NEWS

whatifbandit 1.0.2 (2026-07-23)

Fixing an error in contrast estimation for "control" when control_augment = 0 in mab_from_rct()
Changed estimator "AIPW" to "AW-AIPW" to emphasize adaptive weighting.=

multiple_mab_simulation() and single_mab_simulation() are removed in favor of mab_from_rct() which performs the same functions. Multiple simulations specified via the r argument.
mab_from_rct() now accepts bare column names and no longer accepts strings
multiple.mab and mab classes have been removed in favor of multi_rct_mab, multi_param_mab, single_rct_mab, single_param_mab, and .mab.
plot(), summary(), and print() methods have been removed pending new versions.

Joint hypothesis testing is now provided by joint_test() a function accepting single_*_mab classes. Methods are "bootstrap" and "randomization" inspired by Offer-Westort et. al (2021), see docs for more.
simulate_mab() is a new function that simulates a Multi-Arm-Bandit trial from provided population parameters. Uses similar options to mab_from_rct(). See the simulate_mab() documentation for details.
Inverse Probability Weighted estimates (IPW) are now provided, and cluster-robust standard errors supported
Linear contrast estimation supported. simulate_mab() and mab_from_rct() can now produce linear contrasts (e.g. treatment effects) in final output along with traditional mean estimates for all estimators. See the documentation for which options are available.
New discount rate parameter: Information from previous periods is now weighted by a discount rate, which can be useful for non-stationary bandits. See simulate_mab() and mab_from_rct() documentation for more info.

random_assign_prop now adjusts the assignment probabilities instead of splitting the data; see docs for more details.
Updating UCB1 formula and assignment mechanism to match the canonical paper Auer et al. (2002)
Major internal optimizations for runtime. Runtime decreased for packaged tanf dataset with individual assignment from 20-30 seconds on v.0.3.0 to 9 seconds.
Vignette has been removed pending new version.

Aligning output structure for data.table versus tibble data.frame classes. Ensuring results are data.frame library agnostic. Fixing miscellaneous bugs for data.table inputs.

plot.mab() for type = "assign" now displays the proportion of total observations assigned to each treatment for each period, instead of the individual probability of assignment.

multiple_mab_simulation() calculates the number of observations assigned to each treatment for each trial, and provides support for plotting them has been added to plot.multiple.mab() via the type = "hist" and quantity = "assignment" arguments.
summary.mab() now includes a new column with the number of observations assigned to each treatment.
summary.multiple.mab() now includes two new columns with the mean and standard deviation for the number of observations assigned to each treatment across the simulations.
Month-based assignment, time_unit = "month" can now be specified with and without an appropriate month_col, resulting in either time-based (no month_col) or calendar-based (provided month_col) assignments. See the time_unit documentation for more details.
plot.multiple.mab() now accepts arguments for ggplot2::facet_grid() for more precise customizations.

Fixed handling of numeric and factor types in condition_col of the data_cols argument.
Weighting AIPW by group size along with adaptive weights.
Fixed inconsistent results across with data.frames, tibbles, and data.tables. Running single_mab_simulation() or multiple_mab_simulation(), with the same seeds on the same system, now results in the same outcome regardless of input data class.

multiple_mab_simulation() supports parallel processing via future.
single_mab_simulation() and multiple_mab_simulation() support data.table for larger data sets.
summary(), print(), and plot() generics for mab and multiple.mab class objects.
single_mab_simulation() and multiple_mab_simulation() throw informative error messages, relating to argument specification, and data types passed.

single_mab_simulation() and multiple_mab_simulation() simulate successfully.