Testing and Plotting Procedures for Biostatistics

Description

Contains miscellaneous functions useful in biostatistics, mostly univariate and multivariate testing procedures with a special emphasis on permutation tests. Many functions intend to simplify user's life by shortening existing procedures or by implementing plotting functions that can be used with as many methods from different packages as possible.

Details

Author(s)

Maxime HERVE

Maintainer: Maxime HERVE <maxime.herve@univ-rennes.fr>

References

Document : "Aide-memoire de statistique appliquee a la biologie - Construire son etude et analyser les resutats a l'aide du logiciel R" (available on CRAN)

Anova Tables for Cumulative Link (Mixed) Models

Description

These functions are methods for Anova to calculate type-II or type-III analysis-of-deviance tables for model objects produced by clm and clmm. Likelihood-ratio tests are calculated in both cases.

Usage

## S3 method for class 'clm'
Anova(mod, type = c("II", "III", 2, 3), ...)

## S3 method for class 'clmm'
Anova(mod, type = c("II", "III", 2, 3), ...)

Arguments

Details

See help of the Anova for a detailed explanation of what "type II" and "typ III" mean.

Value

See Anova.

Author(s)

Maxime HERVE <maxime.herve@univ-rennes1.fr>

See Also

Anova, clm, clmm

Cross validation

Description

Performs cross validation with correspondence discriminant analyses.

Usage

CDA.cv(X, Y, repet = 10, k = 7, ncomp = NULL, method = c("mahalanobis",
  "euclidian"))

Arguments

Details

The training sets are generated in respect to the relative proportions of the levels of Y in the original data set (see splitf).

Value

Author(s)

Maxime HERVE <maxime.herve@univ-rennes1.fr>

See Also

discrimin.coa

Examples

require(ade4)
data(perthi02)
## Not run: CDA.cv(perthi02$tab,perthi02$cla)

Significance test for CDA

Description

Performs a significance test for correspondence discriminant analysis. See Details.

Usage

CDA.test(X, fact, ncomp = NULL, ...)

Arguments

Details

CDA consists in two steps: building a correspondence analysis (CA) on X, then using row coordinates on all CA components as input variables for a linear discriminant analysis. CDA.test builds the intermediate CA, then uses the first ncomp components to test for an effect of fact. If 1 component is used the test is an ANOVA, if more than 1 component are used the test is a MANOVA.

Value

An ANOVA or MANOVA table.

Author(s)

Maxime HERVE <maxime.herve@univ-rennes1.fr>

See Also

discrimin.coa, summary.manova

Examples

require(ade4)
data(perthi02)

CDA.test(perthi02$tab,perthi02$cla)

Cross validation

Description

Performs cross validation with DIABLO (block.plsda or block.splsda).

Usage

DIABLO.cv(x, method = c("mahalanobis.dist", "max.dist", "centroids.dist"),
  validation = c("Mfold", "loo"), k = 7, repet = 10, ...)

Arguments

Details

The function uses the weighted predicted classification error rate (see perf).

Value

Author(s)

Maxime HERVE <mx.herve@gmail.com>

See Also

block.plsda, block.splsda, perf

Examples

## Not run: 
require(mixOmics)
data(nutrimouse)
data <- list(gene=nutrimouse$gene,lipid=nutrimouse$lipid,Y=nutrimouse$diet)
DIABLO <- block.plsda(X=data,indY=3)
DIABLO.cv(DIABLO)

## End(Not run)

Significance test based on cross-validation

Description

Performs a permutation significance test based on cross-validation with DIABLO (block.plsda or block.splsda).

Usage

DIABLO.test(x, method = c("mahalanobis.dist", "max.dist", "centroids.dist"),
  validation = c("Mfold", "loo"), k = 7, nperm = 999, progress = TRUE, ...)

Arguments

Details

The function uses the weighted predicted classification error rate (see perf).

Value

Author(s)

Maxime HERVE <maxime.herve@univ-rennes1.fr>

See Also

block.plsda, block.splsda, perf

Examples

## Not run: 
require(mixOmics)
data(nutrimouse)
data <- list(gene=nutrimouse$gene,lipid=nutrimouse$lipid,Y=nutrimouse$diet)
DIABLO <- block.plsda(X=data,indY=3)
DIABLO.test(DIABLO)

## End(Not run)

G-test for binary variables

Description

Performs a G-test for comparing response probabilities (i.e. when the response variable is a binary variable). The function is in fact a wrapper to the G-test for comparison of proportions on a contingency table. If the p-value of the test is significant, the function performs pairwise comparisons by using G-tests.

Usage

G.bintest(formula, data, alpha = 0.05, p.method = "fdr")

Arguments

Details

If the response is a 0/1 variable, the probability of the '1' group is tested. In any other cases, the response is transformed into a factor and the probability of the second level is tested.

Since a G-test is an approximate test, an exact test is preferable when the number of individuals is small (200 is a reasonable minimum). See fisher.bintest in that case.

Value

Author(s)

Maxime HERVE <maxime.herve@univ-rennes1.fr>

See Also

chisq.bintest, fisher.bintest

Examples

response <- c(rep(0:1,c(40,60)),rep(0:1,c(55,45)),rep(0:1,c(65,35)))
fact <- gl(3,100,labels=LETTERS[1:3])
G.bintest(response~fact)

Pairwise comparisons after a G-test

Description

Performs pairwise comparisons after a global G-test.

Usage

G.multcomp(x, p.method = "fdr")

Arguments

Details

Since a G-test is an approximate test, an exact test is preferable when the number of individuals is small (200 is a reasonable minimum). See multinomial.multcomp in that case.

Value

Author(s)

Maxime HERVE <maxime.herve@univ-rennes1.fr>

See Also

G.test, multinomial.test, multinomial.multcomp

Examples

counts <- c(49,30,63,59)
G.test(counts)
G.multcomp(counts)

G-test

Description

Perfoms a G-test on a contingency table or a vector of counts.

Usage

G.test(x, p = rep(1/length(x), length(x)))

Arguments

Details

If x is matrix, it must be constructed like this:

- 2 columns giving number of successes (left) and fails (right)

- 1 row per population.

The function works as chisq.test :

- if x is a vector and theoretical proportions are not given, equality of counts is tested

- if x is a vector and theoretical proportions are given, equality of counts to theoretical counts (given by theoretical proportions) is tested

- if x is a matrix with two columns, equality of proportion of successes between populations is tested.

- if x is a matrix with more than two columns, independence of rows and columns is tested.

Since a G-test is an approximate test, an exact test is preferable when the number of individuals is small (200 is a reasonable minimum). See multinomial.test in that case with a vector, fisher.test with a matrix.

Value

Author(s)

Maxime HERVE <maxime.herve@univ-rennes1.fr>

See Also

chisq.test, multinomial.test, fisher.test G.multcomp, G.theo.multcomp, pairwise.G.test

Examples

counts <- c(49,30,63,59)
G.test(counts)

Pairwise comparisons after a G-test for given probabilities

Description

Performs pairwise comparisons after a global G-test for given probabilities.

Usage

G.theo.multcomp(x, p = rep(1/length(x), length(x)), p.method = "fdr")

Arguments

Details

Since a G-test is an approximate test, an exact test is preferable when the number of individuals is small (200 is a reasonable minimum). See multinomial.theo.multcomp in that case.

Value

Author(s)

Maxime HERVE <maxime.herve@univ-rennes1.fr>

See Also

G.test, multinomial.test, multinomial.theo.multcomp

Examples

counts <- c(49,30,63,59)
p.theo <- c(0.2,0.1,0.45,0.25)
G.test(counts,p=p.theo)
G.theo.multcomp(counts,p=p.theo)

Significance test for GPA

Description

Performs a permutation significance test based on total variance explained for Generalized Procrustes Analysis. The function uses GPA.

Usage

GPA.test(df, group, tolerance = 10^-10, nbiteration = 200, scale = TRUE,
  nperm = 999, progress = TRUE)

Arguments

Details

Rows of each data frame are randomly and independently permuted.

The function deals with the limitted floating point precision, which can bias calculation of p-values based on a discrete test statistic distribution.

Value

Author(s)

Maxime HERVE <maxime.herve@univ-rennes1.fr>

References

Wakeling IN, Raats MM and MacFie HJH (1992) A new significance test for consensus in Generalized Procrustes Analysis. Journal of Sensory Studies 7:91-96.

See Also

GPA

Examples

require(FactoMineR)
data(wine)

## Not run: GPA.test(wine[,-(1:2)],group=c(5,3,10,9,2))

Type II permutation test for constrained multivariate analyses

Description

This function is a wrapper to anova.cca(...,by="terms") but performs type II tests (whereas anova.cca performs type I).

Usage

MVA.anova(object, ...)

Arguments

Details

See anova.cca for detailed explanation of what is done. The only difference with anova.cca is that MVA.anova performs type II tests instead of type I.

See example of adonis.II for the difference between type I (sequential) and type II tests.

Value

a data frame of class "anova".

Author(s)

Maxime HERVE <maxime.herve@univ-rennes1.fr>

Biplot of multivariate analyses

Description

Displays a biplot of a multivariate analysis. This just consists in superimposing a score plot and a correlation circle (plus centroids of factor levels in constrained analyses, RDA or CCA). The correlation circle is adjusted to fit the size of the score plot.

Usage

MVA.biplot(x, xax = 1, yax = 2, scaling = 2, sco.set = c(12, 1, 2),
  cor.set = c(12, 1, 2), space = 1, ratio = 0.9, weights = 1,
  constraints = c("nf", "n", "f", NULL), sco.args = list(),
  cor.args = list(), f.col = 1, f.cex = 1)

Arguments

Details

This function should not be use directly. Prefer the general MVA.plot, to which all arguments can be passed.

