DstarM (version 0.4.0)

estND: Estimate nondecision density

Description

Estimate nondecision density

Usage

estND(
  res,
  tt = NULL,
  data = NULL,
  h = res$h,
  zp = 5,
  upper.bound = 1,
  lower.bound = 0,
  Optim = list(),
  verbose = TRUE,
  dist = NULL,
  NDindex,
  max = 100,
  useRcpp = TRUE
)

Arguments

res

an object of class D*M.

tt

optional timegrid if the nondecision density is to be estimated at a different grid than the model density.

data

if tt is specified then the original dataset must be supplied too.

h

Optional smoothing parameter to be used when estimating the nondecision model on a different time grid than the decision model. If omitted, the smoothing parameter of the decision model is used.

zp

Zero padding the estimated nondecision density by this amount to avoid numerical artefacts.

upper.bound

An upper bound for the nondecision density. Defaults to one. Lowering this bound can increase estimation speed, at the cost of assuming that the density of the nondecision distribution is zero past this value.

lower.bound

A lower bound for the nondecision density. Defaults to zero. Increasing this bound can increase estimation speed, at the cost of assuming that the density of the nondecision distribution is zero past this value.

Optim

a named list with identical arguments to DEoptim.control. In addition, if verbose == TRUE Optim$steptol can be a vector, i.e. c(200, 50, 10) means: Do 200 iterations then check for convergence, do 50 iterations then check for convergence, check every 10 iterations for convergence until itermax is reached. If there are multiple nondecision distributions to estimate, one can supply different estimation parameters for every nondecision distribution by supplying Optim as a list of lists. Every sublists then corresponds to parameters for one nondecision distribution and should consist of arguments for DEoptim.control. Defaults to Optim = list(reltol = 1e-6, itermax = 1e4, steptol = 200, CR = .9, trace = 0).

verbose

Numeric, should intermediate output be printed? Defaults to 1, higher values result in more progress output. Estimation will speed up if set to 0. If nonzero, Optim$trace will be forced to 0, hereby disabling the build in printing of DEoptim. To enable the printing of DEoptim, set verbose to 0 and specify Optim$trace.

dist

A matrix where columns represent nondecision distributions. If this argument is supplied then the objective function will be evaluated in these values.

NDindex

A vector containing indices of which nondecision distributions to estimate. If omitted, all nondecision distributions that complement the results in res are estimated.

max

A positive float which indicates the maximum height of the nondecision distribution. If estimated nondecision distributions appear chopped of or have a lot of values at this max value it is recommended to re-estimate the nondecision distributions with a higher max value. Increasing the max value without reason will increase the size of the parameter space and slow the estimation procedure.

useRcpp

Logical, setting this to true will make use of an Rcpp implementation of the objective function. This gains speed at the cost of flexibility.

Details

When verbose is set to 1, the ETA is an estimated of the time it takes to execute ALL iterations. Convergence can (and is usually) reached before then.

Examples

# simulate data with three stimuli of different difficulty.
# this implies different drift rates across conditions.
# define a time grid. A more reasonable stepsize is .01; this is just for speed.
tt = seq(0, 5, .1)
pars = c(.8, 2, .5, .5, .5, # condition 1
        .8, 3, .5, .5, .5, # condition 2
        .8, 4, .5, .5, .5) # condition 3
pdfND = dbeta(tt, 10, 30)
# simulate data
dat = simData(n = 3e5, pars = pars, tt = tt, pdfND = pdfND)
# define restriction matrix
restr = matrix(1:5, 5, 3)
restr[2, 2:3] = 6:7 # allow drift rates to differ
# fix variance parameters
fixed = matrix(c('sz1', .5, 'sv1', .5), 2, 2)
if (FALSE) {
# Run D*M analysis
res = estDstarM(data = dat, tt = tt, restr = restr, fixed = fixed)
# Estimate nondecision density
resND = estND(res)
plot(resND)
lines(tt, pdfND, type = 'b', col = 2)
}