spm {stpm}

A central function that estimates Stochastic Process Model parameters a from given dataset.

Description

A central function that estimates Stochastic Process Model parameters a from given dataset.

Usage

spm(
  x,
  model = "discrete",
  formulas = list(at = "a", f1t = "f1", Qt = "Q", ft = "f", bt = "b", mu0t = "mu0"),
  start = NULL,
  tol = NULL,
  stopifbound = FALSE,
  lb = NULL,
  ub = NULL,
  pinv.tol = 0.01,
  theta.range = seq(0.01, 0.2, by = 0.001),
  verbose = FALSE,
  gomp = FALSE,
  opts = list(algorithm = "NLOPT_LN_NELDERMEAD", maxeval = 100, ftol_rel = 1e-08)
)

Arguments

x

A dataset: is the output from prepare_data(...) function and consists of two separate data tables: (1) a data table for continuous-time model and (2) a data table for discrete-time model.

model

A model type. Choices are: "discrete", "continuous" or "time-dependent".

formulas

A list of parameter formulas used in the "time-dependent" model. Default: formulas=list(at="a", f1t="f1", Qt="Q", ft="f", bt="b", mu0t="mu0").

start

A starting values of coefficients in the "time-dependent" model.

tol

A tolerance threshold for matrix inversion (NULL by default).

stopifbound

A flag (default=FALSE) if it is set then the optimization stops when any of the parametrs achives lower or upper boundary.

lb

Lower boundary, default NULL.

ub

Upper boundary, default NULL.

pinv.tol

A tolerance threshold for matrix pseudo-inverse. Default: 0.01.

theta.range

A user-defined range of the parameter theta used in discrete-time optimization and estimating of starting point for continuous-time optimization.

verbose

A verbosing output indicator (FALSE by default).

gomp

A flag (FALSE by default). When it is set, then time-dependent exponential form of mu0 and Q are used: mu0 = mu0*exp(theta*t), Q = Q*exp(theta*t).

opts

A list of options for nloptr. Default value: opt=list(algorithm="NLOPT_LN_NELDERMEAD", maxeval=100, ftol_rel=1e-8). Please see nloptr documentation for more information.

Value

For "discrete" (dmodel) and "continuous" (cmodel) model types: (1) a list of model parameter estimates for the discrete model type described in "Life tables with covariates: Dynamic Model for Nonlinear Analysis of Longitudinal Data", Akushevich et al, 2005.<DOI:10.1080/08898480590932296>, and (2) a list of model parameter estimates for the continuous model type described in "Stochastic model for analysis of longitudinal data on aging and mortality", Yashin et al, 2007, Math Biosci.<DOI:10.1016/j.mbs.2006.11.006>.

For the "time-dependent" model (model parameters depend on time): a set of model parameter estimates.

References

Yashin, A. et al (2007), Stochastic model for analysis of longitudinal data on aging and mortality. Mathematical Biosciences, 208(2), 538-551.

Akushevich I., Kulminski A. and Manton K. (2005). Life tables with covariates: Dynamic model for Nonlinear Analysis of Longitudinal Data. Mathematical Popu-lation Studies, 12(2), pp.: 51-80. <DOI: 10.1080/08898480590932296>.

Yashin, A. et al (2007), Health decline, aging and mortality: how are they related? Biogerontology, 8(3), 291-302.<DOI:10.1007/s10522-006-9073-3>.

Examples

## Not run:  
library(stpm)
data.continuous <- simdata_cont(N=1000)
data.discrete <- simdata_discr(N=1000)
data <- list(data.continuous, data.discrete)
p.discr.model <- spm(data)
p.discr.model
p.cont.model <- spm(data, model="continuous")
p.cont.model
p.td.model <- spm(data, 
model="time-dependent",f=list(at="aa*t+bb", f1t="f1", Qt="Q", ft="f", bt="b", mu0t="mu0"), 
start=list(a=-0.001, bb=0.05, f1=80, Q=2e-8, f=80, b=5, mu0=1e-3))
p.td.model

## End(Not run)