[R] Error in MCMCpack in fitting a 2PL-IRT-model with independent priors on the item-parameters

[Tobias Konitzer]

I am trying to run a simple one-dimensional Bayesian IRT-model on a 9
(respondents)*43(items) dataset, but I want to specify separate normal
priors for my item-parameters.

With sample data, I run

data(SupremeCourt)
priors<-matrix(NA,ncol=43,nrow=2)
priors[1,]<-0
priors[2,]<-seq(1,43,1)
posterior3 <- MCMCirtKd(t(SupremeCourt),dimensions=1,
                    theta.constraints=list(Scalia="+", Ginsburg="-"),
                    b0=priors,
                    B0=1000,
                    burnin=5, mcmc=2000, thin=5, verbose=500,
                    store.item=TRUE,store.ability=F)

I assume from the documentary that b0 "Can be either a scalar or a matrix
of dimension (K + 1) × items". But if I run this, no matter how I specify
it, I get "l0 neither matrix, list, nor scalar".

It seems like a package-bug to me, but I have no clue how to specify
separate priors for item-parameters.

Thanks, Tobi

	[[alternative HTML version deleted]]