

this class defines the fixed effects design matrix X. The mean function is defined by a model formula, data, and parameters. A new instance can be generated with $new(). The class will generate the relevant matrix X automatically. See glmmrBase for a detailed guide on model specification.

Specification of the mean function follows standard model formulae in R. For example for a stepped-wedge cluster trial model, a typical mean model is $E(y_{ijt}|\delta)=\beta_0 + \tau_t + \beta_1 d_{jt} + z_{ijt}\delta$ where $\tau_t$ are fixed effects for each time period. The formula specification for this would be '~ factor(t) + int' where 'int' is the name of the variable indicating the treatment.

One can also include non-linear functions of variables in the mean function, and name the parameters. The resulting X matrix is then a matrix of first-order partial derivatives. For example, one can specify '~ int + b_1*exp(b_2*x)'.

Using 'update_parameters()' is the preferred way of updating the parameters of the mean or covariance objects as opposed to direct assignment, e.g. 'self$parameters <- c(...)'. The function calls check functions to automatically update linked matrices with the new parameters.

Public fields

Methods

Public methods

• MeanFunction$n()

• MeanFunction$new()

• MeanFunction$print()

• MeanFunction$update_parameters()

• MeanFunction$colnames()

• MeanFunction$subset_rows()

• MeanFunction$linear_predictor()

• MeanFunction$any_nonlinear()

• MeanFunction$clone()

Method n()

Returns the number of observations

Usage

MeanFunction$n()

Arguments

Returns

The number of observations in the model

Examples

\dontshow{
setParallel(FALSE) # for the CRAN check
}
df <- nelder(~(cl(4)*t(5)) > ind(5))
df$int <- 0
df[df$cl <= 2, 'int'] <- 1
mf1 <- MeanFunction$new(formula = ~ int ,
                        data=df,
                        parameters = c(-1,1)
                        )
mf1$n()

Method new()

Create a new MeanFunction object

Usage

MeanFunction$new(
  formula,
  data,
  parameters = NULL,
  offset = NULL,
  verbose = FALSE
)

Arguments

Returns

A MeanFunction object

Examples

\dontshow{
setParallel(FALSE) # for the CRAN check
}
df <- nelder(~(cl(4)*t(5)) > ind(5))
df$int <- 0
df[df$cl <= 2, 'int'] <- 1
mf1 <- MeanFunction$new(formula = ~ int ,
                        data=df,
                        parameters = c(-1,1),
                        )

Method print()

Prints details about the object

Usage

MeanFunction$print()

Arguments

Method update_parameters()

Updates the model parameters

Usage

MeanFunction$update_parameters(parameters)

Arguments

Method colnames()

Returns or replaces the column names of the data in the object

Usage

MeanFunction$colnames(names = NULL)

Arguments

Examples

\dontshow{
setParallel(FALSE) # for the CRAN check
}
df <- nelder(~(cl(4)*t(5)) > ind(5))
df$int <- 0
df[df$cl <= 5, 'int'] <- 1
mf1 <- MeanFunction$new(formula = ~ int ,
                        data=df,
                        parameters = c(-1,1)
                        )
mf1$colnames(c("cluster","time","individual","treatment"))
mf1$colnames()

Method subset_rows()

Keeps a subset of the data and removes the rest

All indices not in the provided vector of row numbers will be removed from both the data and fixed effects design matrix X.

Usage

MeanFunction$subset_rows(index)

Arguments

Returns

NULL

Examples

\dontshow{
setParallel(FALSE) # for the CRAN check
}
df <- nelder(~(cl(4)*t(5)) > ind(5))
df$int <- 0
df[df$cl <= 5, 'int'] <- 1
mf1 <- MeanFunction$new(formula = ~ int ,
                        data=df,
                        parameters = c(-1,1)
                        )
mf1$subset_rows(1:20) 

Method linear_predictor()

Returns the linear predictor

Returns the linear predictor, X * beta

Usage

MeanFunction$linear_predictor()

Returns

A vector

Method any_nonlinear()

Returns a logical indicating whether the mean function contains non-linear functions of model parameters. Mainly used internally.

Usage

MeanFunction$any_nonlinear()

Returns

None. Called for effects

Method clone()

The objects of this class are cloneable with this method.

Usage

MeanFunction$clone(deep = FALSE)

Arguments

Examples

## ------------------------------------------------
## Method `MeanFunction$n`
## ------------------------------------------------


df <- nelder(~(cl(4)*t(5)) > ind(5))
df$int <- 0
df[df$cl <= 2, 'int'] <- 1
mf1 <- MeanFunction$new(formula = ~ int ,
                        data=df,
                        parameters = c(-1,1)
                        )
mf1$n()

## ------------------------------------------------
## Method `MeanFunction$new`
## ------------------------------------------------


df <- nelder(~(cl(4)*t(5)) > ind(5))
df$int <- 0
df[df$cl <= 2, 'int'] <- 1
mf1 <- MeanFunction$new(formula = ~ int ,
                        data=df,
                        parameters = c(-1,1),
                        )

