Skip to content

tramarobin/fctSnPM

master
Switch branches/tags
Code

Latest commit

 

Git stats

Files

Permalink
Failed to load latest commit information.
Type
Name
Latest commit message
Commit time
 
 
 
 
 
 
src
 
 
 
 
 
 

fctSnPM DOI View fctSnPM on File Exchange

Using spm1d package (v.0.4.3), computes ANOVA and post-hoc tests from anova1 to anova3rm, with a non-parametric approach (permutation tests). The type of ANOVA (if required) and post-hoc tests are chosen regarding the independent or repeated measure effects given in parameters. The function automatically adapts to 1D and 2D data.

The general usage is:

snpmAnalysis=fctSnPM(data, independantEffects, repeatedMeasuresEffects, varargin)

Table of contents

Warnings

  • Unbalanced two- and three-way repeated-measures ANOVA results have not been verified. Please refer to the spm1d documentation about unbalanced design
  • Post-hoc tests with Bonferonni correction are only approximate and conservative. Refer to the spm1d documentation about post hoc tests for more information.
  • Some warnings are displayed if the number of permutations is automatically modified or not sufficient to perform the analysis with the defined alpha risk.
  • Effect sizes and confidence intervals are not validated and are just given as supplementary information.
  • For more information, please visit the smp1d documentation

Caution

  • For independantEffects and repeatedMeasuresEffects, avoid underscore (_) or minus (-) signs. Spaces are OK. It affects the recognition of sub-data for the multiples comparisons
  • Use '/' when indicating a saving directory (savedir option)
  • Once you are ok with the created figure, increase the number of permutations (multiPerm or Perm options) to achieve numerical stability. 1000 permutations are a good start but if your hardware allows it, the more is the better

Statement of need

Most of physiological data measured during human movement are continuous and expressed in function of time. However, researchers predominantly analyze extracted scalar values from the continuous measurement, as the mean, the maximum, the amplitude, or the integrated value over the time. Analyzing continuous values (i.e., time series) can provide more information than extracted indicators, as the later discards one dimension of the data among the magnitude and localization in time. In addition, oscillatory signals such as muscle vibrations and electromyograms contain information in the temporal and frequency domains. Once again, scalar analysis reduces the information at only one dimension by discarding two dimensions among the magnitude and the localization in the time and/or frequency domain.

Factorial SnPM (fctSnPM) allows MATLAB users to create figures of 1D and 2D SnPM analysis for ANOVA and post-hoc designs. The statistical analysis is also saved in .mat files.
This function synthetizes the main and interaction effects to display only the significant post-hoc regarding the results of the ANOVA.
For post-hoc for interaction effects, the main effect is also displayed if located elsewhere than the interaction effect.

Citing fctSnPM

  • This function: Trama et al., (2021). fctSnPM: Factorial ANOVA and post-hoc tests for Statistical nonParametric Mapping in MATLAB. Journal of Open Source Software, 6(63), 3159, https://doi.org/10.21105/joss.03159
  • For spm1d: Pataky TC (2010). Generalized n-dimensional biomechanical field analysis using statistical parametric mapping. Journal of Biomechanics 43, 1976-1982.
  • For permutation tests: Nichols TE, Holmes AP (2002). Nonparametric permutation tests for functional neuroimaging: a primer with examples. Human Brain Mapping 15(1), 1–25.
  • spm1d package for matlab can be downloaded at: https://github.com/0todd0000/spm1dmatlab

MATLAB Release Compatibility

Compatible from Matlab R2017b

Outputs

snpmAnalysis

snpmAnalysis.mat is a structure composed of the results of the statistical analysis (ANOVA + Post Hoc).

snpmAnalysis.anova is composed of different fields:

  • type is the type of ANOVA performed
  • effectNames is a structure (one cell for each effect) that represent the names of the effects tested (mains and interactions)
  • alpha is the alpha risk choosen for the anova (default is 0.05 (5%)).
  • pCritical is the alpha risk used for the anova. Warning message is displayed if this value is modified.
  • nPermutations is the number of permutations performed for the anova.
  • maxPermutations is the number of maximal permutations possible for the anova.
  • Fcontinuum is a structure that represent the F-value for each node
  • Fthreshold is a structure that represent the statistical threshold for the F-values (statistical inference)
  • Fsignificant is a structure that contains the logical for the significance (Fcontinuum > Fthreshold) of each effect of the ANOVA (1 if significant, 0 if not).
  • clusterLocation is a structure (one for each significant cluster) that contains the location (start and end as indexes) of each significant cluster.
  • clusterP is a structure (one for each significant cluster) that contains the p-value of each significant cluster.

clusterLocation and clusterP are created only in one dimension

snpmAnalysis.posthoc is a strucure of cells (one for each effect of the ANOVA) composed of different fields:

  • data.names is a structure that contains the name of the conditions used in the analysis (\cap is the union of different conditions for interactions).
  • data.continuum is a structure that contains the data used in the analysis.

For the following outputs, the number of cells of the structure correspond to the number of t-tests performed, one t-test corresponds to one cell.

  • differences.names is the name of the conditions (the first minus the second) used in the differences and t-tests.

  • differences.continuum is the data used to plot differences.

  • differences.continuumRelative is the data used to plot relative differences

  • differences.ES is the effect size that correspond to the differences (Hedge's g)

  • differences.ESsd is the standard deviation of the effect size.

  • tTests.type is the type of t-test performed (independent or paired)

  • tTests.names is the name of the conditions (the first minus the second) used in the differences and t-tests.

  • tTests.nWarning represents the number of warnings displayed during the analysis: 0 is OK, 1 means the number of permutations was reduced but pCritical = pBonferroni, 2 means that the number of permutations was reduced and pCritical > pBonferroni. In this case, more subjects are required to performed the analysis. %* tTests.alpha is the alpha risk choosen for the post hoc tests before Bonferroni correction (default is the same as the ANOVA).

  • tTests.alpha is the alpha risk choosen for the post hoc tests before Bonferroni correction (default is the same as the ANOVA).

  • tTests.warning: only if alpha is modified with alphaT input.

  • tTests.pBonferroni is the alpha risk choosen for the post hoc tests after Bonferroni correction.

  • tTests.pCritical is the alpha risk used for the post hoc tests. Warning message is displayed if this value does not meet pBonferroni.

  • tTests.nPermutations is the number of permutations performed for the t-test.

  • tTests.maxPermutations is the number of maximal permutations possible for the t-test.

  • tTests.Tcontinuum represents the T-value for each node

  • tTests.Tthreshold represents the statistical threshold for the T-values (statistical inference)

  • tTests.Tsignificant contains the logical for the significance (Tcontinuum > Tthreshold) (1 if significant, 0 if not). This value is corrected with the result of the corresponding ANOVA and previous t-tests.

  • tTests.clusterLocation is a structure (one for each significant cluster) that contains the location (start and end as indexes) of each significant cluster.

  • tTests.clusterP is a structure (one for each significant cluster) that contains the p-value of each significant cluster. This value is corrected with inverse Bonferonni correction. clusterLocation and clusterP are created only in one dimension and are not corrected with the result of the ANOVA.

  • tTests.contourSignificant represents a modified T-value continuum to display smoother contour plots. `tTests.contourSignificant is created only in two dimensions.

  • tTests.combi is only created for 2 and 3-way ANOVA and used in saveNplot to define the data to plot.

Figures

Two folders composed of figures in .TIF and .fig are created

ANOVA

Each effect is also display on a specific figure in .TIF format named after effectNames, a floder named FIG is also created and contains the figures in .fig format.
For 1D: The curve represents the Fcontinuum, the horizontal line is the Fthreshold, the highlighted parts of the curve in blue represents the significant cluster(s), and the vertical lines are the start and end indexes. For 2D: The map represents the Fcontinuum, with a colorbar maximum at Fthreshold, the white clusters represent the significant clusters.

Post Hoc

This folder contains additional folders (0 (for anova1), 3 (for anova2) or 7 (for anova3)) that contain figures and metrics representing the different post-hoc tests.

Interaction folders (AxB or AxBxC) contain 2 or 3 folders in which one effect is investigated.

