Skip to content

Commit

Permalink
Update ibootnhst.m
Browse files Browse the repository at this point in the history
  • Loading branch information
acp29 committed Feb 19, 2022
1 parent 208039a commit 6335b2a
Showing 1 changed file with 68 additions and 57 deletions.
125 changes: 68 additions & 57 deletions inst/ibootnhst.m
View file
Original file line number Diff line number Diff line change
Expand Up @@ -8,8 +8,8 @@
% ibootnhst(...,'bootfun',bootfun);
% ibootnhst(...,'nboot',nboot);
% ibootnhst(...,'ref',ref);
% ibootnhst(...,'block',blocks);
% ibootnhst(...,'nested',clusters);
% ibootnhst(...,'block',blocks); % for bootfun 'mean' or 'robust' only
% ibootnhst(...,'nested',clusters); % for bootfun 'mean' or 'robust' only
% ibootnhst(...,'Options',paropt);
% ibootnhst(...,'alpha',alpha);
% p = ibootnhst(...,'dim',dim)
Expand Down Expand Up @@ -74,7 +74,7 @@
% jackknife resampling instead. Since jackknife resampling is not random,
% setting nboot(2) to 0 can also reduce monte carlo error but jackknife
% is only suitable where bootfun is a smooth function of the data (e.g.
% 'mean' or 'smoothmedian').
% 'mean' or 'smoothmedian', a.k.a. 'robust').
%
% ibootnhst(...,'ref',ref) also sets the GROUP to use as the reference
% GROUP for post hoc tests. For a one-way experimental design or family of
Expand Down Expand Up @@ -109,8 +109,8 @@
% indicate block membership of the data rows. Data in a data block are
% centered and the resampling is stratified to impose restrictions on the
% exchangeability of data to within blocks. Since the data must be centered
% using bootfun, this feature only supports the following values for bootfun:
% 'mean', 'median', 'pseudomedian', 'smoothmedian' (or 'robust'). This option
% using bootfun, this feature only supports location parameters, of which
% ibootnhst supports bootfun being either 'mean' or 'robust'. This option
% is appropriate when the family of tests has a randomized block design or
% one-way repeated measures layout. See end of this help for an example. The
% 'block' option here should not be confused with the block option in ibootci.
Expand All @@ -125,14 +125,16 @@
% specifying clusters causes ibootnhst to resample residuals, the bootstrap
% becomes semi-parametric. The clusters input argument can be used to
% accomodate for a single level of nesting in heirarchical data structures.
% This resampling strategy is appropriate when the family of tests has a
% split plot design layout, and is a bootstrap version of a nested 1-way
% ANOVA. See end of this help for an example of this application. This
% function will return an error if any GROUP is not represented by more than
% one cluster, but there is no restriction on the number of data rows in any
% cluster. Note that the value in nboot(2) is ignored since specifying
% cluster identifiers enforces cluster-jackknife to compute standard errors
% and studentized test statistics. If empty, this argument is ignored.
% Since the bootstrap becomes semiparametric, this feature only supports
% location parameters, of which ibootnhst supports bootfun being either
% 'mean' or 'robust'. This resampling strategy is appropriate when the
% family of tests has a split plot design layout, and is a bootstrap
% version of a nested 1-way ANOVA. See end of this help for an example of
% this application. This function will return an error if any GROUP is not
% represented by more than one cluster, but there is no restriction on the
% number of data rows in any cluster. Note that the value in nboot(2) is
% ignored since specifying cluster identifiers enforces cluster-jackknife
% to compute standard errors. If empty, this argument is ignored.
%
% ibootnhst(...,'Options',paropt) specifies options that govern if and
% how to perform bootstrap iterations using multiple processors (if the
Expand Down Expand Up @@ -219,7 +221,7 @@
% variety of experimental designs:
%
%
% EXAMPLE 1:
% EXAMPLE 1A:
% ONE-WAY ANOVA WITH EQUAL SAMPLE SIZES: Treatment vs. Control (1)
%
% >> y = [111.39 110.21 89.21 76.64 95.35 90.97 62.78;
Expand Down Expand Up @@ -264,6 +266,50 @@
% | 6 | 6 |
% | 7 | 7 |
%
% EXAMPLE 1B:
% ROBUST ONE-WAY ANOVA WITH EQUAL SAMPLE SIZES: Treatment vs. Control (1)
%
% >> y = [111.39 110.21 89.21 76.64 95.35 90.97 62.78;
% 112.93 60.36 92.29 59.54 98.93 97.03 79.65;
% 85.24 109.63 64.93 75.69 95.28 57.41 75.83;
% 111.96 103.40 75.49 76.69 77.95 93.32 78.70];
% >> g = [1 2 3 4 5 6 7;
% 1 2 3 4 5 6 7;
