trajeR: Fitting longitudinal mixture models
In trajeR: Group Based Modeling Trajectory

trajeR

R Documentation

Fitting longitudinal mixture models

Description

Fitting longitudinal mixture models

Usage

trajeR(
  Y,
  A = NULL,
  Risk = NULL,
  TCOV = NULL,
  data = NULL,
  ng = NULL,
  degre = NULL,
  degre.nu = 0,
  degre.phi = 0,
  Model,
  Method = "L",
  ssigma = FALSE,
  ymax = NULL,
  ymin = NULL,
  hessian = TRUE,
  itermax = 100,
  paraminit = NULL,
  ProbIRLS = TRUE,
  refgr = 1,
  fct = NULL,
  diffct = NULL,
  nbvar = NULL,
  ng.nl = NULL,
  nls.lmiter = 50,
  control = list()
)

Arguments

`Y`	Matrix, Formula, or a list of Formulas. A matrix containing the response variables, a single formula (e.g., `Y ~ A + I(A^2)`), or a list of formulas for models with different degrees.
`A`	Matrix. A matrix containing the time variable data. Can be omitted if specified in the RHS of the `Y` formula.
`Risk`	Matrix or Formula. An optional matrix (or one-sided formula like `~ X1 + X2`) of covariates that modify the probability of belonging to a group.
`TCOV`	Matrix or Formula. An optional matrix (or one-sided formula) containing time-dependent covariates that influence the trajectories.
`data`	data.frame. An optional data frame containing the variables named in the formulas.
`ng`	Integer. The number of groups. Required if using a single formula or if 'degre' is not provided.
`degre`	Vector of integers. The degree of each polynomial function. The code adds 1 to these values internally.
`degre.nu`	Vector of integers. The degree of the polynomial for the zero-inflation part of a ZIP model.
`degre.phi`	Vector of integers. The degree of the polynomial for the precision parameter (phi) of a BETA model. The code adds 1 to these values internally.
`Model`	String. The model to be used. One of "LOGIT", "CNORM", "ZIP", "BETA", "POIS", or a non-linear model.
`Method`	String. The estimation method. "L" for Likelihood, "EM" for Expectation-Maximization.
`ssigma`	Logical. For the CNORM model, if TRUE, a single shared standard deviation (sigma) is estimated for all groups. Default is FALSE.
`ymax`	Real. For the CNORM model, the maximum value for censoring. Defaults to max(Y) + 1.
`ymin`	Real. For the CNORM model, the minimum value for censoring. Defaults to min(Y) - 1.
`hessian`	Logical. If TRUE, the Hessian matrix is computed. Default is TRUE.
`itermax`	Integer. The maximum number of iterations for the optimization algorithm (like 'optim' or the EM loop). Default is 100.
`paraminit`	Vector. The vector of initial parameters.
`ProbIRLS`	Logical. Indicates the method to use for searching predictor probabilities. Default is TRUE.
`refgr`	Integer. The reference group number. Default is 1.
`fct`	Function. The definition of the function f in the definition in nonlinear model.
`diffct`	Function. The differential of the function f in the nonlinear model.
`nbvar`	Integer. The number of variable in the nonlinear model.
`ng.nl`	Integer. The number of group for a non linear model.
`nls.lmiter`	Integer. In the case of non linear model, the maximum number of iterations allowed.
`control`	A list of control parameters for the 'ucminf' optimization algorithm. 'trace': If positive, detailed tracing information is printed for each iteration. 'grtol': The algorithm stops when the largest absolute value of the gradient is less than 'grtol'. Default is 1e-6. 'xtol': The algorithm stops when the last relative step length is sufficiently small. Default is 1e-12. 'stepmax': Initial maximal allowed step length (radius of trust-region). The value is updated during the optimization. Default is 1. 'maxeval': The maximum number of function evaluations (objective and gradient). Default is 500. ‘grad': Method for finite difference approximation of the gradient if not provided. Can be ’forward' (default) or 'central'. 'gradstep': Vector of length 2 for the step length in finite difference approximation. Default is c(1e-6, 1e-8).

Value

Returns an object of a class corresponding to the chosen model (e.g., 'Trajectory.CNORM', 'Trajectory.LOGIT').

Returns an object of a class corresponding to the chosen model (e.g., 'Trajectory.CNORM', 'Trajectory.LOGIT'). This object, referred to as a 'Trajectory' object, contains the following elements:

'beta': Vector of the final beta parameters for the trajectory shapes.
'sigma': Vector of the final sigma parameters (standard deviations), for CNORM and NL models.
'nu': Vector of final nu parameters for the zero-inflation part of a ZIP model.
'phi': Vector of final phi parameters (precision) for a BETA model.
'delta': Vector of the final delta parameters for time-dependent covariates (if any).
'theta': Vector of the final theta parameters for group membership probabilities.
'sd': Vector of the standard deviations of the estimated parameters.
'tab': A data frame with parameter estimates, standard errors, T-values, and p-values.
'Model': A string indicating the fitted model (e.g., "CNORM", "LOGIT").
'groups': The number of groups.
'Names': The names of the parameters.
'Method': The estimation method used ("L" for Likelihood or "EM" for Expectation-Maximization).
'Size': The number of individuals (observations).
'Likelihood': The final log-likelihood value.
'Time': The time points for the first individual, or min/max time for ZIP model.
‘degre': A vector with the polynomial degrees for each group’s trajectory.
'degre.nu': For ZIP models, a vector with the polynomial degrees for the zero-inflation part.
'degre.phi': For BETA models, a vector with the polynomial degrees for the precision part.
'min', 'max': For CNORM models, the censoring limits used.
'fct': For NL models, the non-linear function provided.
'varcov': The variance-covariance matrix of the parameters.
'convergence': An integer code indicating the convergence status of the optimization algorithm. The meaning depends on the method used.
- For 'Method = "L"' (using 'ucminf'):
- '1': Stopped by small gradient ('grtol'). The current iterate is probably a solution.
- '2': Stopped by small step ('xtol'). Successive iterates are within tolerance.
- '3': Stopped by function evaluation limit ('maxeval').
- '4': Stopped by zero step from line search. May be a solution.
- '-2': Computation did not start: length of parameters is 0.
- '-4': Computation did not start: 'stepmax' is too small.
- '-5': Computation did not start: 'grtol' or 'xtol' is not positive.
- '-6': Computation did not start: 'maxeval' is not positive.
- '-7': Computation did not start: The provided Hessian was not positive definite.
- For 'Method = "EM"':
- '1': Indicates that the EM algorithm has completed its run (convergence is assessed by the user by checking the stability of parameters across iterations).