All multivariate analyses covered by MVA.corplot can be used for biplots.

Author(s)

Maxime HERVE <maxime.herve@univ-rennes1.fr>

Examples

require(vegan)
data(iris)
RDA <- rda(iris[,1:4]~Species,data=iris)
MVA.plot(RDA,"biplot",cor.args=list(col="purple"),ratio=0.8,f.col=c("red","green","blue"))

Cross model validation

Description

Performs cross model validation (2CV) with different PLS analyses.

Usage

MVA.cmv(X, Y, repet = 10, kout = 7, kinn = 6, ncomp = 8, scale = TRUE,
  model = c("PLSR", "CPPLS", "PLS-DA", "PPLS-DA", "PLS-DA/LDA", "PLS-DA/QDA",
  "PPLS-DA/LDA", "PPLS-DA/QDA"), crit.inn = c("RMSEP", "Q2", "NMC"),
  Q2diff = 0.05, lower = 0.5, upper = 0.5, Y.add = NULL, weights = rep(1, nrow(X)),
  set.prior = FALSE, crit.DA = c("plug-in", "predictive", "debiased"), ...)

Arguments

Details

Cross model validation is detailed is Szymanska et al (2012). Some more details about how this function works:

- when a discriminant analysis is used ("PLS-DA", "PPLS-DA", "PLS-DA/LDA", "PLS-DA/QDA", "PPLS-DA/LDA" or "PPLS-DA/QDA"), the training sets (test set itself in the inner loop, test+validation sets in the outer loop) are generated in respect to the relative proportions of the levels of Y in the original data set (see splitf).

- "PLS-DA" is considered as PLS2 on a dummy-coded response. For a PLS-DA based on the CPPLS algorithm, use "PPLS-DA" with lower and upper limits of the power parameters set to 0.5.

- if a second analysis is used ("PLS-DA/LDA", "PLS-DA/QDA", "PPLS-DA/LDA" or "PPLS-DA/QDA"), a LDA or QDA is built on scores of the first analysis (PLS-DA or PPLS-DA) also in the inner loop. The classification error rate, based on this second analysis, is used to choose the number of components.

If scale = TRUE, the scaling is done as this:

- for each step of the outer loop (i.e. kout steps), the rest set is pre-processed by centering and unit-variance scaling. Means and standard deviations of variables in the rest set are then used to scale the test set.

- for each step of the inner loop (i.e. kinn steps), the training set is pre-processed by centering and unit-variance scaling. Means and standard deviations of variables in the training set are then used to scale the validation set.

Value

Author(s)

Maxime HERVE <maxime.herve@univ-rennes1.fr>

References

Szymanska E, Saccenti E, Smilde AK and Westerhuis J (2012) Double-check: validation of diagnostic statistics for PLS-DA models in metabolomics studies. Metabolomics (2012) 8:S3-S16.

See Also

predict.MVA.cmv, mvr, lda, qda

Examples

require(pls)
require(MASS)

# PLSR
data(yarn)
## Not run: MVA.cmv(yarn$NIR,yarn$density,model="PLSR")

# PPLS-DA coupled to LDA
data(mayonnaise)
## Not run: MVA.cmv(mayonnaise$NIR,factor(mayonnaise$oil.type),model="PPLS-DA/LDA",crit.inn="NMC")

Correlations of multivariate analyses

Description

Returns correlations of a multivariate analysis.

Usage

MVA.cor(x, xax = 1, yax = 2, set = c(12, 1, 2), space = 1, ...)

Arguments

Details

Many multivariate analyses are supported, from various packages:

- PCA: dudi.pca, rda.

- sPCA: spca.

- IPCA: ipca.

- sIPCA: sipca.

- LDA: lda, discrimin.

- PLS-DA (PLS2 on a dummy-coded factor): plsda. X space only.

- sPLS-DA (sPLS2 on a dummy-coded factor): splsda. X space only.

- CPPLS: mvr. Set 1 is X, set 2 is Y. If set=12 (default), fac is not available and pch,cex, col, lwd can be defined differently for each set. X space only.

- PLSR: mvr, pls, plsR (plsRglm package). Set 1 is X, set 2 is Y. If set=12 (default), fac is not available and pch,cex, col, lwd can be defined differently for each set. X space only.

- sPLSR: pls. Set 1 is X, set 2 is Y. If set=12 (default), fac is not available and pch,cex, col, lwd can be defined differently for each set. X space only.

- PLS-GLR: plsRglm (plsRglm package). Set 1 is X, set 2 is Y. If set=12 (default), fac is not available and pch,cex, col, lwd can be defined differently for each set. Correlations are computed with Y on the link scale.

- PCR: mvr. Set 1 is X, set 2 is Y. If set=12 (default), fac is not available and pch,cex, col, lwd can be defined differently for each set.

- CDA: discrimin, discrimin.coa.

- NSCOA: dudi.nsc. For NSCOA there is no real correlation, but the classical representation of columns is arrows. This is why MVA.corplot was made able to deal with this analysis.

- CCA: cca, pcaiv. Constraints (only quantitative constraints are extracted) in constrained space only.

- Mix analysis: dudi.mix, dudi.hillsmith. Only quantitative variables are displayed.

- RDA (or PCAIV): pcaiv, pcaivortho, rda. With rda, space 1 is constrained space, space 2 is unconstrained space. Only constrained space is available with pcaiv, the opposite for pcaivortho. Set 1 is constraints (only quantitative constraints are extracted), set 2 is dependent variables (only set 2 is available for pcaivortho). If set=12 (default), fac is not available and pch,cex, col, lwd can be defined differently for each set.

- CCorA: CCorA, rcc. Space 1 is X, space 2 is Y. With rcc a third space is available, in which coordinates are means of X and Y coordinates. In this third space, set 1 is X, set 2 is Y. If set=12 (default), fac is not available and pch,cex, col, lwd can be defined differently for each set.

- rCCorA: rcc. Space 1 is X, space 2 is Y, space 3 is a "common" space in which coordinates are means of X and Y coordinates. In space 3, set 1 is X and set 2 is Y. If set=12 (default), fac is not available and pch,cex, col, lwd can be defined differently for each set.

- CIA: coinertia. Space 1 is X, space 2 is Y, space 3 is a "common" space where X and Y scores are normed. In space 3, set 1 is X and set 2 is Y. If set=12 in space 3 (default), fac is not available and pch,cex, col, lws can be defined differently for each set.

- GPA: GPA. Only the consensus ordination can be displayed.

- 2B-PLS: pls. Space 1 is X, space 2 is Y, space 3 is a "common" space in which coordinates are means of X and Y coordinates. In space 3, set 1 is X and set 2 is Y. If set=12 (default), fac is not available and pch,cex, col, lwd can be defined differently for each set.

- 2B-sPLS: pls. Space 1 is X, space 2 is Y, space 3 is a "common" space in which coordinates are means of X and Y coordinates. In space 3, set 1 is X and set 2 is Y. If set=12 (default), fac is not available and pch,cex, col, lwd can be defined differently for each set.

- rGCCA: wrapper.rgcca. Space can be 1 to n, the number of blocks (i.e. datasets).

- sGCCA: wrapper.sgcca. Space can be 1 to n, the number of blocks (i.e. datasets).

- DIABLO: block.plsda, block.splsda. Space can be 1 to n, the number of blocks (i.e. datasets).

Author(s)

Maxime HERVE <maxime.herve@univ-rennes1.fr>

Correlation circle of multivariate analyses

Description

Displays a correlation circle of a multivariate analysis.

Usage

MVA.corplot(x, xax = 1, yax = 2, thresh = 0, fac = NULL, set = c(12, 1, 2), space = 1,
  xlab = NULL, ylab = NULL, main = NULL, circle = TRUE, intcircle = 0.5, points = TRUE,
  ident = TRUE, arrows = TRUE, labels = NULL, main.pos = c("bottomleft", "topleft",
  "bottomright", "topright"), main.cex = 1.3, legend = FALSE, legend.pos = c("topleft",
  "topright", "bottomleft", "bottomright"), legend.title = NULL, legend.lab = NULL,
  pch = 16, cex = 1, col = 1, lwd = 1, drawintaxes = TRUE, add = FALSE, add.const = 1,
  keepmar = FALSE)

Arguments

Details

This function should not be use directly. Prefer the general MVA.plot, to which all arguments can be passed.

Many multivariate analyses are supported, from various packages:

- PCA: dudi.pca, rda.

- sPCA: spca.

- IPCA: ipca.

- sIPCA: sipca.

- LDA: lda, discrimin.

- PLS-DA (PLS2 on a dummy-coded factor): plsda. X space only.

- sPLS-DA (sPLS2 on a dummy-coded factor): splsda. X space only.

- CPPLS: mvr. Set 1 is X, set 2 is Y. If set=12 (default), fac is not available and pch,cex, col, lwd can be defined differently for each set. X space only.

- PLSR: mvr, pls, plsR (plsRglm package). Set 1 is X, set 2 is Y. If set=12 (default), fac is not available and pch,cex, col, lwd can be defined differently for each set. X space only.

- sPLSR: pls. Set 1 is X, set 2 is Y. If set=12 (default), fac is not available and pch,cex, col, lwd can be defined differently for each set. X space only.