## ------------------------------------------------
## Method `MeanFunction$colnames`
## ------------------------------------------------


df <- nelder(~(cl(4)*t(5)) > ind(5))
df$int <- 0
df[df$cl <= 5, 'int'] <- 1
mf1 <- MeanFunction$new(formula = ~ int ,
                        data=df,
                        parameters = c(-1,1)
                        )
mf1$colnames(c("cluster","time","individual","treatment"))
mf1$colnames()

## ------------------------------------------------
## Method `MeanFunction$subset_rows`
## ------------------------------------------------


df <- nelder(~(cl(4)*t(5)) > ind(5))
df$int <- 0
df[df$cl <= 5, 'int'] <- 1
mf1 <- MeanFunction$new(formula = ~ int ,
                        data=df,
                        parameters = c(-1,1)
                        )
mf1$subset_rows(1:20)

Description

A GLMM Model

A GLMM Model

Details

A generalised linear mixed model

See glmmrBase-package for a more in-depth guide.

The generalised linear mixed model is:

$Y \sim F(\mu,\sigma)$

$\mu = h^-1(X\beta + Zu)$

$u \sim MVN(0,D)$

where F is a distribution with scale parameter

$\sigma$

, h is a link function, X are the fixed effects with parameters

$\beta$

, Z is the random effect design matrix with multivariate Gaussian distributed effects u. The class provides access to all of the elements of the model above and associated calculations and functions including model fitting, power analysis, and various relevant matrices, including information matrices and related corrections. The object is an R6 class and so can serve as a parent class for extended functionality.

The currently support families (links) are Gaussian (identity, log), Binomial (logit, log, probit, identity), Poisson (log, identity), Gamma (logit, identity, inverse), and Beta (logit). The class also supports mixed quantile regression models, although functionality is currently experimental. Quantile models use an asymmetrical Laplace distribution for the likelihood and support all five link functions listed before.

Many calculations use the covariance matrix of the observations, such as the information matrix, which is used in power calculations and other functions. For non-Gaussian models, the class uses the first-order approximation proposed by Breslow and Clayton (1993) based on the marginal quasilikelihood:

$\Sigma = W^{-1} + ZDZ^T$

where W is a diagonal matrix with the GLM iterated weights for each observation equal to, for individual i $\left( \frac{(\partial h^{-1}(\eta_i))}{\partial \eta_i}\right) ^2 Var(y|u)$ (see Table 2.1 in McCullagh and Nelder (1989)). The modification proposed by Zegers et al to the linear predictor to improve the accuracy of approximations based on the marginal quasilikelihood is also available, see use_attenuation().

See glmmrBase for a detailed guide on model specification. A detailed vingette for this package is also available onlinedoi:10.48550/arXiv.2303.12657.

Attenuation For calculations such as the information matrix, the first-order approximation to the covariance matrix proposed by Breslow and Clayton (1993), described above, is used. The approximation is based on the marginal quasilikelihood. Zegers, Liang, and Albert (1988) suggest that a better approximation to the marginal mean is achieved by "attenuating" the linear predictor. Setting use equal to TRUE uses this adjustment for calculations using the covariance matrix for non-linear models.

Calls the respective print methods of the linked covariance and mean function objects.

The matrices X and Z both have n rows, where n is the number of observations in the model/design.

Using update_parameters() is the preferred way of updating the parameters of the mean or covariance objects as opposed to direct assignment, e.g. self$covariance$parameters <- c(...). The function calls check functions to automatically update linked matrices with the new parameters.

Stochastic maximum likelihood Fits generalised linear mixed models using one of several algorithms: Markov Chain Newton Raphson (MCNR), Markov Chain Expectation Maximisation (MCEM), or stochastic approximation expectation maximisation (SAEM) with or without Polyak-Ruppert averaging. MCNR and MCEM are described by McCulloch (1997) doi:10.1080/01621459.1997.10473613. For each iteration of the algorithms the unobserved random effect terms ($\gamma$) are simulated using Markov Chain Monte Carlo (MCMC) methods, and then these values are conditioned on in the subsequent steps to estimate the covariance parameters and the mean function parameters ($\beta$). SAEM uses a Robbins-Munroe approach to approximating the likelihood and requires fewer MCMC samples and may have lower Monte Carlo error, see Jank (2006)doi:10.1198/106186006X157469. The option alpha determines the rate at which succesive iterations "forget" the past and must be between 0.5 and 1. Higher values will result in lower Monte Carlo error but slower convergence. The options mcem.adapt and mcnr.adapt will modify the number of MCMC samples during each step of model fitting using the suggested values in Caffo, Jank, and Jones (2006)doi:10.1111/j.1467-9868.2005.00499.x as the estimates converge.

The accuracy of the algorithm depends on the user specified tolerance. For higher levels of tolerance, larger numbers of MCMC samples are likely need to sufficiently reduce Monte Carlo error. However, the SAEM approach does overcome reduce the required samples, especially with R-P averaging. As such a lower number (20-50) samples per iteration is normally sufficient to get convergence.

There are several stopping rules for the algorithm. Either the algorithm will terminate when succesive parameter estimates are all within a specified tolerance of each other (conv.criterion = 1), or when there is a high probability that the estimated log-likelihood has not been improved. This latter criterion can be applied to either the overall log-likelihood (conv.criterion = 2), the likelihood just for the fixed effects (conv.criterion = 3), or both the likelihoods for the fixed effects and covariance parameters (conv.criterion = 4; default).