% 1 2 3 4 5 6 7;
% 1 2 3 4 5 6 7];
% >> ibootnhst (y(:),g(:),'ref',1,'bootfun','robust');
%
% Summary of bootstrap null hypothesis (H0) significance test(s)
% ******************************************************************************
% Overall hypothesis test from single-step maxT procedure
% H0: Groups of data are all sampled from the same population as data in ref
%
% Maximum t(21) = 3.14, p-adj = .023
% ------------------------------------------------------------------------------
%
% POST HOC TESTS with control of the FWER by the single-step maxT procedure
% ------------------------------------------------------------------------------
% | Comparison | Reference # | Test # | Difference | t(df)| p-adj |
% |------------|--------------|--------------|-------------|---------|---------|
% | 1 | 1 | 2 | -6.66e+00 | 0.59 | .974 |
% | 2 | 1 | 3 | -3.02e+01 | 2.66 | .069 |
% | 3 | 1 | 4 | -3.56e+01 | 3.14 | .023 |*
% | 4 | 1 | 5 | -1.62e+01 | 1.43 | .562 |
% | 5 | 1 | 6 | -2.00e+01 | 1.76 | .373 |
% | 6 | 1 | 7 | -3.49e+01 | 3.07 | .025 |*
%
% Where degrees of freedom (df) = 21
%
% ------------------------------------------------------------------------------
% | GROUP # | GROUP label |
% |------------|---------------------------------------------------------------|
% | 1 | 1 |
% | 2 | 2 |
% | 3 | 3 |
% | 4 | 4 |
% | 5 | 5 |
% | 6 | 6 |
% | 7 | 7 |
%
% EXAMPLE 2A:
% COMPARISON OF TWO INDEPENDENT GROUPS WITH UNEQUAL SAMPLE SIZES
Expand Down Expand Up @@ -639,7 +685,7 @@
% Sampling vs. Smoothing, Proceedings of the Section on Statistics
% and the Environment, American Statistical Association, 2924-2930.
%
% ibootnhst v1.7.0.0 (11/02/2022)
% ibootnhst v1.7.1.0 (19/02/2022)
% Author: Andrew Charles Penn
% https://www.researchgate.net/profile/Andrew_Penn/
%
Expand Down Expand Up @@ -755,19 +801,13 @@
else
bootfun = @smoothmedian;
end
elseif all(bootfun(data) == median(data))
if (nboot(2) == 0)
error('jacknife resampling cannot be used to calculate standard errors of the median')
else
if ~isempty(strata)
error('bootfun must be ''mean'' or ''robust'' for block or repeated measures designs.')
end
if ~isempty(clusters)
error('cluster-jacknife resampling cannot be used to calculate standard errors of the median')
error('bootfun must be ''mean'' or ''robust'' for nested designs.')
end
if nvar > 1
% Grand median for multivariate data
bootfun = @(data) median(median(data,dim));
else
bootfun = @median;
end
end
elseif isa(bootfun,'char')
if strcmpi(bootfun,'mean')
Expand All @@ -784,41 +824,12 @@
else
bootfun = @smoothmedian;
end
if ~isempty(clusters)
fprintf('\nNote: Switching to parametric resampling. Cluster resampling here uses residuals from the column-wise mean vector.')
end
elseif strcmpi(bootfun,'pseudomedian')
if nvar > 1
% Grand pseudomedian for multivariate data
bootfun = @(data) pseudomedian(pseudomean(data,dim));
else
bootfun = @pseudomedian;
end
if ~isempty(clusters)
fprintf('\nNote: Switching to parametric resampling. Cluster resampling here uses residuals from the column-wise mean vector.')
end
elseif strcmpi(bootfun,'median')
if (nboot(2) == 0)
error('jacknife resampling cannot be used to calculate standard errors of the median')
end
if ~isempty(clusters)
error('cluster-jacknife resampling cannot be used to calculate standard errors of the median')
end
if nvar > 1
% Grand median for multivariate data
bootfun = @(data) median(median(data,dim));
else
bootfun = @median;
end
if ~isempty(clusters)
fprintf('\nNote: Switching to parametric resampling. Cluster resampling here uses residuals from the column-wise mean vector.')
end
else
if ~isempty(strata)
error('bootfun is not (and must be) a recognised location parameter for the center of the data distribution.')
error('bootfun must be ''mean'' or ''robust'' for block or repeated measures designs.')
end
if ~isempty(clusters)
fprintf('\nNote: Switching to parametric resampling. Cluster resampling here uses residuals from the column-wise mean vector.')
error('bootfun must be ''mean'' or ''robust'' for nested designs.')
end
end
end
Expand Down Expand Up @@ -898,7 +909,7 @@
l = numel(unique(strata)); % number of strata
end

% If applicable, center each stratum on stratum (grand) mean
% If applicable, center each stratum on it's respective (grand) mean or smoothed median
% Note that the bootstrap becomes semiparametric
if ~isempty(strata)
sk = unique(strata);
Expand Down

0 comments on commit 6335b2a

Please sign in to comment.