- PLS-GLR: plsRglm (plsRglm package). Set 1 is X, set 2 is Y. If set=12 (default), fac is not available and pch,cex, col, lwd can be defined differently for each set. Correlations are computed with Y on the link scale.

- PCR: mvr. Set 1 is X, set 2 is Y. If set=12 (default), fac is not available and pch,cex, col, lwd can be defined differently for each set.

- CDA: discrimin, discrimin.coa.

- NSCOA: dudi.nsc. For NSCOA there is no real correlation, but the classical representation of columns is arrows. This is why MVA.corplot was made able to deal with this analysis.

- CCA: cca, pcaiv. Constraints (only quantitative constraints are extracted) in constrained space only.

- Mix analysis: dudi.mix, dudi.hillsmith. Only quantitative variables are displayed.

- RDA (or PCAIV): pcaiv, pcaivortho, rda. With rda, space 1 is constrained space, space 2 is unconstrained space. Only constrained space is available with pcaiv, the opposite for pcaivortho. Set 1 is constraints (only quantitative constraints are extracted), set 2 is dependent variables (only set 2 is available for pcaivortho). If set=12 (default), fac is not available and pch,cex, col, lwd can be defined differently for each set.

- db-RDA: capscale, dbrda. Constraints (only quantitative constraints are extracted) in constrained space only.

- CCorA: CCorA, rcc. Space 1 is X, space 2 is Y. With rcc a third space is available, in which coordinates are means of X and Y coordinates. In this third space, set 1 is X, set 2 is Y. If set=12 (default), fac is not available and pch,cex, col, lwd can be defined differently for each set.

- rCCorA: rcc. Space 1 is X, space 2 is Y, space 3 is a "common" space in which coordinates are means of X and Y coordinates. In space 3, set 1 is X and set 2 is Y. If set=12 (default), fac is not available and pch,cex, col, lwd can be defined differently for each set.

- CIA: coinertia. Space 1 is X, space 2 is Y, space 3 is a "common" space where X and Y scores are normed. In space 3, set 1 is X and set 2 is Y. If set=12 in space 3 (default), fac is not available and pch,cex, col, lws can be defined differently for each set.

- PCIA: procuste. Set 1 is X, set 2 is Y.

- 2B-PLS: pls. Space 1 is X, space 2 is Y, space 3 is a "common" space in which coordinates are means of X and Y coordinates. In space 3, set 1 is X and set 2 is Y. If set=12 (default), fac is not available and pch,cex, col, lwd can be defined differently for each set.

- 2B-sPLS: pls. Space 1 is X, space 2 is Y, space 3 is a "common" space in which coordinates are means of X and Y coordinates. In space 3, set 1 is X and set 2 is Y. If set=12 (default), fac is not available and pch,cex, col, lwd can be defined differently for each set.

- rGCCA: wrapper.rgcca. Space can be 1 to n, the number of blocks (i.e. datasets).

- sGCCA: wrapper.sgcca. Space can be 1 to n, the number of blocks (i.e. datasets).

- DIABLO: block.plsda, block.splsda. Space can be 1 to n, the number of blocks (i.e. datasets).

Author(s)

Maxime HERVE <maxime.herve@univ-rennes1.fr>

Examples

require(ade4)
data(olympic)
PCA <- dudi.pca(olympic$tab,scannf=FALSE)
MVA.plot(PCA,"corr")

Cross validation

Description

Performs cross validation with different PLS and/or discriminant analyses.

Usage

MVA.cv(X, Y, repet = 10, k = 7, ncomp = 8, scale = TRUE, model = c("PLSR",
  "CPPLS", "PLS-DA", "PPLS-DA", "LDA", "QDA", "PLS-DA/LDA", "PLS-DA/QDA",
  "PPLS-DA/LDA", "PPLS-DA/QDA"), lower = 0.5, upper = 0.5, Y.add = NULL,
  weights = rep(1, nrow(X)), set.prior = FALSE, crit.DA = c("plug-in",
  "predictive", "debiased"), ...)

Arguments

Details

When a discriminant analysis is used ("PLS-DA", "PPLS-DA", "LDA", "QDA", "PLS-DA/LDA", "PLS-DA/QDA", "PPLS-DA/LDA" or "PPLS-DA/QDA"), the training sets are generated in respect to the relative proportions of the levels of Y in the original data set (see splitf).

"PLS-DA" is considered as PLS2 on a dummy-coded response. For a PLS-DA based on the CPPLS algorithm, use "PPLS-DA" with lower and upper limits of the power parameters set to 0.5.

If scale = TRUE, the scaling is done as this: for each step of the validation loop (i.e. k steps), the training set is pre-processed by centering and unit-variance scaling. Means and standard deviations of variables in the training set are then used to scale the test set.

Value

Author(s)

Maxime HERVE <maxime.herve@univ-rennes1.fr>

See Also

predict.MVA.cmv, mvr, lda, qda

Examples

require(pls)
require(MASS)

# PLSR
data(yarn)
## Not run: MVA.cv(yarn$NIR,yarn$density,model="PLSR")

# PPLS-DA coupled to LDA
data(mayonnaise)
## Not run: MVA.cv(mayonnaise$NIR,factor(mayonnaise$oil.type),model="PPLS-DA/LDA")

Loadings of multivariate analyses

Description

Returns loadings of a multivariate analysis.

Usage

MVA.load(x, xax = 1, yax = 2, set = c(12, 1, 2), space = 1, ...)

Arguments

Details

Many multivariate analyses are supported, from various packages:

- PCA: prcomp, princomp, dudi.pca, rda, pca, pca.

- sPCA: spca.

- IPCA: ipca.

- sIPCA: sipca.

- LDA: lda, discrimin.

- PLS-DA (PLS2 on a dummy-coded factor): plsda. X space only.

- sPLS-DA (sPLS2 on a dummy-coded factor): splsda. X space only.

- CPPLS: mvr. X space only.

- PLSR: mvr, pls, plsR (plsRglm package). X space only.

- sPLSR: pls. X space only.

- PLS-GLR: plsRglm (plsRglm package).

- PCR: mvr.

- CDA: discrimin, discrimin.coa.

- NSCOA: dudi.nsc.

- MCA: dudi.acm.

- Mix analysis: dudi.mix, dudi.hillsmith.

- PCIA: procuste. Set 1 is X, set 2 is Y.

- RDA (or PCAIV): pcaiv, pcaivortho, rda. With rda, space 1 is constrained space, space 2 is unconstrained space. Only constrained space is available with pcaiv, the opposite for pcaivortho.

- CCorA: rcc. Space 1 is X, space 2 is Y.

- rCCorA: rcc. Space 1 is X, space 2 is Y.

- CIA: coinertia. Space 1 is X, space 2 is Y.

- 2B-PLS: pls. Space 1 is X, space 2 is Y.

- 2B-sPLS: pls. Space 1 is X, space 2 is Y.

- rGCCA: wrapper.rgcca. Space can be 1 to n, the number of blocks (i.e. datasets).

- sGCCA: wrapper.sgcca. Space can be 1 to n, the number of blocks (i.e. datasets).

- DIABLO: block.plsda, block.splsda. Space can be 1 to n, the number of blocks (i.e. datasets).

Author(s)

Maxime HERVE <maxime.herve@univ-rennes1.fr>

Loading plot of multivariate analyses

Description

Displays a loading plot of a multivariate analysis.

Usage

MVA.loadplot(x, xax = 1, yax = 2, fac = NULL, set = c(12, 1, 2), space = 1, map = TRUE,
  xlab = NULL, ylab = NULL, main = NULL, points = TRUE, ident = TRUE, links = TRUE, 
  line = TRUE, labels = NULL, main.pos = c("bottomleft", "topleft","bottomright",
  "topright"), main.cex = 1.3, legend = FALSE, legend.pos = c("topleft", "topright",
  "bottomleft", "bottomright"), legend.title = NULL, legend.lab = NULL, pch = 16,
  cex = 1, col = 1, lwd = 1, lty = 1, drawextaxes = TRUE, drawintaxes = TRUE, xlim = NULL,
  ylim = NULL)

Arguments

Details

This function should not be use directly. Prefer the general MVA.plot, to which all arguments can be passed.

Many multivariate analyses are supported, from various packages:

- PCA: prcomp, princomp, dudi.pca, rda, pca, pca.

- sPCA: spca.

- IPCA: ipca.

- sIPCA: sipca.

- LDA: lda, discrimin.

- PLS-DA (PLS2 on a dummy-coded factor): plsda. X space only.

- sPLS-DA (sPLS2 on a dummy-coded factor): splsda. X space only.

- CPPLS: mvr. X space only.

- PLSR: mvr, pls, plsR (plsRglm package). X space only.

- sPLSR: pls. X space only.

- PLS-GLR: plsRglm (plsRglm package).

- PCR: mvr.

- CDA: discrimin, discrimin.coa.

- NSCOA: dudi.nsc.

- MCA: dudi.acm.

- Mix analysis: dudi.mix, dudi.hillsmith.

- PCIA: procuste. Set 1 is X, set 2 is Y.

- RDA (or PCAIV): pcaiv, pcaivortho, rda. With rda, space 1 is constrained space, space 2 is unconstrained space. Only constrained space is available with pcaiv, the opposite for pcaivortho.

- CCorA: rcc. Space 1 is X, space 2 is Y.

- rCCorA: rcc. Space 1 is X, space 2 is Y.

- CIA: coinertia. Space 1 is X, space 2 is Y.

- 2B-PLS: pls. Space 1 is X, space 2 is Y.

- 2B-sPLS: pls. Space 1 is X, space 2 is Y.

