v0.99.1

This is the second beta release; most features are now implemented and stable. New version of the working paper is also now available describing all the new features in this release: https://osf.io/preprints/osf/8j2bt

v0.99 (v1.0 beta release)

This release is a beta for v1.0. The package's new features & functionality are documented in the two vignettes, Package_Introduction and Time_Series (see README file in the repository for more information). The only feature that is not implemented as of yet are ideal point marginal effects.

Because the new version relies on cmdstanr, which is not on CRAN, idealstan will remain only on Github until cmdstanr can be put on CRAN as well.

Some of the new features include:

1. Mixed outcomes -- both discrete and continuous distributions can be used in the same model for different items (continuous, ordinal, binary). You need to pass a column model_id to id_make to make this work, as well as specify discrete outcome/response as outcome_disc and any continuous outcomes/responses as outcome_cont.
2. Within-chain parallelization -- you can now specify the number of cores ncores as a multiple of nchains to use multiple cores per chain and speed up processing.
3. A variety of new defaults/priors/processes to improve & speed up dynamic ideal point estimation.

Idealstan v0.7.2

This release adds two new functions for plotting covariates, id_plot_cov for covariate marginal effects, and id_plot_irf for impulse-response functions for AR(1) models. It also updates idealstan to work with the latest version of rstan.

Gaussian process for ideal point models

This release implements Gaussian process priors for ideal point models, which permits semi-parametric inference of the time-varying nature of ideal points. Compared to existing approaches, Gaussian processes allow for more flexible fits to time-series while also allowing for over-time covariates that measure the effect of independent variables on the change in ideal points over time.

Bug Fix Release

This release fixes plotting issues introduced by v.0.5.0. It also resolves some variance restriction issues for the AR(1) model.

Idealstan v0.5.0 Released

This version of idealstan employs several new outcomes/response distributions, including ordinal-graded response, Poisson (count) variables, Normal (continuous), Log-Normal (positive-continuous), and the latent space formulation of ideal point models. In addition, new time-varying processes are included that allow you to fit for any provided model random-walk and AR(1) (stationary) ideal points.

This version also includes the ability to fit hierarchical covariates for ideal points and also for discrimination parameters in either time-varying or static models.

Data is now handled in long data frames rather than matrices, which facilitates all of the new options.

CRAN Release

This release matches the version currently on CRAN. Additional features will be added through this branch (master) on Github, and eventually these will make their way to CRAN every few months.

CRAN Release

This release has been submitted to CRAN for review.

Minor Release (V0.2.1)

This release changes the defaults for automatic identification to use person rather than discrimination parameters. The id_estimate function should work better out of the box without having to change the defaults. Also, vignettes updated with more examples.

First Stable Release

This release incorporates most of the basic functions of the idealstan package for Bayesian inference of ordinal and binary IRT ideal point models.