statistics-bootstrap-v5.1.6

Minor changes in v5.1.6 since v5.1.5:

• Made BOOTSAM a documented input argument
• Minor tweaks to coding style and documentation

statistics-bootstrap-v5.1.5

Changes in v5.1.5 since v5.1.4:

• Fixed bug in 'boot' function (specifically the windows versions of boot.mex and it's source code boot.cpp). This bug prevented uniform resampling across the whole sample when n*nboot became too large. This bug was introduced in commit f07cb46 (8 Dec 2022) and was present in releases 5.1.0-5.1.4. The fix has reintroduced C++11 dependency for compiling boot.mex.
• Fixed incorrect precompiled binaries for Octave mex files, which were 32-bit, when they should have been 64-bit windows
• Added a series of significant performance enhancements in bootknife by vectorizing the code for computing bias, standard errors and confidence intervals for each element of the return value of bootfun. The performance enhancement is also achieved by reducing the number of (bootfun) function evaluations.
• Improved algorithm for computing quantiles (by linear interpolation) after double bootstrap in bootknife and bootci.
• Added message printed to stdout when kernel density estimation fails while computing quantiles after single bootstrap (bootknife).
• Improved the consistency of coding style across the functions in the package.

statistics-bootstrap-v5.1.4

!!! IMPORTANT NOTICE: This version has a significant bug introduced in 5.1.0 that affects bootstrap resampling, particularly for Windows users. Please use version 5.1.5 or above instead. !!!

Changes in v5.1.4 since v5.1.3:

• The semiparametric method used before in bootcoeff and bootemm was sensitive to heteroskedasticity. The non-parametric resampling now implemented is robust to violations of the assumptions of homoskedasticity and normality. These changes also widen the options in terms of how the intervals can be constructed (e.g. BCa, calibration etc). This approach is also taken by the R function bootstrap_parameters for lm models.
• Fixed bug that caused excessively large STATS structure array to be preallocated in bootknife function when bootfun on the data input argument returns a vector
• Fixed a bug in the bootknife function that could lead to errors if bootfun on the data input argument returns a vector AND when the some bootstrap statistics contain NaN
• Fixed minor bug in bootknife during stratified sampling that threw an error if any stratum had only 1 observation/row
• Replaced call to regress function in bootknife to remove it's dependency on the Statistics package/toolbox
• Added linear interpolation as fail safe for kernel density estimation when constructing bootstrap confidence intervals using bootknife.

statistics-bootstrap-v5.1.3

!!! IMPORTANT NOTICE: This version has a significant bug introduced in 5.1.0 that affects bootstrap resampling, particularly for Windows users. Please use version 5.1.5 or above instead. !!!

Minor performance and reporting enhancements:

• Performance enhancement in bootknife when evaluating univariate bootfun over matrix input
• bootknife: Added more complete information to printed output about the resampling method and confidence interval type used
• bootknife: Improved consistency in the reporting of the percentiles used to construct confidence intervals
• Simplified bootcoeff and bootemm by removing expanded probabilities; further development on these functions expected in the next update or so
• Minor changes to help info, comments and demos

statistics-bootstrap-v5.1.2

!!! IMPORTANT NOTICE: This version has a significant bug introduced in 5.1.0 that affects bootstrap resampling, particularly for Windows users. Please use version 5.1.5 or above instead. !!!

Bug fix

• fixes bug introduced in v5.1.1 when printing output when bootfun is mean

statistics-bootstrap-v5.1.1

!!! IMPORTANT NOTICE: This version has a significant bug introduced in 5.1.0 that affects bootstrap resampling, particularly for Windows users. Please use version 5.1.5 or above instead. !!!

Minor changes

• corrected printed function name when bootfun is provided as function name and additional input arguments in a cell array
• corrected printed nominal coverage of interval when bootfun is 'mean'

statistics-bootstrap-v5.1.0

!!! IMPORTANT NOTICE: This version has a significant bug introduced in 5.1.0 that affects bootstrap resampling, particularly for Windows users. Please use version 5.1.5 or above instead. !!!

Some of the changes include:

Bug fixes:

• Enabled handling of multiple data arguments with varying number of columns (e.g. for case resampling in linear regression) using bootknife, bootci and bootstrp
• Fixed minor bug in make file
• Made boot and smoothmedian accept empty optional input arguments and set them to default values
• bootci, bootstrp and bootknife now omit and gracefully accommodate for instances of NaN in the bootstrap statistics returned by bootfun

Feature changes:

• Swapped 4th and 5th input arguments to boot
• Added more BISTs and demos to functions (which can be used in Octave)
• Added more interval types to bootci function
• Added ability to construct 1-sided or (symmetric) 2-sided double bootstrap confidence intervals
• Added expansion of probabilities to intervals constructed by bootcoeff and bootemm
• Switched to using kernel density estimation with shrinkage correction for obtaining quantiles from the bootstrap statistics (Gaussian kernel)
• Added double bootstrap calculation of standard error
• Added option in make file to choose between compiling source code or using precompiled binaries
• Added new function cor in param folder: vectorized function for more rapid computation of Pearsons correlation coefficient from pairs of matrices

Other changes:

• Some formatting changes to code or printed function output.
• Some changes to organisation of coded
• Updated help documentation and comments in code
• Minor performance enhancement to double bootstrap (by avoiding error checking in the nested level of resampling)

statistics-bootstrap-v5.0.2

Improved the handling of NaN values in the smoothmedian MEX (and .m) function files. Instead of throwing an error, the MEX files omit NaN values before performing calculations, whereas the m-file includes NaN values in it's calculations.

statistics-bootstrap-v5.0.1

• Added demos to all functions
• Made BISTs more robust
• Fixed some minor bugs and backwards compatibility

statistics-bootstrap-v5.0.0

• fixed minor bugs in bootknife and bootnhst
• revamped the MATLAB-like bootci and bootstrp functions - they are now wrappers for bootknife and no longer require BOOTFUN to return a scalar value
• removed all legacy functions
• added octave-specific function (bootcoeff) for bootstrapping the coefficients from a linear model (requires Statistics package v 1.5+)
• added octave-specific function (bootemms) for bootstrapping the estimated marginal means from a linear model (requires Statistics package v 1.5+)
• added functionality to bootmode to accelerate computations using parallel computing
• added some BISTs or demos to some functions (but the tests and demos can only be run in Octave)