- rGCCA: wrapper.rgcca. Space can be 1 to n, the number of blocks (i.e. datasets).

- sGCCA: wrapper.sgcca. Space can be 1 to n, the number of blocks (i.e. datasets).

- DIABLO: block.plsda, block.splsda. Space can be 1 to n, the number of blocks (i.e. datasets).

Author(s)

Maxime HERVE <maxime.herve@univ-rennes1.fr>

Examples

require(ade4)
data(olympic)
PCA <- dudi.pca(olympic$tab,scannf=FALSE)
MVA.plot(PCA,"load")

Paired plot of multivariate analyses

Description

Displays a paired plot (i.e. a score plot of paired points) of a multivariate analysis.

Usage

MVA.pairplot(x, xax = 1, yax = 2, pairs = NULL, scaling = 2, space = 1, fac = NULL,
  xlab = NULL, ylab = NULL, main = NULL, ident = TRUE, labels = NULL, cex = 0.7, col = 1,
  lwd = 1, main.pos = c("bottomleft", "topleft", "bottomright", "topright"),
  main.cex = 1.3, legend = FALSE, legend.pos = c("topleft", "topright", "bottomleft",
  "bottomright"), legend.title = NULL, legend.lab = NULL, drawextaxes = TRUE,
  drawintaxes = TRUE, xlim = NULL, ylim = NULL)

Arguments

Details

This function should not be use directly. Prefer the general MVA.plot, to which all arguments can be passed.

All multivariate analyses supported by MVA.scoreplot can be used for a paired plot.

Author(s)

Maxime HERVE <maxime.herve@univ-rennes1.fr>

Examples

require(ade4)
data(macaca)
PCIA <- procuste(macaca$xy1,macaca$xy2)
MVA.plot(PCIA,"pairs")

Plotting of multivariate analyses

Description

Displays several kinds of plots for multivariate analyses.

Usage

MVA.plot(x, type = c("scores", "loadings", "correlations", "biplot", "pairs",
  "trajectories"), ...)

Arguments

Details

Different subfunctions are used depending on the type of plot to be displayed: MVA.scoreplot, MVA.loadplot, MVA.corplot, MVA.biplot, MVA.pairplot or MVA.trajplot. These functions should not be used directly (everything can be done with the general MVA.plot) but for convenience, arguments and analyses supported are detailed in separate help pages.

Warning: the use of attach before running a multivariate analysis can prevent MVA.plot to get the values it needs, and make it fail.

Author(s)

Maxime HERVE <maxime.herve@univ-rennes1.fr>

Score plot of multivariate analyses

Description

Displays a score plot of a multivariate analysis.

Usage

MVA.scoreplot(x, xax = 1, yax = 2, scaling = 2, set = c(12, 1, 2), space = 1,
  byfac = TRUE, fac = NULL, barycenters = TRUE, stars = TRUE, contours = FALSE,
  dhist = TRUE, weights = 1, xlab = NULL, ylab = NULL, main = NULL, pch = 16,
  cex = 1, col = 1, points = TRUE, labels = NULL, main.pos = c("bottomleft",
  "topleft", "bottomright", "topright"), main.cex = 1.3, fac.lab = NULL,
  fac.cex = 1, legend = FALSE, legend.pos = c("topleft", "topright", "bottomleft",
  "bottomright"), legend.title = NULL, legend.lab = NULL, legend.cex = 1,
  drawextaxes = TRUE, drawintaxes = TRUE, xlim = NULL, ylim = NULL,
  keepmar = FALSE)

Arguments

Details

This function should not be use directly. Prefer the general MVA.plot, to which all arguments can be passed.

Many multivariate analyses are supported, from various packages:

- PCA: prcomp, princomp (if scores=TRUE), dudi.pca, rda, pca, pca. scaling can be defined for rda (see scores.rda).

- sPCA: spca.

- IPCA: ipca.

- sIPCA: sipca.

- PCoA: cmdscale (with at least on non-default argument), dudi.pco, wcmdscale (with at least one non-default argument), capscale, pco, pcoa.

- nMDS: monoMDS, metaMDS, nmds, isoMDS.

- LDA: lda, discrimin.

- PLS-DA (PLS2 on a dummy-coded factor): plsda. X space only.

- sPLS-DA (sPLS2 on a dummy-coded factor): splsda. X space only.

- CPPLS: mvr. X space only.

- PLSR: mvr, pls, plsR (plsRglm package). X space only.

- sPLSR: pls. X space only.

- PLS-GLR: plsRglm (plsRglm package).

- PCR: mvr.

- CDA: discrimin, discrimin.coa.

- NSCOA: dudi.nsc.

- MCA: dudi.acm.

- Mix analysis: dudi.mix, dudi.hillsmith.

- COA: dudi.coa, cca. Set 1 is rows, set 2 is columns. If set=12 (default), fac is not available and pch,cex, col can be defined differently for each set. scaling can be defined for cca (see scores.cca).

- DCOA: dudi.dec. Set 1 is rows, set 2 is columns. If set=12 (default), fac is not available and pch,cex, col can be defined differently for each set.

- PCIA: procuste. Set 1 is X, set 2 is Y. If set=12 (default), fac is not available and pch,cex, col can be defined differently for each set.

- Procrustean superimposition: procrustes. Set 1 is X, set 2 is Y. If set=12 (default), fac is not available and pch,cex, col can be defined differently for each set.

- GPA: GPA. Only the consensus ordination can be displayed.

- DPCoA: dpcoa. Set 1 is categories, set 2 is collections. If set=12 (default), fac is not available and pch,cex, col can be defined differently for each set.

- RDA (or PCAIV): pcaiv, pcaivortho, rda. With rda, space 1 is constrained space, space 2 is unconstrained space. Only constrained space is available with pcaiv, the opposite for pcaivortho. scaling can be defined for rda (see scores.rda).

- db-RDA (or CAP): capscale, dbrda. Space 1 is constrained space, space 2 is unconstrained space.

- CCA: pcaiv, cca. With rda, space 1 is constrained space, space 2 is unconstrained space. Only constrained space is available with pcaiv. Set 1 is rows, set 2 is columns. scaling can be defined for cca (see scores.cca).

- CCorA: CCorA, rcc. Space 1 is X, space 2 is Y. With rcc a third space is available, in which coordinates are means of X and Y coordinates.

- rCCorA: rcc. Space 1 is X, space 2 is Y, space 3 is a "common" space in which coordinates are means of X and Y coordinates.

- CIA: coinertia. Space 1 is X, space 2 is Y, space 3 is a "common" space where X and Y scores are normed. In space 3, set 1 is X and set 2 is Y. If set=12 in space 3 (default), fac is not available and pch,cex, col can be defined differently for each set.

- 2B-PLS: pls. Space 1 is X, space 2 is Y, space 3 is a "common" space in which coordinates are means of X and Y coordinates.

- 2B-sPLS: pls. Space 1 is X, space 2 is Y, space 3 is a "common" space in which coordinates are means of X and Y coordinates.

- rGCCA: rgcca, wrapper.rgcca. Space can be 1 to n, the number of blocks (i.e. datasets).

- sGCCA: rgcca, wrapper.sgcca. Space can be 1 to n, the number of blocks (i.e. datasets).

- DIABLO: block.plsda, block.splsda. Space can be 1 to n, the number of blocks (i.e. datasets).

Author(s)

Maxime HERVE <maxime.herve@univ-rennes1.fr>

Examples

data(iris)
PCA <- prcomp(iris[,1:4])
MVA.plot(PCA,"scores")
MVA.plot(PCA,"scores",fac=iris$Species,col=1:3,pch=15:17)

Scores of multivariate analyses

Description

Returns scores of a multivariate analysis.

Usage

MVA.scores(x, xax = 1, yax = 2, scaling = 2, set = c(12, 1, 2), space = 1, ...)

Arguments

Details

Many multivariate analyses are supported, from various packages:

- PCA: prcomp, princomp (if scores=TRUE), dudi.pca, rda, pca, pca. scaling can be defined for rda (see scores.rda).

- sPCA: spca.

- IPCA: ipca.

- sIPCA: sipca.

- PCoA: cmdscale (with at least on non-default argument), dudi.pco, wcmdscale (with at least one non-default argument), capscale, pco, pcoa.

- nMDS: monoMDS, metaMDS, nmds, isoMDS.

- LDA: lda, discrimin.

- PLS-DA (PLS2 on a dummy-coded factor): plsda. X space only.

- sPLS-DA (sPLS2 on a dummy-coded factor): splsda. X space only.

- CPPLS: mvr. X space only.

- PLSR: mvr, pls, plsR (plsRglm package). X space only.

- sPLSR: pls. X space only.

- PLS-GLR: plsRglm (plsRglm package).

- PCR: mvr.

- CDA: discrimin, discrimin.coa.

- NSCOA: dudi.nsc.

- MCA: dudi.acm.

- Mix analysis: dudi.mix, dudi.hillsmith.

- COA: dudi.coa, cca. Set 1 is rows, set 2 is columns. If set=12 (default), fac is not available and pch,cex, col can be defined differently for each set. scaling can be defined for cca (see scores.cca).

- DCOA: dudi.dec. Set 1 is rows, set 2 is columns. If set=12 (default), fac is not available and pch,cex, col can be defined differently for each set.

- PCIA: procuste. Set 1 is X, set 2 is Y. If set=12 (default), fac is not available and pch,cex, col can be defined differently for each set.

- Procrustean superimposition: procrustes. Set 1 is X, set 2 is Y. If set=12 (default), fac is not available and pch,cex, col can be defined differently for each set.