Options for the MCMC sampler are set by changing the values in self$mcmc_options. The information printed to the console during model fitting can be controlled with the self$set_trace() function.

To provide weights for the model fitting, store them in self$weights. To set the number of trials for binomial models, set self$trials.

Laplace approximation Fits generalised linear mixed models using Laplace approximation to the log likelihood. For non-Gaussian models the covariance matrix is approximated using the first order approximation based on the marginal quasilikelihood proposed by Breslow and Clayton (1993). The marginal mean in this approximation can be further adjusted following the proposal of Zeger et al (1988), use the member function use_attenuated() in this class, see Model. To provide weights for the model fitting, store them in self$weights. To set the number of trials for binomial models, set self$trials. To control the information printed to the console during model fitting use the self$set_trace() function.

Public fields

Methods

Public methods

• Model$use_attenuation()

• Model$fitted()

• Model$residuals()

• Model$predict()

• Model$new()

• Model$print()

• Model$n()

• Model$subset_rows()

• Model$sim_data()

• Model$update_parameters()

• Model$information_matrix()

• Model$sandwich()

• Model$small_sample_correction()

• Model$box()

• Model$power()

• Model$w_matrix()

• Model$dh_deta()

• Model$Sigma()

• Model$MCML()

• Model$LA()

• Model$sparse()

• Model$mcmc_sample()

• Model$gradient()

• Model$partial_sigma()

• Model$u()

• Model$log_likelihood()

• Model$calculator_instructions()

• Model$marginal()

• Model$update_y()

• Model$set_trace()

• Model$clone()

Method use_attenuation()

Sets the model to use or not use "attenuation" when calculating the first-order approximation to the covariance matrix.

Usage

Model$use_attenuation(use)

Arguments

Returns

None. Used for effects.

Method fitted()

Return fitted values. Does not account for the random effects. For simulated values based on resampling random effects, see also sim_data(). To predict the values including random effects at a new location see also predict().

Usage

Model$fitted(type = "link", X, u, sample = FALSE, sample_n = 100)

Arguments

Returns

Fitted values as either a vector or matrix depending on the number of samples

Method residuals()

Generates the residuals for the model

Generates one of several types of residual for the model. If conditional = TRUE then the residuals include the random effects, otherwise only the fixed effects are included. For type, there are raw, pearson, and standardized residuals. For conditional residuals a matrix is returned with each column corresponding to a sample of the random effects.

Usage

Model$residuals(type = "standardized", conditional = TRUE)

Arguments

Returns

A matrix with either one column is conditional is false, or with number of columns corresponding to the number of MCMC samples.

Method predict()

Generate predictions at new values

Generates predicted values using a new data set to specify covariance values and values for the variables that define the covariance function. The function will return a list with the linear predictor, conditional distribution of the new random effects term conditional on the current estimates of the random effects, and some simulated values of the random effects if requested.

Usage

Model$predict(newdata, offset = rep(0, nrow(newdata)), m = 0)

Arguments

Returns

A list with the linear predictor, parameters (mean and covariance matrices) for the conditional distribution of the random effects, and any random effect samples.

Method new()

Create a new Model object. Typically, a model is generated from a formula and data. However, it can also be generated from a previous model fit.

Usage

Model$new(
  formula,
  covariance,
  mean,
  data = NULL,
  family = NULL,
  var_par = NULL,
  offset = NULL,
  weights = NULL,
  trials = NULL,
  model_fit = NULL
)

Arguments

Returns

A new Model class object

Examples

\dontshow{
setParallel(FALSE) 
}
#create a data frame describing a cross-sectional parallel cluster
#randomised trial
df <- nelder(~(cl(10)*t(5)) > ind(10))
df$int <- 0
df[df$cl > 5, 'int'] <- 1
mod <- Model$new(
  formula = ~ factor(t) + int - 1 + (1|gr(cl)) + (1|gr(cl,t)),
  data = df,
  family = stats::gaussian()
)

# We can also include the outcome data in the model initialisation. 
# For example, simulating data and creating a new object:
df$y <- mod$sim_data()

mod <- Model$new(
  formula = y ~ factor(t) + int - 1 + (1|gr(cl)) + (1|gr(cl,t)),
  data = df,
  family = stats::gaussian()
)

# Here we will specify a cohort study
df <- nelder(~ind(20) * t(6))
df$int <- 0
df[df$t > 3, 'int'] <- 1

des <- Model$new(
  formula = ~ int + (1|gr(ind)),
  data = df,
  family = stats::poisson()
)
  
# or with parameter values specified
  
des <- Model$new(
  formula = ~ int + (1|gr(ind)),
  covariance = c(0.05),
  mean = c(1,0.5),
  data = df,
  family = stats::poisson()
)

#an example of a spatial grid with two time points

df <- nelder(~ (x(10)*y(10))*t(2))
spt_design <- Model$new(formula = ~ 1 + (1|ar0(t)*fexp(x,y)),
                        data = df,
                        family = stats::gaussian())

# A quantile regression model for the 25th percentile (experimental)
df <- nelder(~(cl(10)*t(5)) > ind(10))
df$int <- 0
df[df$cl > 5, 'int'] <- 1

qmod <- Model$new(
  formula = ~ factor(t) + int - 1 + (1|gr(cl)) + (1|gr(cl,t)),
  data = df,
  family = Quantile(link = "identity", scaled = TRUE, q = 0.25)
)

Method print()

Print method for Model class

Usage

Model$print()

Arguments

Method n()

Returns the number of observations in the model

Usage

Model$n(...)