In one dimension

A total of 5 figures with the name of the effect represent the means and standard deviations between subject for each condition (grouped in function of the effect investigated).
Under this plot, a second graph display the result of the ANOVA (or ANOVAs for interaction) and the significant post-hoc tests for pairwise comparisons.
The differences between the 5 figures are the representation of the ANOVA and the disposition of the statistical analysis (subplot or same plot, see optional inputs)

Subfolders: Contains the pairwise comparison results

  • DIFF: Differences plots. Filenames with '%' at the end are the relative differences
  • ES: Effect size plots. Bold blue lines are located at the significant differences (corrected with the ANOVA).
  • SnPM: Tcontinuum and statistical inferences plots. Bold blue lines are located at the significant differences (corrected with the ANOVA).
  • FIG folder contains the above mentioned folder with the figures in .fig format.
In two dimensions

Means maps for each condition are represented in one figure each. The global effect of the post hoc procedure is display on a figure with the name of the effect. Mean maps are represented on the diagonal, pairwise differences on the top-right panel, and pairwise SnPM inferences on the bottom-left panel.

Subfolders: Contains the pairwise comparison results (In one folder for ANOVA1, in 2 or 3 folders for ANOVA2 and ANOVA3)

  • SD: standard deviation of the maps for each condition.
  • DIFF: Differences plots. Filenames with '%' at the end are the relative differences. White clusters represent the significant effect (corrected with ANOVA)
  • ES: Effect size plots. Whites clusters represent the significant effect (corrected with ANOVA)
  • SnPM: Tcontinuum and statistical inferences plots. Whites clusters represent the significant effect (no correction with the ANOVA)
  • FIG folder contains the above mentioned folder with the figures in .fig format.

Examples

In one dimension

The output of D1_ANOVA2_1rm.m for a 2way ANOVA with 1 repeated measure in 1 dimension in the Examples folder is as follows The curves represent the ratio between Quadriceps and Hamstrings during isokinetic tests for two sides (Left and Right) and for two sexes (M and F).

ANOVA results: There is a "Side" effect, with a F-value above the significant threshold of 7.14 between 30 and 85°. alt text

Post-hoc results: The curves represent the means and the standard deviations. Right > Left between 30 and 85°. The significant differences are displayed below the curves using the curves to display the pairwise comparisons. The ANOVA results are also displayed if wanted. alt text

In two dimensions

The output of D2_ANOVA2_2rm.m for a 2way ANOVA with 2 repeated measures in 2 dimensions is as follows The maps represent the time-frequency analysis of vibratory signal quantified with two devices (ACC and US), and for two muscle activations (Relaxed and Contracted).

ANOVA results: Red clusters circled in white are the zone of significant effects. There are mains and interaction effects. alt text

Post-hoc results for main effects: Mean maps for each condition are displayed. Statistical results are shown within each map of differences on the top (Device effect) and at the bottom (Activation effect). The significant cluster are circled in white. alt text

Post-hoc results for interaction effects: Mean maps for each condition are displayed. Statistical results are shown within each map of differences on the right (Activation effect) and at the bottom (Device effect). The significant cluster are circled in white. Note that the main effects are also displayed. alt text

Using fctSnPM

Installation

Install this package by adding the src directory and its subdirectories to the MATLAB path. One way to do this is to call: addpath(genpath("./fctSnPM/src")), where ./fctSnPM/src is the full path to the src directory. Refer to the MATLAB documentation regarding search paths for alternative ways to set the path for current and future sessions.

Usage

snpmAnalysis=fctSnPM(data, independantEffects, repeatedMeasuresEffects, varargin)

Obligatory inputs

  • data is a x by y cell array, with x corresponding to subjects and y corresponding to repeated measures. Each cell contains a column vector (1D) or a matrix (2D) corresponding to the mean of the subject and condition.
  • independantEffects is a cell array defining the independent effects. There must be 1 cell by effect. Each cell contains the name ('char') of each modality for the given subject and must correspond to the number of subjects/rows (x). For instance independantEffects{1}={'Male' 'Male' 'Female' 'Female'} if the two first subjects are Male and the two last are Female.
  • repeatedMeasuresEffects is a cell array defining the repeated measure effects. There must be 1 cell by effect. Each cell contains the name ('char') of each modality for the given condition and must correspond to the number of conditions/columns (y). For instance repeatedMeasuresEffects{1}={'PRETEST' 'POSTTEST' 'RETEST'} if there was 3 times of measurement.