- GPA: GPA. Only the consensus ordination can be displayed.

- DPCoA: dpcoa. Set 1 is categories, set 2 is collections. If set=12 (default), fac is not available and pch,cex, col can be defined differently for each set.

- RDA (or PCAIV): pcaiv, pcaivortho, rda. With rda, space 1 is constrained space, space 2 is unconstrained space. Only constrained space is available with pcaiv, the opposite for pcaivortho. scaling can be defined for rda (see scores.rda).

- db-RDA (or CAP): capscale, dbrda. Space 1 is constrained space, space 2 is unconstrained space.

- CCA: pcaiv, cca. With rda, space 1 is constrained space, space 2 is unconstrained space. Only constrained space is available with pcaiv. Set 1 is rows, set 2 is columns. scaling can be defined for cca (see scores.cca).

- CCorA: CCorA, rcc. Space 1 is X, space 2 is Y. With rcc a third space is available, in which coordinates are means of X and Y coordinates.

- rCCorA: rcc. Space 1 is X, space 2 is Y, space 3 is a "common" space in which coordinates are means of X and Y coordinates.

- CIA: coinertia. Space 1 is X, space 2 is Y, space 3 is a "common" space where X and Y scores are normed. In space 3, set 1 is X and set 2 is Y. If set=12 in space 3 (default), fac is not available and pch,cex, col can be defined differently for each set.

- 2B-PLS: pls. Space 1 is X, space 2 is Y, space 3 is a "common" space in which coordinates are means of X and Y coordinates.

- 2B-sPLS: pls. Space 1 is X, space 2 is Y, space 3 is a "common" space in which coordinates are means of X and Y coordinates.

- rGCCA: rgcca, wrapper.rgcca. Space can be 1 to n, the number of blocks (i.e. datasets).

- sGCCA: rgcca, wrapper.sgcca. Space can be 1 to n, the number of blocks (i.e. datasets).

- DIABLO: block.plsda, block.splsda. Space can be 1 to n, the number of blocks (i.e. datasets).

Author(s)

Maxime HERVE <maxime.herve@univ-rennes1.fr>

Synthesis quality of multivariate analyses

Description

Gives a simple estimator of the quality of the (descriptive) synthesis performed by a wide range of multivariate analyses.

Usage

MVA.synt(x, rows = 5)

Arguments

Details

Many multivariate analyses are supported, from various packages.

- PCA: prcomp, princomp, dudi.pca, rda, pca, pca: % of total variance explained by each axis.

- sPCA: spca: % of total variance explained by each axis.

- IPCA: ipca: kurtosis of each axis.

- sIPCA: sipca: kurtosis of each axis.

- PCoA: cmdscale (with eig=TRUE), dudi.pco, wcmdscale (with eig=TRUE), capscale, pco, pcoa: % of total variance explained by each axis.

- nMDS: monoMDS, metaMDS, nmds, isoMDS: stress.

- RDA: pcaiv, pcaivortho, rda: % of constrained and unconstrained total variance, % of constrained variance explained by constrained axes (pcaiv and rda), % of unconstrained variance explained by unconstrained axes (pcaivortho and rda).

- db-RDA (or CAP): capscale, dbrda: % of constrained and unconstrained total variance, % of constrained variance explained by constrained axes, % of unconstrained variance explained by unconstrained axes.

- COA: dudi.coa, cca: % of total inertia explained by each axis.

- CCA: pcaiv, cca: % of constrained and unconstrained total inertia, % of constrained inertia explained by constrained axes, % of unconstrained inertia explained by unconstrained axes (cca only).

- CPPLS: mvr: % of X and Y variances explained by each axis.

- PLSR: mvr, plsR (plsRglm package): % of X and Y variances explained by each axis (only Y for the moment with plsR).

- 2B-PLS: pls: % of X/Y square covariance explained by each pair of axes, correlation between each pair of axes (canonical correlations).

- CCorA: CCorA, rcc: correlation between each pair of axes (canonical correlations).

- rCCorA: rcc: correlation between each pair of axes (canonical correlations).

- PCR: mvr: % of X and Y variances explained by each axis.

- MCA: dudi.acm: % of total inertia explained by each axis.

- Mix analysis: dudi.mix, dudi.hillsmith: % of total inertia explained by each axis.

- GPA: GPA: % of consensus and residual variance, % of total variance exlained by each axis, % of consensus variance explained by each axis, % of residual variance coming from each group of variables.

- RGCCA: rgcca, wrapper.rgcca: % of total intra-block variance explained by each axis, correlation between each pair of axes (canonical correlations).

- DIABLO: block.plsda, block.splsda: % of total intra-block variance explained by each axis, correlation between each pair of axes (canonical correlations).

- CIA: coinertia: RV coefficient, % of co-inertia explained by each pair of axes, correlation between each pair of axes (canonical correlations).

- PCIA: procuste: m2.

Author(s)

Maxime HERVE <maxime.herve@univ-rennes1.fr>

Examples

data(iris)
PCA <- prcomp(iris[,1:4])
MVA.synt(PCA)

Significance test based on cross (model) validation

Description

Performs a permutation significance test based on cross (model) validation with different PLS and/or discriminant analyses. See MVA.cv and MVA.cmv for more details about how cross (model) validation is performed.

Usage

MVA.test(X, Y, cmv = FALSE, ncomp = 8, kout = 7, kinn = 6, scale = TRUE,
  model = c("PLSR", "CPPLS", "PLS-DA", "PPLS-DA", "LDA", "QDA", "PLS-DA/LDA",
  "PLS-DA/QDA", "PPLS-DA/LDA","PPLS-DA/QDA"), Q2diff = 0.05, lower = 0.5,
  upper = 0.5, Y.add = NULL, weights = rep(1, nrow(X)), set.prior = FALSE,
  crit.DA = c("plug-in", "predictive", "debiased"), p.method = "fdr",
  nperm = 999, progress = TRUE, ...)

Arguments

Details

When Y consists in quantitative response(s), the null hypothesis is that each response is not predicted better than what would happen by chance. In this case, Q2 is used as the test statistic. When Y contains several responses, a p-value is computed for each response and p-values are corrected for multiple testing.

When Y is a factor, the null hypothesis is that the factor has no discriminant ability. In this case, the classification error rate (NMC) is used as the test statistic.

Whatever the response, the reference value of the test statistics is obtained by averaging 20 values coming from independently performed cross (model) validation on the original data.

The function deals with the limitted floating point precision, which can bias calculation of p-values based on a discrete test statistic distribution.

Value

Author(s)

Maxime HERVE <maxime.herve@univ-rennes1.fr>

References

Westerhuis J, Hoefsloot HCJ, Smit S, Vis DJ, Smilde AK, van Velzen EJJ, van Duijnhoven JPM and van Dorsten FA (2008) Assessment of PLSDA cross validation. Metabolomics 4:81-89.

See Also

MVA.cv, MVA.cmv

Examples

require(pls)
require(MASS)

# PLSR
data(yarn)
## Not run: MVA.test(yarn$NIR,yarn$density,cmv=TRUE,model="PLSR")

# PPLS-DA coupled to LDA
data(mayonnaise)
## Not run: MVA.test(mayonnaise$NIR,factor(mayonnaise$oil.type),model="PPLS-DA/LDA")

Trajectory plot of multivariate analyses

Description

Displays a trajectory plot (i.e. a score plot with trajectories linking defined points) of a multivariate analysis.

Usage

MVA.trajplot(x, xax = 1, yax = 2, trajects, trajlab = NULL, scaling = 2,
  set = c(12, 1, 2), space = 1, xlab = NULL, ylab = NULL, main = NULL,
  pch = 16, cex = 1, trajlab.cex = 1, col = 1, lwd = 1, lty = 1,
  points = TRUE, allpoints = TRUE, arrows = TRUE, labels = NULL,
  main.pos = c("bottomleft", "topleft", "bottomright", "topright"),
  main.cex = 1.3, legend = FALSE, legend.pos = c("topleft", "topright",
  "bottomleft", "bottomright"), legend.title = NULL, legend.lab = NULL,
  legend.cex = 1, drawextaxes = TRUE, drawintaxes = TRUE, xlim = NULL,
  ylim = NULL)

Arguments

Details

This function should not be use directly. Prefer the general MVA.plot, to which all arguments can be passed.

All multivariate analyses supported by MVA.scoreplot can be used for a paired plot.

Author(s)

Maxime HERVE <maxime.herve@univ-rennes1.fr>

Examples

require(ade4)
data(olympic)
PCA <- dudi.pca(olympic$tab,scannf=FALSE)
MVA.plot(PCA,"traject",trajects=list(1:10,25:30),col=c(2,3,1),trajlab=c("T1","T2"))

Odds-ratio (multinomial regression)

Description

Computes the odds ratios and their confidence interval for a predictor of a model fitted with multinom.

Usage

OR.multinom(model, variable, conf.level = 0.95)

Arguments

Author(s)

Maxime HERVE <maxime.herve@univ-rennes1.fr>

Variable Importance in the Projection (VIP)

Description

Returns VIP score of each X-variable in a PLS-DA (obtained from plsda).

Usage

PLSDA.VIP(model, graph = FALSE)

Arguments

Value

Author(s)

Maxime HERVE <maxime.herve@univ-rennes1.fr>

See Also

plsda

Examples

if (require(mixOmics)) {
  data(yeast)
  model.PLSDA <- plsda(t(yeast$data),yeast$strain.cond)
  PLSDA.VIP(model.PLSDA)
}

Type II permutation MANOVA using distance matrices

Description

This function is a wrapper to adonis2 but performs type II tests (whereas adonis2 performs type I).