Arguments

Method subset_rows()

Subsets the design keeping specified observations only

Given a vector of row indices, the corresponding rows will be kept and the other rows will be removed from the mean function and covariance

Usage

Model$subset_rows(index)

Arguments

Returns

The function updates the object and nothing is returned.

Method sim_data()

Generates a realisation of the design

Generates a single vector of outcome data based upon the specified GLMM design.

Usage

Model$sim_data(type = "y")

Arguments

Returns

Either a vector, a data frame, or a list

Examples

df <- nelder(~(cl(10)*t(5)) > ind(10))
df$int <- 0
df[df$cl > 5, 'int'] <- 1
\dontshow{
setParallel(FALSE) # for the CRAN check
}
des <- Model$new(
  formula = ~ factor(t) + int - 1 + (1|gr(cl)*ar0(t)),
  covariance = c(0.05,0.8),
  mean = c(rep(0,5),0.6),
  data = df,
  family = stats::binomial()
)
ysim <- des$sim_data()

Method update_parameters()

Updates the parameters of the mean function and/or the covariance function

Usage

Model$update_parameters(mean.pars = NULL, cov.pars = NULL, var.par = NULL)

Arguments

Examples

\dontshow{
setParallel(FALSE) # for the CRAN check
}
df <- nelder(~(cl(10)*t(5)) > ind(10))
df$int <- 0
df[df$cl > 5, 'int'] <- 1
des <- Model$new(
  formula = ~ factor(t) + int - 1 + (1|gr(cl)*ar0(t)),
  data = df,
  family = stats::binomial()
)
des$update_parameters(cov.pars = c(0.1,0.9))

Method information_matrix()

Generates the information matrix of the mixed model GLS estimator (X'S^-1X). The inverse of this matrix is an estimator for the variance-covariance matrix of the fixed effect parameters. For various small sample corrections see small_sample_correction() and box(). For models with non-linear functions of fixed effect parameters, a correction to the Hessian matrix is required, which is automatically calculated or optionally returned or disabled.

Usage

Model$information_matrix(
  include.re = FALSE,
  theta = FALSE,
  hessian.corr = "add",
  adj.nonspd = TRUE
)

Arguments

Returns

A matrix

Method sandwich()

Returns the robust sandwich variance-covariance matrix for the fixed effect parameters

Usage

Model$sandwich()

Returns

A PxP matrix

Method small_sample_correction()

Returns a small sample correction. The option "KR" returns the Kenward-Roger bias-corrected variance-covariance matrix for the fixed effect parameters and degrees of freedom. Option "KR2" returns an improved correction given in Kenward & Roger (2009) doi:j.csda.2008.12.013. Note, that the corrected/improved version is invariant under reparameterisation of the covariance, and it will also make no difference if the covariance is linear in parameters. Exchangeable covariance structures in this package (i.e. gr()) are parameterised in terms of the variance rather than standard deviation, so the results will be unaffected. Option "sat" returns the "Satterthwaite" correction, which only includes corrected degrees of freedom, along with the GLS standard errors.

Usage

Model$small_sample_correction(type)

Arguments

Returns

A PxP matrix

Method box()

Returns the inferential statistics (F-stat, p-value) for a modified Box correction doi:10.1002/sim.4072 for Gaussian-identity models.

Usage

Model$box(y)

Arguments

Returns

A data frame.

Method power()

Estimates the power of the design described by the model using the square root of the relevant element of the GLS variance matrix:

$(X^T\Sigma^{-1}X)^{-1}$

Note that this is equivalent to using the "design effect" for many models.

Usage

Model$power(alpha = 0.05, two.sided = TRUE, alternative = "pos")

Arguments

Returns

A data frame describing the parameters, their values, expected standard errors and estimated power.

Examples

\dontshow{
setParallel(FALSE) # for the CRAN check
}
df <- nelder(~(cl(10)*t(5)) > ind(10))
df$int <- 0
df[df$cl > 5, 'int'] <- 1
des <- Model$new(
  formula = ~ factor(t) + int - 1 + (1|gr(cl)) + (1|gr(cl,t)),
  covariance = c(0.05,0.1),
  mean = c(rep(0,5),0.6),
  data = df,
  family = stats::gaussian(),
  var_par = 1
)
des$power() #power of 0.90 for the int parameter

Method w_matrix()

Returns the diagonal of the matrix W used to calculate the covariance matrix approximation

Usage

Model$w_matrix()

Returns

A vector with values of the glm iterated weights

Method dh_deta()

Returns the derivative of the link function with respect to the linear preditor

Usage

Model$dh_deta()

Returns

A vector

Method Sigma()

Returns the (approximate) covariance matrix of y

Returns the covariance matrix Sigma. For non-linear models this is an approximation. See Details.

