Package: birtRcpp 0.3.0

birtRcpp: BIvariate Response Time Modeling with 'Rcpp'

Item response and response time models fitted from their full conditionals. A built-in model is chosen with ecm(data, model = ...) or gibbs(data, model = ...): the two-parameter logistic model, alone or with log-normal response times (van der Linden, 2007, <doi:10.1007/s11336-006-1478-z>), latent regressions of ability and speed, cross-relations between ability and the log response times, and their quantile (asymmetric Laplace) versions. The main estimator is ECM (ecm(); Meng and Rubin, 1993, <doi:10.1093/biomet/80.2.267>): one conditional maximization per block of parameters, Polya-Gamma bounds for logistic likelihoods (Polson, Scott and Windle, 2013, <doi:10.1080/01621459.2013.829001>), adaptive or Bock-Aitkin quadrature (Bock and Aitkin, 1981, <doi:10.1007/BF02293801>) with normal-linear latent variables integrated exactly, and Louis' standard errors; quantile models maximize the asymmetric Laplace working likelihood, with sandwich standard errors. The same steps drive Gibbs sampling (gibbs(); conjugate updates and Metropolis-adjusted Langevin steps), with samplers in C++ ('RcppArmadillo') that port the 'Julia' package 'ExtendedRtIrtModeling.jl' (its function names are kept as aliases). Both estimators have a check: check_ecm(), and check_sampler() with the Geweke (2004, <doi:10.1198/016214504000001132>) test and simulation-based calibration. equations() and algorithm() write out the model and every step, before or after fitting.

Authors:Jie-Wen Tsai [aut, cre]