Usage

adonis.II(formula, data, ...)

Arguments

Details

See adonis2 for detailed explanation of what is done. The only difference with adonis2 is that adonis.II performs type II tests instead of type I.

Value

a data frame of class "anova".

Author(s)

Maxime HERVE <maxime.herve@univ-rennes1.fr>

Examples

require(vegan)
data(dune)
data(dune.env)

# Compare:
adonis2(dune~Management*A1,by="terms",data=dune.env,permutations=99)
adonis2(dune~A1*Management,by="terms",data=dune.env,permutations=99)

# With:
adonis.II(dune~Management*A1,data=dune.env,permutations=99)
adonis.II(dune~A1*Management,data=dune.env,permutations=99)

Back-transformation of EMMeans

Description

Back-transforms EMMeans (produced by emmeans) when the model was built on a transformed response variable. This is typically the case when a LM(M) with log(x+1) as response variable gives a better fitting than a GLM(M) for count data, or when a beta regression takes as response a variable on the [0;1] interval that has been rescaled to the (0;1) interval using p.beta.

Usage

back.emmeans(emm, transform = c("log", "logit", "sqrt", "4rt", "inverse", "p.beta"),
  base = exp(1), add = 0, n = NULL, C = 2, ord = FALSE, decreasing = TRUE)

Arguments

Author(s)

Maxime HERVE <maxime.herve@univ-rennes1.fr>

See Also

emmeans

Examples

require(emmeans)

set.seed(1149)
tab <- data.frame(
  response <- c(rpois(30,0),rpois(30,2),rpois(30,4)),
  fact <- gl(3,30,labels=LETTERS[1:3])
)

model <- lm(log(response+1)~fact,data=tab)
EMM <- emmeans(model,~fact)
back.emmeans(EMM,transform="log",add=1)

Bootstrap

Description

Simplified version of the boot function.

Usage

bootstrap(x, fun, nrep = 1000, conf.level = 0.95, ...)

Arguments

Details

See help of the boot function for more explanations.

Value

Author(s)

Maxime HERVE <maxime.herve@univ-rennes1.fr>

See Also

boot

Examples

# Confidence interval of a mean
samp <- sample(1:50,10,replace=TRUE)
bootstrap(samp,function(x,i) mean(x[i]))

# Confidence interval of the standard error of the mean
bootstrap(samp,function(x,i) sd(x[i])/sqrt(length(x[i])))

Histogram for factor levels

Description

Draws a histogram of a numeric variable per level of a factor.

Usage

byf.hist(formula, data, sep = FALSE, density = TRUE, xlab = NULL, ylab = NULL,
  col = NULL)

Arguments

Author(s)

Maxime HERVE <maxime.herve@univ-rennes1.fr>

See Also

hist

Examples

data(iris)
byf.hist(Sepal.Length~Species,data=iris)

QQ-plot for factor levels

Description

Draws a multivariate QQ-plot of numeric variables per level of a factor.

Usage

byf.mqqnorm(formula, data)

Arguments

Author(s)

Maxime HERVE <maxime.herve@univ-rennes1.fr>

See Also

mqqnorm, byf.mshapiro, qqPlot

Examples

data(iris)
byf.mqqnorm(as.matrix(iris[,1:4])~Species,data=iris)

Shapiro-Wilk test for factor levels

Description

Performs a multivariate Shapiro-Wilk test on numeric variables per level of a factor.

Usage

byf.mshapiro(formula, data)

Arguments

Value

Author(s)

Maxime HERVE <maxime.herve@univ-rennes1.fr>

See Also

byf.mqqnorm, mshapiro.test, qqPlot

Examples

data(iris)
byf.mshapiro(as.matrix(iris[,1:4])~Species,data=iris)

QQ-plot for factor levels

Description

Draws a QQ-plot of a numeric variable per level of a factor.

Usage

byf.qqnorm(formula, data, ...)

Arguments

Author(s)

Maxime HERVE <maxime.herve@univ-rennes1.fr>

See Also

link[RVAideMemoire]{byf.shapiro}, qqPlot

Examples

data(iris)
byf.qqnorm(Sepal.Length~Species,data=iris)

Shapiro-Wilk test for factor levels

Description

Performs a Shapiro-Wilk test on a numeric variable per level of a factor.

Usage

byf.shapiro(formula, data)

Arguments

Value

Author(s)

Maxime HERVE <maxime.herve@univ-rennes1.fr>

See Also

byf.qqnorm, shapiro.test

Examples

data(iris)
byf.shapiro(Sepal.Length~Species,data=iris)

Cumulative Distribution Function of a known discrete distribution

Description

Returns an object similar to what is produced by ecdf, but based on a known discrete distribution.

Usage

cdf.discrete(x, dist = c("binom", "geom", "hyper", "nbinom", "pois"), ...)

Arguments

Details

The function is intended to be used in goodness-of-fits tests for discrete distributions, such as proposed in the dgof package.

Author(s)

Maxime HERVE <maxime.herve@univ-rennes1.fr>

Examples

if(require(dgof)) {
  set.seed(1124)
  resp <- rpois(20,2)
  cvm.test(resp,cdf.discrete(resp,"pois",2))
}

Expected counts for comparison of response probabilities to given values

Description

Returns expected counts before comparing response probabilities (i.e. when the response variable is a binary variable) to given values by a chi-squared test. The function is in fact a wrapper to the chi-squared test for comparison of proportions to given values on a contingency table.

Usage

chisq.bin.exp(formula, data, p, graph = FALSE)

Arguments

Details

The function returns how many counts can be < 5 to respect Cochran's rule (80% of counts must be >= 5).

Value

Author(s)

Maxime HERVE <maxime.herve@univ-rennes1.fr>

See Also

prop.test, chisq.theo.bintest, mosaicplot

Examples

response <- c(rep(0:1,c(40,60)),rep(0:1,c(55,45)),rep(0:1,c(65,35)))
fact <- gl(3,100,labels=LETTERS[1:3])
p.theo <- c(0.5,0.45,0.2)
chisq.bin.exp(response~fact,p=p.theo)

Pearson's Chi-squared test for binary variables

Description

Performs a Pearson's Chi-squared test for comparing response probabilities (i.e. when the response variable is a binary variable). The function is in fact a wrapper to the chi-squared test for comparison of proportions on a contingency table. If the p-value of the test is significant, the function performs pairwise comparisons by using Pearson's Chi-squared tests.

Usage

chisq.bintest(formula, data, correct = TRUE, alpha = 0.05, p.method = "fdr")

Arguments

Details

If the response is a 0/1 variable, the probability of the '1' group is tested. In any other cases, the response is transformed into a factor and the probability of the second level is tested.

Since a chi-squared test is an approximate test, an exact test is preferable when the number of individuals is small (200 is a reasonable minimum). See fisher.bintest in that case.

Value

Author(s)

Maxime HERVE <maxime.herve@univ-rennes1.fr>

See Also

G.bintest, fisher.bintest

Examples

response <- c(rep(0:1,c(40,60)),rep(0:1,c(55,45)),rep(0:1,c(65,35)))
fact <- gl(3,100,labels=LETTERS[1:3])
chisq.bintest(response~fact)

Expected counts for comparison of proportions to given values

Description

Returns expected counts before comparing proportions to given values by a chi-squared test.

Usage

chisq.exp(data, p, graph = FALSE)

Arguments

Details

The function returns how many counts can be < 5 to respect Cochran's rule (80% of counts must be >= 5).

Value

Author(s)

Maxime HERVE <maxime.herve@univ-rennes1.fr>

See Also

prop.test, chisq.test, mosaicplot

Examples

proportions <- sample(c(0,1),200,replace=TRUE)
populations <- sample(LETTERS[1:3],200,replace=TRUE)
tab.cont <- table(populations,proportions)
p.theo <- c(0.4,0.5,0.7)
chisq.exp(tab.cont,p=p.theo)

Pairwise comparisons after a chi-squared goodness-of-fit test

Description

Performs pairwise comparisons after a global chi-squared goodness-of-fit test.

Usage

chisq.multcomp(x, p.method = "fdr")

Arguments

Details

Since a chi-squared test is an approximate test, an exact test is preferable when the number of individuals is small (200 is a reasonable minimum). See multinomial.multcomp in that case.

Value

Author(s)

Maxime HERVE <maxime.herve@univ-rennes1.fr>

See Also

chisq.test, multinomial.test, multinomial.multcomp

Examples

counts <- c(49,30,63,59)
chisq.test(counts)
chisq.multcomp(counts)

Pearson's Chi-squared test for comparison of response probabilities to given values

Description

Performs a Pearson's Chi-squared test for comparing response probabilities (i.e. when the response variable is a binary variable) to given values. The function is in fact a wrapper to the chi-squared test for comparison of proportions to given values on a contingency table.

Usage

chisq.theo.bintest(formula, data, p)

Arguments

Details

If the response is a 0/1 variable, the probability of the '1' group is tested. In any other cases, the response is transformed into a factor and the probability of the second level is tested.

Value

Author(s)

Maxime HERVE <maxime.herve@univ-rennes1.fr>

See Also

prop.test, chisq.bin.exp, prop.bin.multcomp

Examples

response <- c(rep(0:1,c(40,60)),rep(0:1,c(55,45)),rep(0:1,c(65,35)))
fact <- gl(3,100,labels=LETTERS[1:3])
p.theo <- c(0.5,0.45,0.2)
chisq.theo.bintest(response~fact,p=p.theo)

Pairwise comparisons after a chi-squared test for given probabilities

Description

Performs pairwise comparisons after a global chi-squared test for given probabilities.

