Changes in version 0.4.0 (2026-07-29) - Fixed the adjusted trimmed means estimator for higher value trimming (side = "HIGH"), which produced incorrect estimates for outcomes not centred near zero - The permutation p-value is now two-sided, so negative treatment effects are detected, and uses the (b+1)/(n_perm+1) estimator so it is never exactly zero - Fixed an off-by-one in higher value trimming which removed one more observation per group than the specified trimming fraction - Fixed print.tm_bias() erroring when the dropout spread was left unspecified - tm() and tm_bias() now work with any treatment variable name, not just TR - side now defaults to "LOW" and invalid values give a clear error - tm_bias() gains the documented trF default - tm_bias() no longer errors when a treatment group has no dropout - summary.tm() output now shows dropout and post-trimming sample sizes for both treatment groups - tm() no longer perturbs the random number generator state beyond the draws needed for the permutation test - The trimming fraction and treatment variable are now validated up front with clear error messages - Treatment groups are defined by factor level order, fixing numeric group codes such as 2 and 10 and respecting user-set reference levels - The permutation test is around 9 times faster - tm_bias objects gain a groups component giving the treatment and comparator group values - Corrected the author name spelling and publication year in the package citation Changes in version 0.3.1 (2024-08-29) - Minor fixes to improve the code, the documentation, and the package website Changes in version 0.3.0 (2023-05-10) - Simplified the NAMESPACE so functions from other packages are not imported - Modified the tm() helpfile example to run faster Changes in version 0.2.0 (2023-03-30) - Bump roxygen2 version - Update maintainer email address Changes in version 0.1.0 (2022-02-03) - Initial submission to CRAN