Usage

Model$Sigma(inverse = FALSE)

Arguments

Returns

A matrix.

Method MCML()

Stochastic Maximum Likelihood model fitting

Usage

Model$MCML(
  y = NULL,
  method = "saem",
  tol = 0.01,
  max.iter = 50,
  se = "gls",
  mcmc.pkg = "rstan",
  se.theta = TRUE,
  algo = ifelse(self$mean$any_nonlinear(), 2, 3),
  lower.bound = NULL,
  upper.bound = NULL,
  lower.bound.theta = NULL,
  upper.bound.theta = NULL,
  alpha = 0.8,
  convergence.prob = 0.95,
  pr.average = FALSE,
  conv.criterion = 2
)

Arguments

Returns

A mcml object

Examples

\dontrun{
#create example data with six clusters, five time periods, and five people per cluster-period
df <- nelder(~(cl(6)*t(5)) > ind(5))
# parallel trial design intervention indicator
df$int <- 0
df[df$cl > 3, 'int'] <- 1 
# specify parameter values in the call for the data simulation below
des <- Model$new(
  formula= ~ factor(t) + int - 1 +(1|gr(cl)*ar0(t)),
  covariance = c(0.05,0.7),
  mean = c(rep(0,5),0.2),
  data = df,
  family = gaussian()
)
ysim <- des$sim_data() # simulate some data from the model
fit1 <- des$MCML(y = ysim) # Default model fitting with SAEM-PR
# use MCEM instead and stop when parameter values are within 1e-2 on successive iterations
fit2 <- des$MCML(y = ysim, method="mcem",tol=1e-2,conv.criterion = 1)
}

Method LA()

Maximum Likelihood model fitting with Laplace Approximation

Usage

Model$LA(
  y = NULL,
  start,
  method = "nr",
  se = "gls",
  max.iter = 40,
  tol = 1e-04,
  se.theta = TRUE,
  algo = 2,
  lower.bound = NULL,
  upper.bound = NULL,
  lower.bound.theta = NULL,
  upper.bound.theta = NULL
)

Arguments

Returns

A mcml object

Examples

\dontshow{
setParallel(FALSE) # for the CRAN check
}
#create example data with six clusters, five time periods, and five people per cluster-period
df <- nelder(~(cl(6)*t(5)) > ind(5))
# parallel trial design intervention indicator
df$int <- 0
df[df$cl > 3, 'int'] <- 1 
# specify parameter values in the call for the data simulation below
des <- Model$new(
  formula = ~ factor(t) + int - 1 + (1|gr(cl)*ar0(t)),
  covariance = c(0.05,0.7),
  mean = c(rep(0,5),-0.2),
  data = df,
  family = stats::binomial()
)
ysim <- des$sim_data() # simulate some data from the model
fit1 <- des$LA(y = ysim)

Method sparse()

Set whether to use sparse matrix methods for model calculations and fitting. By default the model does not use sparse matrix methods.

Usage

Model$sparse(sparse = TRUE, amd = TRUE)

Arguments

Returns

None, called for effects

Method mcmc_sample()

Generate an MCMC sample of the random effects

Usage

Model$mcmc_sample(mcmc.pkg = "rstan")

Arguments

Returns

A matrix of samples of the random effects

Method gradient()

The gradient of the log-likelihood with respect to either the random effects or the model parameters. The random effects are on the N(0,I) scale, i.e. scaled by the Cholesky decomposition of the matrix D. To obtain the random effects from the last model fit, see member function ⁠$u⁠

Usage

Model$gradient(y, u, beta = FALSE)

Arguments

Returns

A vector of the gradient

Method partial_sigma()

The partial derivatives of the covariance matrix Sigma with respect to the covariance parameters. The function returns a list in order: Sigma, first order derivatives, second order derivatives. The second order derivatives are ordered as the lower-triangular matrix in column major order. Letting 'd(i)' mean the first-order partial derivative with respect to parameter i, and d2(i,j) mean the second order derivative with respect to parameters i and j, then if there were three covariance parameters the order of the output would be: (sigma, d(1), d(2), d(3), d2(1,1), d2(1,2), d2(1,3), d2(2,2), d2(2,3), d2(3,3)).

Usage

Model$partial_sigma()

Returns

A list of matrices, see description for contents of the list.

Method u()

Returns the sample of random effects from the last model fit, or updates the samples for the model.

Usage

Model$u(scaled = TRUE, u)

Arguments

Returns

A matrix of random effect samples

Method log_likelihood()

The log likelihood for the GLMM. The random effects can be left unspecified. If no random effects are provided, and there was a previous model fit with the same data y then the random effects will be taken from that model. If there was no previous model fit then the random effects are assumed to be all zero.

Usage

Model$log_likelihood(y, u)

Arguments

Returns

The log-likelihood of the model parameters

Method calculator_instructions()

Prints the internal instructions and data used to calculate the linear predictor. Internally the class uses a reverse polish notation to store and calculate different functions, including user-specified non-linear mean functions. This function will print all the steps. Mainly used for debugging and determining how the class has interpreted non-linear model specifications.

Usage

Model$calculator_instructions()

Returns

None. Called for effects.