Usage

chisq.theo.multcomp(x, p = rep(1/length(x), length(x)), p.method = "fdr")

Arguments

Details

Since a chi-squared test is an approximate test, an exact test is preferable when the number of individuals is small (200 is a reasonable minimum). See multinomial.theo.multcomp in that case.

Value

Author(s)

Maxime HERVE <maxime.herve@univ-rennes1.fr>

See Also

chisq.test, multinomial.test, multinomial.theo.multcomp

Examples

counts <- c(49,30,63,59)
p.theo <- c(0.2,0.1,0.45,0.25)
chisq.test(counts,p=p.theo)
chisq.theo.multcomp(counts,p=p.theo)

Cochran's Q test

Description

Performs the Cochran's Q test for unreplicated randomized block design experiments with a binary response variable and paired data. If the p-value of the test is significant, the function performs pairwise comparisons by using the Wilcoxon sign test.

Usage

cochran.qtest(formula, data, alpha = 0.05, p.method = "fdr")

Arguments

Details

If the response is a 0/1 variable, the probability of the '1' group is tested. In any other cases, the response is transformed into a factor and the probability of the second level is tested.

Value

Author(s)

Maxime HERVE <maxime.herve@univ-rennes1.fr>

Examples

response <- c(0,1,1,0,0,1,0,1,1,1,1,1,0,0,1,1,0,1,0,1,1,0,0,1,0,1,1,0,0,1)
fact <- gl(3,1,30,labels=LETTERS[1:3])
block <- gl(10,3,labels=letters[1:10])
cochran.qtest(response~fact|block)

Condition number of the Hessian matrix of a multinomial log-linear model

Description

Computes the condition number of the Hessian matrix of a model fitted with multinom.

Usage

cond.multinom(model)

Arguments

Author(s)

Maxime HERVE <maxime.herve@univ-rennes1.fr>

Coordinates of projected points

Description

Returns the coordinates of a set of points when orthogonally projected on a new axis.

Usage

coord.proj(coord,slp)

Arguments

Author(s)

Maxime HERVE <maxime.herve@univ-rennes1.fr>

Examples

data(iris)

# Original coordinates
plot(Petal.Length~Sepal.Length,pch=16,col=as.numeric(iris$Species),data=iris)

# New axis
abline(-6,1.6)

# Coordinates on new axis
new.coord <- coord.proj(iris[,c("Sepal.Length","Petal.Length")],1.6)
stripchart(new.coord~Species,data=iris,col=1:3)

Comparison of 2 Pearson's linear correlation coefficients

Description

Performs the test for equality of 2 Pearson's correlation coefficients. If the difference is not significant, the function returns the common coefficient, its confidence interval and performs the test for equality to a given value.

Usage

cor.2comp(var1, var2, var3, var4, alpha = 0.05, conf.level = 0.95, theo = 0)

Arguments

Value

Author(s)

Maxime HERVE <maxime.herve@univ-rennes1.fr>

See Also

cor.test

Examples

cor1.var1 <- 1:30+rnorm(30,0,2)
cor1.var2 <- 1:30+rnorm(30,0,3)
cor2.var1 <- (-1):-30+rnorm(30,0,2)
cor2.var2 <- (-1):-30+rnorm(30,0,3)
cor.2comp(cor1.var1,cor1.var2,cor2.var1,cor2.var2)

Equality of a Pearson's linear correlation coefficient to a given value

Description

Performs a test for equality of a Pearson's linear correlation coefficient to a given value.

Usage

cor.conf(var1, var2, theo)

Arguments

Value

Author(s)

Maxime HERVE <maxime.herve@univ-rennes1.fr>

See Also

cor.test

Examples

var1 <- 1:30+rnorm(30,0,4)
var2 <- 1:30+rnorm(30,0,4)
cor.conf(var1,var2,theo=0.5)

Comparison of several Pearson's linear correlation coefficients

Description

Performs comparisons of several Pearson's linear correlation coefficients. If no difference, the function returns the common correlation coefficient, its confidence interval and test for its equality to a given value. If the difference is significant, the functions performs pairwise comparisons between coefficients.

Usage

cor.multcomp(var1, var2, fact, alpha = 0.05, conf.level = 0.95, theo = 0,
  p.method = "fdr")

Arguments

Value

Author(s)

Maxime HERVE <maxime.herve@univ-rennes1.fr>

See Also

cor.test

Examples

var1 <- c(1:15+rnorm(15,0,4),1:15+rnorm(15,0,1),1:15+rnorm(15,0,8))
var2 <- c(-1:-15+rnorm(15,0,4),1:15+rnorm(15,0,1),1:15+rnorm(15,0,8))
fact <- gl(3,15,labels=LETTERS[1:3])
cor.multcomp(var1,var2,fact)

var3 <- c(1:15+rnorm(15,0,1),1:15+rnorm(15,0,3),1:15+rnorm(15,0,2))
cor.multcomp(var1,var3,fact)

Significance test for the covariance between two datasets

Description

Performs a permutation test based on the sum of square covariance between variables of two datasets, to test wether the (square) covariance is higher than expected under random association between the two datasets. The test is relevent parallel to a 2B-PLS.

Usage

cov.test(X, Y, scale.X = TRUE, scale.Y = TRUE, nperm = 999, progress = TRUE)

Arguments

Details

The function deals with the limitted floating point precision, which can bias calculation of p-values based on a discrete test statistic distribution.

Value

Author(s)

Maxime HERVE <maxime.herve@univ-rennes1.fr>

Martingale residuals of a Cox model

Description

Plots martingale residuals of a Cox model against fitted values, to check for log-linearity of covariates.

Usage

cox.resid(model)

Arguments

Author(s)

Maxime HERVE <maxime.herve@univ-rennes1.fr>, based on an idea of John Fox.

References

Fox, J. 2002 Cox Proportional-Hazards Regression for Survival Data.

See Also

coxph

Examples

# 'kidney' dataset of package 'survival'
require(survival)
data(kidney)
model <- coxph(Surv(time,status)~age+factor(sex),data=kidney)
cox.resid(model)

Cramer's association coefficient

Description

Computes the Cramer's association coefficient between 2 nominal variables.

Usage

cramer(x, y)

Arguments

Author(s)

Maxime HERVE <maxime.herve@univ-rennes1.fr>

Examples

var1 <- sample(LETTERS[1:3],30,replace=TRUE)
var2 <- sample(letters[1:3],30,replace=TRUE)
cramer(var1,var2)
# or cramer(table(var1,var2))

Cramer's association coefficient

Description

Computes the Cramer's association coefficient between 2 nominal variables, its confidence interval (by bootstraping) and tests for its significance.

Usage

cramer.test(x, y, nrep = 1000, conf.level = 0.95)

Arguments

Value

Author(s)

Maxime HERVE <maxime.herve@univ-rennes1.fr>

See Also

boot

Examples

var1 <- sample(LETTERS[1:3],30,replace=TRUE)
var2 <- sample(letters[1:3],30,replace=TRUE)
cramer.test(var1,var2)
# or cramer.test(table(var1,var2))

Coefficient of variation

Description

Computes the coefficient of variation of a vector.

Usage

cv(x, abs = TRUE, pc = TRUE)

Arguments

Details

The function deals with missing values.

Author(s)

Maxime HERVE <maxime.herve@univ-rennes1.fr>

Examples

cv(rnorm(30))

Dendrogram and number of groups to be chosen

Description

Draws a dendrogram and an additional bar plot helping to choose the number of groups to be retained (based on the dendrogram).

Usage

dendro.gp(dend)

Arguments

Author(s)

Maxime HERVE <maxime.herve@univ-rennes1.fr>

See Also

hclust

Examples

data(iris)
distances <- dist(iris[,1:4],method="euclidian")
dendro <- hclust(distances,method="ward.D2")
dendro.gp(dendro)

Deprecated functions in RVAideMemoire package

Description

Functions that are not usable anymore, and will be entirely removed from the package in future versions.

Usage

back.lsmeans(...)
byf.normhist(...)
cor.sparse(...)
CvM.test(...)
DA.confusion(...)
DA.valid(...)
DA.var(...)
dunn.test(...)
fc.multcomp(...)
friedman.rating.test(...)
kruskal.rating.test(...)
pairwise.manova(...)
pairwise.to.groups(...)
pairwise.wilcox.rating.test(...)
plot1comp.ind(...)
plot1comp.var(...)
PLSDA.ncomp(...)
PLSDA.test(...)
rating.lsmeans(...)
s.corcircle2(...)
scat.mix.categorical(...)
scat.mix.numeric(...)
scatter.coa2(...)
wilcox.paired.rating.multcomp(...)
wilcox.rating.signtest(...)
wilcox.rating.test(...)

Arguments

Details

back.lsmeans and rating.lsmeans are replaced by back.emmeans and rating.emmeans. More generally, stop using package lsmeans and change to package emmeans, its new version.

byf.normhist was not very useful and byf.hist does nearly the same job.

cor.sparse is replaced by the more generic MVA.plot.

CvM.test did actually not perform a Cramer-von Mises test but an alternative Cramer test. Use cramer.test from package cramer directly, on which CvM.test was based.

DA.confusion and DA.valid are replaced by the more generic MVA.cmv and MVA.cv.