Optional inputs

Optional inputs are available to personalize the figures.

snpmAnalysis=fctSnPM(data, independantEffects, repeatedMeasuresEffects, 'Optional Input Name', value)

Utilities

These options act on the name of the created folders.

  • savedir is the path to the save directory. Default is empty and ask you to choose or create a folder. If filled, the existing data might be overwritten (@ischar).
  • effectsNames are the names of the effects (the independent effects must be named first). Default names the effect {'A','B','C'} (@iscell).
  • plotSub is an option (1 to activate) to plot the individual mean for each subject. Default = 0 and don't plot the mean (@isnumeric).
  • nameSub is an option to name the different subjects. Default names the subject are '1', '2', etc... (@iscell).

Statistical parameters

These options act at a statistical level, modifying the alpha error or the number of permutations.

  • alpha is the alpha error risk for the ANOVA. Default is 0.05 (@isnumeric).
  • alphaT. Do not modify except for exploratory purposes. alphaT is the original alpha used for post hoc tests (Bonferonni correction is applied after as alphaT/number of comparisons. Default is the same as alpha (@isnumeric).
  • multiPerm define the number of permutations as multiPerm/alpha. Default is 10 and corresponds to 200 permutations for 5% risk (@isnumeric).
    Must be increased for better reproductibility.
  • Perm is a fixed number of permutations. This input overrides the multiPerm ans is not recommended (@isnumeric).
    Specified either multiPerm or Perm, but not both.
  • maximalPerm is the limit of the number of maximal permutations in case of too many multiple comparisons. By default this value is set to 10000 to not take too many computation time (@isnumeric).
  • doAllInteractions By default, all post hoc tested are made even if ANOVA did not revealed interaction. Use 0 to performed only post-hoc when interaction was found (@isnumeric).
  • ignoreAnova By default, consider the ANOVA significant location to interpret post-hoc. Use 1 to interpret only post-hoc tests but it is not recommended (@isnumeric).

General plot parameters

These options can modify the general aspect of the figures for 1D and 2D.

  • ylabel are the labels of Y-axis. No name is display by default (@ischar).
  • xlabel are the labels of X-axis. No name is display by default (@ischar).
  • samplefrequency changes the xticks to correspond at the specified frequency. The default value is 1 (@isnumeric).
  • xlimits changes the xticks to correspond to the specified range (it can be negative). The format is, for instance [0 100] (@isnumeric).
    Specified either samplefrequency or xlimits, but not both.
  • ylimits changes the yticks to correspond to the specified range (it can be negative). The format is, for instance [0 100] (@isnumeric).
  • nTicksX is the number of xticks displayed. By default it depends on the size of the figure in 1D, and corresonds to 5 in 2D (@isnumeric).
  • nTicksY is the number of yticks displayed. By default it depends on the size of the figure in 1D, and corresonds to 4 in 2D (@isnumeric).
  • imageResolution is the resolution in ppp of the tiff images. The default value is 96ppp (@isnumeric).
  • imageSize is the size of the image in cm. For instance, [X] or [X X] creates X by X cm images, [X Y] creates X by Y cm images. The default image size is 720*480 pixels (@isnumeric).
  • imageFontSize is the font size of images. By default it is a size 12 fontsize (@isnumeric).
  • linestyle In 1D: lineStyle for plots (default is a solid line for the mean, a dashed line for the SD). Specify linestyle for each modality in cell, apply each style to each modality (independant effect first). lineStyle{1}={':' '-' '--';':' '-' '--'}; to have a dotted line, a solid line, and a dashed line for the first effect (first row is for the means, second row for the sd) // In 2D: linestyle of the contour plot

2D plot parameters

These option are specific to 2D plots.

