StrategySwitch: Strategy Switch model

Description Usage Arguments Details Value Author(s) References Examples

Description

Estimates the Strategy Switch model (Speekenbrink et al., 2009).

Usage

1
2
3
StrategySwitch(y,X,Z,prior,A,b,tol=1e-4,maxiter=200,
A.est=TRUE,prior.est,b.est,A.group=rep(1,length(y)),
verbose=FALSE,b.min=-Inf)

Arguments

y

A list with each element being a vector with binary responses of a subject.

X

A list with each element a matrix of binary indicators for presence X[i,j] = 1 or absence X[i,j] = 0 of cue j at trial i of a subject.

Z

A matrix with coefficients for

prior

A vector with (starting values) for the prior probabilties of the strategies.

A

A list with starting values for the transition matrices.

b

A vector with (starting) values for the response consistency parameters of each subject.

tol

tolerance for the EM algorithm; estimation stops when the change in logLikelihood is smaller than tol.

maxiter

maximum number of iterations of the EM algorithm.

A.est

A list with each element a matrix with integers to fix (value 0) that parameter in A or freely estimate (value > 1) it. Giving different elements the same integer > 0 will contrain these elements to have the same value.

prior.est

A vector with integers to fix or estimate the prior strategy probabilities. Numbering is similar to that for A.est.

b.est

A vector with integers to fix or estimate the response consistency parameter. Numbering is similar to that for A.est

A.group

a vector indicating for each subject which transition matrix (element in list A) applies.

verbose

(logical) to indicate whether details of the EM iterations should be printed to screen.

b.min

lower bound for the consistency parameter.

Details

This function estimates the Strategy Switch model (Speekenbrink et al., in press). The Strategy Switch model is formulated as a hidden Markov model in which the states are specific strategies of responding to cue patterns. Design matrices for the cue patterns are given in the list X. Each strategy specifies a predicted probability of responses for each cue pattern. These predictions are on a logit scale. The strategies are given as rows in the matrix Z, which contain coefficients such that X[i]][j,]

Value

An list with estimated A, b, prior, etc.

Author(s)

Maarten Speekenbrink

References

Speekenbrink, M., Lagnado, D. A, Wilkinson, L., Jahanshahi, M. & Shanks, D. R. (in press). Models of probabilistic category learning in Parkinson's disease: Strategy use and the effects of L-dopa. Journal of Mathematical Psychology.

Examples

 1
 2
 3
 4
 5
 6
 7
 8
 9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
## open weather prediction data
data(WPT)
## specify the Z matrix for the Constant Error version
Z <- rbind(rep(0,15),
            c(0,1,1,-1,-1,-2,0,0,0,0,2,1,1,-1,-1),
            c(-1,2,rep(0,13)),
            c(-1,0,2,rep(0,12)),
            c(1,0,0,-2,rep(0,11)),
            c(1,0,0,0,-2,rep(0,10)),
            c(0,1,1,-1,-1,-1,1,0,0,-1,1,0,2,-2,0))
## construct y and X lists
y <- X <- list()
for(i in 1:length(levels(WPT$id))) {
    dat <- subset(WPT,id==levels(WPT$id)[i])
    y[[i]] <- as.numeric(dat$r==0)
    X[[i]] <- model.matrix(r~x1*x2*x3*x4 - x1:x2:x3:x4,data=dat)
}
## set initial values for A and prior
A <- matrix(1/7,ncol=7,nrow=7)
prior <- c(1,rep(0,6))
## 
## Not run: mod <- StrategySwitch(y=y,X=X,Z=Z,prior=prior,A=A,b=2.8,prior.est=rep(0,7),A.est=TRUE,b.est=TRUE,tol=1e-5,maxiter=2000,A.group=c(1,2),verbose=T,b.min=0.4054651)
## compute BIC
-2*mod$LL + 2*mod$df
## End(Not run)

StrategySwitch documentation built on May 2, 2019, 5 p.m.