statistics-resampling-v5.7.0

This version includes a new function (bootridge.m), major functional additions to existing functions (bootbayes.m, bootwild.m, bootstrp.m and bootlm.m) and various changes to fix bugs, improve backwards compatibility, and update or improve documentation. Details of changes and additions are provided below.

NEW FUNCTION bootridge.m: Empirical Bayes penalized regression for univariate or multivariate outcomes, with shrinkage tuned to minimize prediction error computed by .632 bootstrap.

• bootbayes.m: extended functionality so that the function can be used to fit multivariate data (muliple outcomes simultaneously), removed unnecessary warning when using the Haldane Prior and fixed the random seed to make results for Haldane prior deterministic, improved the stability of calculations for weighted-least squares, generalised the 'auto' prior to work for regression and updated documentation about the prior.
• bootwild.m: Added capability to perform step-down maxT adjustment of p-values to control family-wise error rate (FWER).
• bootstrp.m: Added option to perform stratified bootstrap resampling, made changes to improve backwards compatibility.
• bootlm.m: Improved calculation of interaction terms to work with edge cases, major improvements to the calculation of bootstrap p-values for Type I ANOVA F-tests for more direct null hypothesis testing of nested models, minor bug fix for handling continuous variables, replaced Holm-Bonferroni adjustment with superior step-down maxT adjustment, added output argument to return out-of-bag bootstrap sample indices, added ID to MAT output, made changes to improve backwards compatibility.
• make.m: Changed mex compilation options for greater optimization of execution at runtime
• boot.cpp: enhancements to random number generation, changes to data types to allow more (or larger) bootstrap resamples, and minor changes to documentation.
• smoothmedian.cpp: Better handling of NaN, improved backwards compatibility fix for C++98 and minor changes to documentation
• credint.m: removed unnecessary warning.
• sampszcalc.m: Better handling of edge cases for testtype 't2'
• Added new tests to test_script.m and updated the README.md and DESCRIPTION

statistics-resampling-v5.6.0

This release includes some new features. In particular, to the bootlm function also includes Extended (Efron) Information Criterion (EIC), relative likelihood ratios and EIC weights for stepwise regression. There is also a bug fix that allows bootbayes to supports NaN value for PROB input argument.

statistics-resampling-v5.5.18

Various improvements relating to the 'auto' prior option of the 'bayesian' method:

• Performance enhancement in cases where sample sizes (and therefore the automatically calculated prior) are equal; only one simulation is necessary, rather than a separate simulation for each sample.
• Rather than using a single prior for bayesian bootstrap of two samples in the posthoc tests, bootlm now uses a separate prior for each sample. This is a better approach when sample sizes are unequal.
  Also added a warning in the bootbayes function when the PRIOR is set to 0
  Edited demo 13 in the bootlm function to avoid constructing credible intervals when using a prior that sets automatically to 0.

statistics-resampling-v5.5.17

Enhancement to the bootwild function - the original calculations of standard errors (using HC0 and CR0) have been replaced with the less biased estimates using HC1 and CR1. The calculation of Cohen's d for estimates relating to posthoc tests in bootlm (when 'method' is 'wild') now uses the HC1/CR1 estimates of standard error.

statistics-resampling-v5.5.16

Fixed bug in bootstrp that prevented it from reshaping output correctly when applying vectorized evaluation of bootfun on the data

statistics-resampling-v5.5.15

Fixed syntax bug that triggered error when printing output for asymmetric bootstrap-t confidence intervals

statistics-resampling-v5.5.14

Added option to compute bias-corrected confidence intervals with the bootint function, and made minor changes to documentation. Updated matlab toolbox and online manual with the recent changes.

statistics-resampling-v5.5.13

Improved numerical accuracy when solving linear systems in bootlm, bootwild, and bootbayes. Corrected dates that files were modified and updated matlab toolbox and online manual

statistics-resampling-v5.5.12

Minor update that includes a fail safe in the bootstrp function when using weighted resampling probabilities. The mean of the bootstrap distribution(s) has been added to the 'stats' output of the bootstrp function. Some minor corrections and additions made to the documentation.

statistics-resampling-v5.5.11

Minor update which includes modification to the calculation of influence values (used for BCa intervals) such that they are consistent with the 'empinf' function in the R 'boot' package for statistics with non-zero bias