Ordered choice models now include an intercept by default.
generate_choice_parameters() now draws missing
coefficients from a standard normal distribution instead of a normal
distribution with variance 10.
Added support for uncorrelated normal and log-normal random effects.
Added latent_class_effects to
choice_formula(), which names the effects that differ
between latent classes, and the column latent_class to
choice_effects().
A named beta is matched to the effects by name
instead of by position, and generate_choice_parameters()
completes a named beta that fixes only some of the
effects.
Probit probabilities now come from
oeli::pmvnorm().
Specialized choiceprob_*() functions are now
internal; use compute_choice_probabilities() as the public
interface.
Improved input validations. Internal validation failures now use
descriptive cli messages.
Added partial rankings, individual choice sets, joint panel outcomes, log-normal random effects, and latent class estimation.
Probability calculations now use Rcpp.
Added the choice data sets
water_conservation_choice, wind_power_choice,
and lichess_berserk_choice.
long_to_wide() now accepts a factor
column of alternatives.
J in choice_alternatives() defaults to
the number of supplied choice alternatives.
New function train_test(), which splits choice data
by deciders or by choice occasions into a train and a test
subset.
Removed unused package dependencies.