Method marginal()

Calculates the marginal effect of variable x. There are several options for marginal effect and several types of conditioning or averaging. The type of marginal effect can be the derivative of the mean with respect to x (dydx), the expected difference E(y|x=a)-E(y|x=b) (diff), or the expected log ratio log(E(y|x=a)/E(y|x=b)) (ratio). Other fixed effect variables can be set at specific values (at), set at their mean values (atmeans), or averaged over (average). Averaging over a fixed effects variable here means using all observed values of the variable in the relevant calculation. The random effects can similarly be set at their estimated value (re="estimated"), set to zero (re="zero"), set to a specific value (re="at"), or averaged over (re="average"). Estimates of the expected values over the random effects are generated using MCMC samples. MCMC samples are generated either through MCML model fitting or using mcmc_sample. In the absence of samples average and estimated will produce the same result. The standard errors are calculated using the delta method with one of several options for the variance matrix of the fixed effect parameters. Several of the arguments require the names of the variables as given to the model object. Most variables are as specified in the formula, factor variables are specified as the name of the variable_value, e.g. t_1. To see the names of the stored parameters and data variables see the member function names().

Usage

Model$marginal(
  x,
  type,
  re,
  se,
  at = c(),
  atmeans = c(),
  average = c(),
  xvals = c(1, 0),
  atvals = c(),
  revals = c()
)

Arguments

Returns

A named vector with elements margin specifying the point estimate and se giving the standard error.

Method update_y()

Updates the outcome data y

Some functions require outcome data, which is by default set to all zero if no model fitting function has been run. This function can update the interval y data.

Usage

Model$update_y(y)

Arguments

Returns

None. Called for effects

Method set_trace()

Controls the information printed to the console for other functions.

Usage

Model$set_trace(trace)

Arguments

Returns

None. Called for effects.

Method clone()

The objects of this class are cloneable with this method.

Usage

Model$clone(deep = FALSE)

Arguments

References

Breslow, N. E., Clayton, D. G. (1993). Approximate Inference in Generalized Linear Mixed Models. Journal of the American Statistical Association<, 88(421), 9–25. doi:10.1080/01621459.1993.10594284

McCullagh P, Nelder JA (1989). Generalized linear models, 2nd Edition. Routledge.

McCulloch CE (1997). “Maximum Likelihood Algorithms for Generalized Linear Mixed Models.” Journal of the American statistical Association, 92(437), 162–170.doi:10.2307/2291460

Zeger, S. L., Liang, K.-Y., Albert, P. S. (1988). Models for Longitudinal Data: A Generalized Estimating Equation Approach. Biometrics, 44(4), 1049.doi:10.2307/2531734

See Also

nelder, MeanFunction, Covariance

Model, Covariance, MeanFunction

Model, Covariance, MeanFunction

Examples

## ------------------------------------------------
## Method `Model$new`
## ------------------------------------------------


#create a data frame describing a cross-sectional parallel cluster
#randomised trial
df <- nelder(~(cl(10)*t(5)) > ind(10))
df$int <- 0
df[df$cl > 5, 'int'] <- 1
mod <- Model$new(
  formula = ~ factor(t) + int - 1 + (1|gr(cl)) + (1|gr(cl,t)),
  data = df,
  family = stats::gaussian()
)

# We can also include the outcome data in the model initialisation. 
# For example, simulating data and creating a new object:
df$y <- mod$sim_data()

mod <- Model$new(
  formula = y ~ factor(t) + int - 1 + (1|gr(cl)) + (1|gr(cl,t)),
  data = df,
  family = stats::gaussian()
)

# Here we will specify a cohort study
df <- nelder(~ind(20) * t(6))
df$int <- 0
df[df$t > 3, 'int'] <- 1

des <- Model$new(
  formula = ~ int + (1|gr(ind)),
  data = df,
  family = stats::poisson()
)
  
# or with parameter values specified
  
des <- Model$new(
  formula = ~ int + (1|gr(ind)),
  covariance = c(0.05),
  mean = c(1,0.5),
  data = df,
  family = stats::poisson()
)

#an example of a spatial grid with two time points

df <- nelder(~ (x(10)*y(10))*t(2))
spt_design <- Model$new(formula = ~ 1 + (1|ar0(t)*fexp(x,y)),
                        data = df,
                        family = stats::gaussian())

# A quantile regression model for the 25th percentile (experimental)
df <- nelder(~(cl(10)*t(5)) > ind(10))
df$int <- 0
df[df$cl > 5, 'int'] <- 1

qmod <- Model$new(
  formula = ~ factor(t) + int - 1 + (1|gr(cl)) + (1|gr(cl,t)),
  data = df,
  family = Quantile(link = "identity", scaled = TRUE, q = 0.25)
)

## ------------------------------------------------
## Method `Model$sim_data`
## ------------------------------------------------

df <- nelder(~(cl(10)*t(5)) > ind(10))
df$int <- 0
df[df$cl > 5, 'int'] <- 1

des <- Model$new(
  formula = ~ factor(t) + int - 1 + (1|gr(cl)*ar0(t)),
  covariance = c(0.05,0.8),
  mean = c(rep(0,5),0.6),
  data = df,
  family = stats::binomial()
)
ysim <- des$sim_data()

