Changelog • hbamr

hbamr 2.3.0

All FBAM-type models have been revised to allow for heteroskedastic errors.
- The FBAM_MINI model has been renamed FBAM to reflect this change.

CRAN release: 2024-02-23

The prior on rho (which appears in all heteroskedastic models) has been made narrower to reduce the risk of divergent transitions when there are few observations per stimuli (which could be the case for e.g. expert surveys).
The default priors on mu_beta and mu_alpha in MULTI-type models and the priors on sigma_beta and sigma_alpha in HBAM-type models have also been made narrower to reduce the risk of sampling issues.

CRAN release: 2024-02-14

The get_est() function now takes the logical argument format_orig, which if TRUE makes the function return posterior summaries for individual-level parameters in a format that matches the rows in the original dataset.
The hbam() and fbam() functions now store the input data within the returned objects, which allows simplifying the interfaces for other functions like get_est() and plot_over_self(). As a result, the plot_over_self() function no longer requires a data argument.
The prep_data(), hbam(), and fbam() functions now allow users to not supply self-placements. In this case, no meaningful respondent positions will be estimated, but all other parameters are unaffected.
The prep_data() function now allows the group_id argument to take various forms, such as factor or character. It also allows missing values in the group_id vector and will drop respondents who do not have a valid group_id. Missing values would previously generate an uninformative error message.
The models now accept even numbers of answering categories, but prep_data() will throw a warning.
The prep_data() function now identifies the left and right poles for the BAM model as the stimuli with the most non-NA observations on each side of the center. This can be advantageous when analyzing datasets where some stimuli have a much higher number of valid observations than others.

CRAN release: 2024-02-07

The behavior of FBAM models now depends on whether they are fit via MCMC or optimization. When fit via MCMC, the models simulate errors in respondents’ self-placements and generate draws for the latent discrete flipping parameters, kappa, using a Bernoulli function – just like other models. When fit via optimization, there is no simulation of self-placement errors and lambda is rounded to yield a MAP estimate of kappa.

CRAN release: 2024-01-25

The hbam_cv() function no longer uses parallel::mclapply() for parallel computation as the latter relies on forking, which is not available on Windows. hbam_cv() has been revised to work with the future package, where the user decides the computational strategy and options are available for parallel computation on all systems.
The return value of hbam_cv() has been changed to comply with the standards of the loo package. The function now returns a list with classes kfold and loo. This allows the user to compare estimated ELPDs and obtain standard errors for their differences via loo::loo_compare().

CRAN release: 2024-01-17

All models in the package now simulate errors in respondents’ self-placements. In the original HBAM model, respondents’ latent positions were treated as parameters, which had several consequences: It yielded a realistic level of uncertainty in the estimated positions, but also led to slower sampling. Furthermore, the hierarchical prior on the latent respondent positions shrunk them toward zero, which was undesirable in most applications as it affected the distances between respondents and stimuli. Finally, the original approach led to a more nuanced posterior distribution where combinations of individual-level parameter values that would lead to implausible values for the respondent positions were weighted down. However, the model’s log-normal prior on the stretch parameters is by itself sufficient to yield meaningful respondent positions estimates, so there is no significant cost to not treating the latent respondent positions as parameters. The current versions of the model instead simulate errors in the self-placements to yield the same level of uncertainty as the original model, while sampling considerably faster.

HBAM_MAX has been removed as it offered little extra after the other models were revised to simulate errors in respondents’ self-placements.

CRAN release: 2024-01-08

HBAM_MULTI is a version that models differences between groups defined by the user. It requires a vector identifying the groups to be supplied as the argument group_id. The model gives each group separate hyperparameters for the locations of the prior distributions for the shift and stretch parameters. Rather than shrinking the estimates toward the mode for the whole dataset, this model shrinks the estimates toward the mode for the group. The vectors of hyperparameters are called mu_alpha and mu_beta and are constructed to have means of 0. The scales of the priors on these hyperparameters can be set by the user via the arguments sigma_mu_alpha and sigma_mu_beta. One potential use for this model is to supply self-placements as group_id, and thus give each self-placement group its own prior distribution for the shift and stretch parameters.
HBAM_MULTI_NF is a version of the HBAM_MULTI model that does not allow for scale flipping.
FBAM_MULTI is a version of the FBAM_MINI model that shares the group-modeling features of the HBAM_MULTI model. It allows the user to set the scales of the priors for the shift and stretch parameters via the arguments sigma_alpha and sigma_beta, and set the scales of the priors on mu_alpha and mu_beta via the arguments sigma_mu_alpha and sigma_mu_beta.
FBAM_MULTI_NF is a version of the FBAM_MULTI model that does not allow for scale flipping.

The original HBAM model has been renamed HBAM_MAX.
The HBAM_NE model has been renamed HBAM as it makes a better default model.
HBAM_0 has been renamed HBAM_NF. To be consistent with the other models and offer faster sampling, the model no longer models the latent respondent positions as parameters.

FBAM_MINI (as well as the new FBAM models) now allows for user-defined prior distributions.
All models (except HBAM_MAX) have been revised to not treat the respondent positions as parameters. This results in considerably faster sampling.
All HBAM models that allow for scale flipping have been given a logit-normal prior on the mixing proportions, lambda (i.e. the expectations of the latent discrete flipping parameters, kappa). This replaces the original beta prior, which could trigger divergent transitions.

HBAM_2 has been replaced by the more general HBAM_MULTI model.
HBAM_HM has been removed as it offered little extra and the number of models should be kept somewhat limited.
HBAM_R has been removed as users would typically be better off with the simpler HBAM_R_MINI.