  • colorMap is the colormap used for means, standard deviations, ANOVA and effect sizes plots. The default colomap is cbrewer('Reds').
  • colorMapDiff is the colormap used for differences, relative differences and post-hoc SnPM plots. The default colomap is cbrewer('RdBu'). Colormaps can be defined with cbrewer (distributed in this funtion): Charles (2020). cbrewer: colorbrewer schemes for Matlab, MATLAB Central File Exchange. Retrieved December 11, 2020.
  • colorbarLabel is the name of the colorbar label. No name is display by default (@ischar).
  • limitMeanMaps defines limit of the colorbar. By default, the maps wont necessary be with the same range but will be automatically scaled at their maximum. A value of X will make the colorbar going from 0 to X for all plots to make them easier to compare (@isnumeric).
  • displaycontour displays contour map on differences and size effect maps. Use 0 to not display the statistical analysis but it is not recommended (@isnumeric).
  • contourcolor is the color of the line for the contour plot. The default color is white. RGB or 'color' is accepted.
  • dashedColor is the color of the non-significant zones of the contour plot. The default color is white black. Use RGB triplet [0 0 0].
  • transparancy is the transparency of the non-significant zones. The default value is 50 and can range from 0 (transparent) to 255 (opaque) (@isnumeric).
  • lineWidth is the linewidth of the contour plot. The default value is 2.5 (@isnumeric).
  • diffRatio scales the difference maps at limitMeanMaps*diffRatio. The default value is 0.33 (@isnumeric).
  • relativeRatio scales the relative differences maps at +-relativeRatio. By default, the maps wont necessary be with the same range but will be automatically scaled at their maximum (@isnumeric).

1D plot parameters

These option are specific to 1D plots.

  • CI is the confidence interval used instead of standard deviation. By default, standard deviations are displayed. Use 0.7 to 0.999 to display 70% to 99.9% confidence interval, 0 to display SEM, or negative value to not dispaly dispersion (@isnumeric).
  • colorLine is the colorline for plots (default is "lines"). Use rgb triplet. If in cell, apply each color to each effect (independant effect first). colorLine{1}=[rgb('blue'); rgb('magenta'); rgb('black')] to have blue, magenta and black lines for the first effect.
  • transparancy1D is the transparency for the SD, CI or SEM. The default value is is 0.1 (@isnumeric).
  • ratioSnPM is the ratio of SnPM subplot relative to total figure. The default value is 1/3 of the figure. Must be indicated as [1 3] to have a 1/3 ratio (@isnumeric).
  • yLimitES is the y-axis limits for ES representation. By default, the maps wont necessary be with the same range but will be automatically scaled at their maximum.
  • SnPMPos is the position of SnPM plot, default SnPM analysis is displayed at the bottom of the figure. Any value will set the position to up.
  • aovColor is the color of ANOVA on SnPM plot. The default color is black. Use 'color' or rgb triplet.

Optional functions

in addition of fctSnPM, this repository contains two similar funtions.

  • fctSnPMS performs the same analysis than fctSnPM, however, the figures are not ploted and saved. The inputs are the same at the exception that there is no savedir and no plot parameters.
  • onlyPlot permits to plot a part of the analysis obtain with fctSnPM and fctSnPMS. In 1D, the ANOVA and the means with SnPM analysis are displayed. In 2D, the ANOVA and the post-hoc tests are displayed.
  • saveNplot permits to save and plot the entire analysis obtain with fctSnPM and fctSnPMS.

The general use of these funtions are:

snpmAnalysis=fctSnPMS(data, independantEffects, repeatedMeasuresEffects, 'Optional Input Name', value)
onlyPlot(snpmAnalysis,'Optional Input Name', value)
saveNplot(snpmAnalysis,'Optional Input Name', value)

You can use onlyPlot to have an insight on the figures before saving to adjust the plot parameters.
It may be useful to use saveNplot when a 2D analysis is performed, it may permit to redo quickly figures without the long time of analysis.

Community guidelines

You can create a new issue in the Issues section of this repository to:

  • Contribute to the software
  • Report issues or problems with the software
  • Seek support

I will try to answer as quickly and accurately as possible. Thank's for using this function.

About

Using spm1d package (v.0.4.3), compute anova and post-hoc tests from anova1 to anova3rm, with a non-parametric approach (permutation tests)

Resources

License

Stars

Watchers

Forks

Packages

No packages published