## ------------------------------------------------
## Method `Model$update_parameters`
## ------------------------------------------------


df <- nelder(~(cl(10)*t(5)) > ind(10))
df$int <- 0
df[df$cl > 5, 'int'] <- 1
des <- Model$new(
  formula = ~ factor(t) + int - 1 + (1|gr(cl)*ar0(t)),
  data = df,
  family = stats::binomial()
)
des$update_parameters(cov.pars = c(0.1,0.9))

## ------------------------------------------------
## Method `Model$power`
## ------------------------------------------------


df <- nelder(~(cl(10)*t(5)) > ind(10))
df$int <- 0
df[df$cl > 5, 'int'] <- 1
des <- Model$new(
  formula = ~ factor(t) + int - 1 + (1|gr(cl)) + (1|gr(cl,t)),
  covariance = c(0.05,0.1),
  mean = c(rep(0,5),0.6),
  data = df,
  family = stats::gaussian(),
  var_par = 1
)
des$power() #power of 0.90 for the int parameter

## ------------------------------------------------
## Method `Model$MCML`
## ------------------------------------------------

## Not run: 
#create example data with six clusters, five time periods, and five people per cluster-period
df <- nelder(~(cl(6)*t(5)) > ind(5))
# parallel trial design intervention indicator
df$int <- 0
df[df$cl > 3, 'int'] <- 1 
# specify parameter values in the call for the data simulation below
des <- Model$new(
  formula= ~ factor(t) + int - 1 +(1|gr(cl)*ar0(t)),
  covariance = c(0.05,0.7),
  mean = c(rep(0,5),0.2),
  data = df,
  family = gaussian()
)
ysim <- des$sim_data() # simulate some data from the model
fit1 <- des$MCML(y = ysim) # Default model fitting with SAEM-PR
# use MCEM instead and stop when parameter values are within 1e-2 on successive iterations
fit2 <- des$MCML(y = ysim, method="mcem",tol=1e-2,conv.criterion = 1)

## End(Not run)

## ------------------------------------------------
## Method `Model$LA`
## ------------------------------------------------


#create example data with six clusters, five time periods, and five people per cluster-period
df <- nelder(~(cl(6)*t(5)) > ind(5))
# parallel trial design intervention indicator
df$int <- 0
df[df$cl > 3, 'int'] <- 1 
# specify parameter values in the call for the data simulation below
des <- Model$new(
  formula = ~ factor(t) + int - 1 + (1|gr(cl)*ar0(t)),
  covariance = c(0.05,0.7),
  mean = c(rep(0,5),-0.2),
  data = df,
  family = stats::binomial()
)
ysim <- des$sim_data() # simulate some data from the model
fit1 <- des$LA(y = ysim)

Description

Generates a data frame expressing a block experimental structure using Nelder's formula

Usage

nelder(formula)

Arguments

Details

Nelder (1965) suggested a simple notation that could express a large variety of different blocked designs. The function 'nelder()' that generates a data frame of a design using the notation. There are two operations:

'>' (or $\to$ in Nelder's notation) indicates "clustered in".

'*' (or $\times$ in Nelder's notation) indicates a crossing that generates all combinations of two factors.

The implementation of this notation includes a string indicating the name of the variable and a number for the number of levels, such as 'abc(12)'. So for example '~cl(4) > ind(5)' means in each of five levels of 'cl' there are five levels of 'ind', and the individuals are different between clusters. The formula '~cl(4) * t(3)' indicates that each of the four levels of 'cl' are observed for each of the three levels of 't'. Brackets are used to indicate the order of evaluation. Some specific examples:

'~person(5) * time(10)': A cohort study with five people, all observed in each of ten periods 'time'

'~(cl(4) * t(3)) > ind(5)': A repeated-measures cluster study with four clusters (labelled 'cl'), each observed in each time period 't' with cross-sectional sampling and five indviduals (labelled 'ind') in each cluster-period.

'~(cl(4) > ind(5)) * t(3)': A repeated-measures cluster cohort study with four clusters (labelled 'cl') wth five individuals per cluster, and each cluster-individual combination is observed in each time period 't'.

'~((x(100) * y(100)) > hh(4)) * t(2)': A spatio-temporal grid of 100x100 and two time points, with 4 households per spatial grid cell.

Value

A list with the first member being the data frame

Examples

nelder(~(j(4) * t(5)) > i(5))
nelder(~person(5) * time(10))

Description

Generate a data frame that nests one data frame in another

Usage

nest_df(df1, df2)

Arguments

Details

For two data frames 'df1' and 'df2', the function will return another data frame that nests 'df2' in 'df1'. So each row of 'df1' will be duplicated 'nrow(df2)' times and matched with 'df2'. The values of each 'df2' will be unique for each row of 'df1'

Value

data frame

Examples

nest_df(data.frame(t=1:4),data.frame(cl=1:3))

Description

Predictions cannot be generated directly from an 'mcml' object, rather new predictions should be generated using the original 'Model'. A message is printed to the user.

Usage

## S3 method for class 'mcml'
predict(object, ...)

Arguments

Value

Nothing. Called for effects.

