plot() has the new option type = "mixture", which draws the estimated
mixing distributions of the random effects.
bayes_factor() now accepts models with a fixed number of up to six latent
classes.
Ordered choice models now include an intercept by default.
The occupancy table of latent_class_diagnostics() lists every number of
classes from one to the maximum, including numbers with probability zero.
Every chain starts at a draw from the prior distributions of the coefficients and thresholds, with the deciders allocated to the classes at random, instead of at the prior means, so that the chains start dispersed.
Cap the progress updates to prevent that many fast updates freeze the RStudio console.
Replaced the former multi-step workflow with fit(), which accepts an
empirical data frame or simulates data when data = NULL.
Added correlated and uncorrelated normal, positive log-normal, and negative
log-normal random effects through named random_effects specifications.
Added latent_class_effects to fit(), which names the effects that
differ between the latent classes.
Added interpret(), which reads the estimates as compensations between
effects, such as the willingness to pay, and as average marginal
effects or marginal effects at the average covariate values, each with
posterior uncertainty.
Added the update() method for fitted models, which refits with a modified
specification.
Added decider-level WAIC and PSIS-LOO and replaced the harmonic-mean marginal-likelihood estimator with bridge-sampling Bayes factors.
Reworked predict() and residuals() around posterior uncertainty.
Added posterior diagnostics and standard bayesplot displays through
summary() and plot().
Improved numerical efficiency of the Gibbs sampler.
Rewrote the vignettes: getting started, model variants, preference heterogeneity, posterior prediction, and model evaluation.
Fix in npar.RprobitB_fit(). Parameters for the covariance matrices of the mixing distributions have been miscounted.
Fix in sufficient_statistics(). There was a bug in calculating sufficient statistics in the ordered case.
Added function mode_approx() to approximate conditional posterior modes.
Documentation improvements.
Small fixes.
Added train_choice data set.
logLik() method now returns a logLik object that can be passed to AIC() and BIC() methods from {stats}.
Small bug fixes for the ordered probit model case.
Ranked probit models
Ordered probit models
Print method for RprobitB_parameter.
Splitting the data set into a train and test part can now be done by the function train_test(). Consequently, the argument test_prob was removed in prepare_data() and simulate_choices().
The function simulate_choices() does not have the argument distr anymore. Instead, covariates can be supplied via the covariates argument. Consequently, the argument standardize was removed as well.
Simplified specifying the utility scale, see the help page of RprobitB_normalization().
The function compare is now called model_selection().
The function prepare is now called prepare_data().
The function simulate is now called simulate_choices() to not mask stats::simulate().
The function mcmc is now called fit_model().
The function choice_probs is now called choice_probabilities().
Integrated S3 classes and methods.
Several new functionalities.