cr.des.pseudo: Allocate treatments according to a (weighted) L-optimal...

In mst1g15/biasedcoin:

Description

Allocate treatments according to a (weighted) L-optimal criterion allowing for a pseudo-nonmyopic approach. We assume a linear model for the response.

Usage

cr.des.pseudo(covar, true.beta, true.sigma, threshold, kappa, init, cr.lossfunc,
  sim, z.probs, k, wt = T, prior.scale = 100, same.start = NULL,
  rand.start = NULL, stoc = T, bayes = T, prior.default = T,
  coordex = NULL, u = NULL, ...)

Arguments

covar	a dataframe for the covariates
true.beta	the true parameter values of the regression coefficients
true.sigma	the true parameter value for the standard deviation
threshold	the cut-off value for hypothesis tests
kappa	the value of probability at which weights are set at zero
init	the number of units in the initial design
cr.lossfunc	loss function appropriate for linear combinations of parameters
sim	number of trajectories to simulate
z.probs	vector of probabilities for each level of covariate z
k	the number of "outer loops" in the coordinate exchange algorithm
wt	set to T if the above lossfunction is weighted, NULL otherwise
prior.scale	the prior scale parameter
same.start	the design matrix to be used for the initial design. If set to NULL, function generates initial design.
rand.start	If set to T, function generates an initial design randomly. Else, coordinate exchange is used.
stoc	set to T if treatments are allocated using a stochastic method where the probability is determined by the optimality crtierion. Set to F if treatments are allocated deterministically.
bayes	set to T if bayesglm is used instead of glm. Default prior assumed.
prior.default	set to T if default priors for bayesglm is used. If set to False and bayes=T, normal priors used.
coordex	set to T if the coordinate exchange algorithm is used to assign treatments in the trajectory. Sequential approach used otherwsise.
u	vector of uniform random numbers for generating responses. If set to NULL, responses generated from the binomial distribution.
...	further arguments to be passed to <lossfunc>

Value

design matrix D, responses y, all estimates of betas, final estimate of beta, all weights, all estimates of standard deviation, beta, probabilities for treatment assignment, all values of optimalities (weighted L, DA, weighted DA), proportion of treatment=1, proportion of covariate in each group Type 1 error, true value of power, empirical value of power.

mst1g15/biasedcoin documentation built on Nov. 26, 2019, 4:01 a.m.