Description

Generates predicted values from a 'Model' object using a new data set to specify covariance values and values for the variables that define the covariance function. The function will return a list with the linear predictor, conditional distribution of the new random effects term conditional on the current estimates of the random effects, and some simulated values of the random effects if requested. Typically this functionality is accessed using 'Model$predict()', which this function provides a wrapper for.

Usage

## S3 method for class 'Model'
predict(object, newdata, offset = rep(0, nrow(newdata)), m = 0, ...)

Arguments

Value

A list with the linear predictor, parameters (mean and covariance matrices) for the conditional distribution of the random effects, and any random effect samples.

Description

Print method for class "'mcml'"

Usage

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

Arguments

Details

'print.mcml' tries to replicate the output of other regression functions, such as 'lm' and 'lmer' reporting parameters, standard errors, and z- and p- statistics. The z- and p- statistics should be interpreted cautiously however, as generalised linear miobjected models can suffer from severe small sample biases where the effective sample size relates more to the higher levels of clustering than individual observations.

Parameters 'b' are the mean function beta parameters, parameters 'cov' are the covariance function parameters in the same order as '$covariance$parameters', and parameters 'd' are the estimated random effects.

Value

No return value, called for side effects.

Description

Prints a progress bar

Usage

progress_bar(i, n, len = 30)

Arguments

Value

A character string

Examples

progress_bar(10,100)

Description

Skeleton list to declare a quantile regression model in a 'Model' object.

Usage

Quantile(link = "identity", scaled = FALSE, q = 0.5)

Arguments

Value

A list with two elements naming the family and link function

Description

Extracts the random effect estimates or samples from an mcml object returned from call of 'MCML' or 'LA' in the Model class.

Usage

random.effects(object)

Arguments

Value

A matrix of dimension (number of fixed effects ) x (number of MCMC samples). For Laplace approximation, the number of "samples" equals one.

Description

Calling residuals on an 'mcml' object directly is not recommended. This function will currently only generate marginal residuals. It will generate a new 'Model' object internally, thus copying all the data, which is not ideal for larger models. The preferred method is to call residuals on either the 'Model' object or using 'Model$residuals()', both of which will also generate conditional residuals.

Usage

## S3 method for class 'mcml'
residuals(object, type, ...)

Arguments

Value

A matrix with either one column is conditional is false, or with number of columns corresponding to the number of MCMC samples.

Description

Return the residuals from a 'Model' object. This function is a wrapper for 'Model$residuals()'. Generates one of several types of residual for the model. If conditional = TRUE then the residuals include the random effects, otherwise only the fixed effects are included. For type, there are raw, pearson, and standardized residuals. For conditional residuals a matrix is returned with each column corresponding to a sample of the random effects.

Usage

## S3 method for class 'Model'
residuals(object, type, conditional, ...)

Arguments

Value

A matrix with either one column is conditional is false, or with number of columns corresponding to the number of MCMC samples.

Description

By default, the package will use multithreading for many calculations if OpenMP is available on the system. For multi-user systems this may not be desired, so parallel execution can be disabled with this function.

Usage

setParallel(parallel_, cores_ = 2L)

Arguments

Value

None, called for effects

Description

Summary method for class "'mcml'"

Usage

## S3 method for class 'mcml'
summary(object, ...)

Arguments

Details

'print.mcml' tries to replicate the output of other regression functions, such as 'lm' and 'lmer' reporting parameters, standard errors, and z- and p- statistics. The z- and p- statistics should be interpreted cautiously however, as generalised linear miobjected models can suffer from severe small sample biases where the effective sample size relates more to the higher levels of clustering than individual observations.

Parameters 'b' are the mean function beta parameters, parameters 'cov' are the covariance function parameters in the same order as '$covariance$parameters', and parameters 'd' are the estimated random effects.

Value

A list with random effect names and a data frame with random effect mean and credible intervals

Description

Summarizes 'Model' object.

Usage

## S3 method for class 'Model'
summary(object, max_n = 10, ...)

Arguments

Value

An object of class 'logLik'. If both 'fixed' and 'covariance' are FALSE then it returns NA.

Description

Returns the calculated variance-covariance matrix for a 'mcml' object. The generating Model object has several methods to calculate the variance-convaariance matrix. For the standard GLS information matrix see 'Model$information_matrix()'. Small sample corrections can be accessed directly from the generating Model using 'Model$small_sample_correction()'. The varaince-covariance matrix including the random effects can be accessed using 'Model$information_matrix(include.re = TRUE)'.

Usage

## S3 method for class 'mcml'
vcov(object, ...)

Arguments

Value

A variance-covariance matrix.

Description

Returns the variance-covariance matrix for a 'Model' object. Specifically, this function will return the inverse GLS information matrix for the fixed effect parameters. Small sample corrections can be accessed directly from the Model using 'Model$small_sample_correction()'. The varaince-covariance matrix including the random effects can be accessed using 'Model$information_matrix(include.re = TRUE)'.

Usage

## S3 method for class 'Model'
vcov(object, ...)

Arguments

Value

A variance-covariance matrix.

Description

Data for first example in Section 3.12 of JSS paper

Description

Data for second example in Section 3.12 of JSS paper

Description

Data for third example in Section 3.12 of JSS paper

Description

Data for model tests