DA.var is replaced by the more generic MVA.synt.

dunn.test is not very useful, prefer dunnTest from package FSA.

fc.multcomp is not useful anymore since emtrends (package emmeans) does the same job in a much more powerful manner (see argument var of emtrends).

friedman.rating.test, kruskal.rating.test, wilcox.rating.test, wilcox.rating.signtest, pairwise.wilcox.rating.test and wilcox.paired.rating.multcomp can be problematic with ratings (in which ties and zeroes are very frequent). The use of CLM(M)s (via clm and clmm) is recommended.

pairwise.manova is not useful anymore since emmeans (package emmeans) does the same job in a much more powerful manner (on "mlm" objects, created by lm and not manova)

pairwise.to.groups was not very useful.

plot1comp.ind, plot1comp.var, s.corcircle2, scat.mix.categorical, scat.mix.numeric and scatter.coa2 are replaced by the more generic MVA.plot.

PLSDA.ncomp was not really useful and mvr does nearly the same job.

PLSDA.test is replaced by the more generic MVA.test.

Dummy responses

Description

Creates a matrix of dummy responses from a factor. Needed in some multivariate analyses.

Usage

dummy(f, simplify = TRUE)

Arguments

Author(s)

Maxime HERVE <maxime.herve@univ-rennes1.fr>

Examples

fac <- gl(3,5,labels=LETTERS[1:3])
dummy(fac)

Empirical logistic transformation

Description

Empirical logistic transformation for logistic models with data showing (quasi-)complete separation. The function is intended to be used as a link function in GLM(M)s.

Usage

elogis()

Author(s)

Formula from McCullagh & Nelder in their seminal book 'Generalized Linear Models'. R code from Eric Wajnberg & Jean-Sebastien Pierre.

Examples

# An example with 3 groups and complete separation (from E. Wajnberg)
tab <- data.frame(case=letters[1:3],yes=c(25,30,0),no=c(1,0,20))
tab
## Not run: 
mod <- glm(cbind(yes,no)~case,family=binomial(link=elogis()),data=tab)
# Warnings are normal
summary(mod)
## End(Not run)

Fisher's exact test for binary variables

Description

Performs a Fisher's exact test for comparing response probabilities (i.e. when the response variable is a binary variable). The function is in fact a wrapper to the Fisher's exact test for count data. If the p-value of the test is significant, the function performs pairwise comparisons by using Fisher's exact tests.

Usage

fisher.bintest(formula, data, alpha = 0.05, p.method = "fdr")

Arguments

Details

If the response is a 0/1 variable, the probability of the '1' group is tested. In any other cases, the response is transformed into a factor and the probability of the second level is tested.

Since chi-squared and G tests are approximate tests, exact tests are preferable when the number of individuals is small (200 is a reasonable minimum).

Value

Author(s)

Maxime HERVE <maxime.herve@univ-rennes1.fr>

See Also

chisq.bintest, G.bintest

Examples

response <- c(0,0,0,0,0,0,1,0,0,0,0,0,1,0,1,1,1,0,0,1,1,1,1,1,1,0,0,1,1,1)
fact <- gl(3,10,labels=LETTERS[1:3])
fisher.bintest(response~fact)

Pairwise comparisons using Fisher's exact test

Description

Performs pairwise comparisons after a comparison of proportions or after a test for independence of 2 categorical variables, by using a Fisher's exact test.

Usage

fisher.multcomp(tab.cont, p.method = "fdr")

Arguments

Details

Since chi-squared and G tests are approximate tests, exact tests are preferable when the number of individuals is small (200 is a reasonable minimum).

Value

Author(s)

Maxime HERVE <maxime.herve@univ-rennes1.fr>

See Also

chisq.test, prop.test, fisher.test

Examples

# 2-column contingency table: comparison of proportions
tab.cont1 <- matrix(c(17,23,12,24,20,10),ncol=2,dimnames=list(c("Control",
  "Treatment1","Treatment2"),c("Alive","Dead")),byrow=TRUE)
fisher.test(tab.cont1)
fisher.multcomp(tab.cont1)

# 3-column contingency table: independence test
tab.cont2 <- as.table(matrix(c(25,10,12,6,15,14,9,16,9),ncol=3,dimnames=list(c("fair",
  "dark","russet"),c("blue","brown","green"))))
fisher.test(tab.cont2)
fisher.multcomp(tab.cont2)

Fligner-Policello test

Description

Performs a Fligner-Policello test of the null that the medians in the two groups (samples) are the same.

Usage

fp.test(x, ...)

## Default S3 method:
fp.test(x, y, delta = 0, alternative = "two.sided", ...)

## S3 method for class 'formula'
fp.test(formula, data, subset, ...)

Arguments

Details

The Fligner-Policello test does not assume that the shape of the distribution is similar in two groups, contrary to the Mann-Whitney-Wilcoxon test. However, it assumes that the the distributions are symmetric.

Value

Author(s)

Maxime HERVE <maxime.herve@univ-rennes1.fr>

See Also

wilcox.test

Examples

x <- rpois(20,3)
y <- rpois(20,5)
fp.test(x,y)

Individual contributions in regression

Description

Computes difference in regression parameters when each individual is dropped, expressed in proportion of the whole regression coefficients. The function deals with lm (including glm) and least.rect models.

Usage

ind.contrib(model, print.diff = FALSE, graph = TRUE, warning=25)

Arguments

Value

Author(s)

Maxime HERVE <maxime.herve@univ-rennes1.fr>

See Also

lm.influence, least.rect

Examples

x <- 1:30
y <- 1:30+rnorm(30,0,4)
model1 <- lm(y~x)
model2 <- least.rect(y~x)
ind.contrib(model1)
ind.contrib(model2)

Least rectangles linear regression

Description

Fits a least rectangle linear regression, possibly for each level of a factor.

Usage

least.rect(formula, data, conf.level = 0.95, theo = 1, adj = TRUE)

Arguments

Value

Author(s)

Maxime HERVE <maxime.herve@univ-rennes1.fr>

Examples

x <- 1:30+rnorm(30,0,3)
y <- 1:30+rnorm(30,0,3)
regression1 <- least.rect(y~x)
summary(regression1)

x2 <- c(1:30,1:30)
y2 <- c(1:30+rnorm(30,0,3),seq(10,22,12/29)+rnorm(30,0,3))
fact <- gl(2,30,labels=LETTERS[1:2])
regression2 <- least.rect(y2~x2|fact)
summary(regression2)

Slope of a hand-defined line

Description

Returns the slope of a line defined by selecting two points on a graph.

Usage

loc.slp()

Author(s)

Maxime HERVE <maxime.herve@univ-rennes1.fr>

Graphical adujstment of a simple binary logistic regression to data

Description

Cuts the data into intervals, compute the response probability and its standard error for each interval and add the results to the regression curve. No test is performed but this permits to have a graphical idea of the adjustment of the model to the data.

Usage

logis.fit(model, int = 5, ...)

Arguments

Author(s)

Maxime HERVE <maxime.herve@univ-rennes1.fr>

See Also

glm

Examples

x <- 1:50
y <- c(rep(0,18),sample(0:1,14,replace=TRUE),rep(1,18))
model <- glm(y~x,family=binomial)
plot(x,y)
lines(x,model$fitted)
logis.fit(model)

Creating a nls model for logistic regression from fitted values of a glm model

Description

Adds some noise to the fitted values of a glm model to create a nls model for logistic regression (creating a nls model from exact fitted values can not be done, see help of nls).

Usage

logis.noise(model, intensity = 25)

Arguments

Author(s)

Maxime HERVE <maxime.herve@univ-rennes1.fr>

See Also

glm, nls

Examples

x <- 1:50
y <- c(rep(0,18),sample(0:1,14,replace=TRUE),rep(1,18))
model <- glm(y~x,family=binomial)
y2 <- logis.noise(model)
# Then model2 <- nls(y2~SSlogis(...))

Mode

Description

Computes the mode of a vector. The function makes the difference between continuous and discontinuous variables (which are made up of integers only). By extention, it also gives the most frequent value in a character vector or a factor.

Usage

mod(x)

Arguments

Author(s)

Maxime HERVE <maxime.herve@univ-rennes1.fr>

See Also

density

Examples

# Continuous variable
x <- rnorm(100)
mod(x)

# Discontinuous variable
y <- rpois(100,2)
mod(y)

# Character vector
z <- sample(LETTERS[1:3],20,replace=TRUE)
mod(z)

Mood's median test

Description

Performs a Mood's median test to compare medians of independent samples.

Usage

mood.medtest(x, ...)

## Default S3 method:
mood.medtest(x, g, exact = NULL, ...)

## S3 method for class 'formula'
mood.medtest(formula, data, subset, ...)

Arguments

Details

If exact=NULL, a Fisher's exact test is used if the number of data values is < 200; otherwise a chi-square test is used, with Yates continuity correction if necessary.

Value

Author(s)

Maxime HERVE <maxime.herve@univ-rennes1.fr>

Examples

set.seed(1716)
response <- c(rnorm(10,3,1.5),rnorm(10,5.5,2))
fact <- gl(2,10,labels=LETTERS[1:2])
mood.medtest(response~fact)

Multivariate normality QQ-Plot

Description

Draws a QQ-plot to assess multivariate normality.

Usage

mqqnorm(x, main = "Multi-normal Q-Q Plot")

Arguments

Author(s)

Maxime HERVE <maxime.herve@univ-rennes1.fr>

See Also

mshapiro.test, qqPlot

Examples

x <- 1:30+rnorm(30)
y <- 1:30+rnorm(30,1,3)
mqqnorm(cbind(x,y))

Shapiro-Wilk multivariate normality test

Description

Performs a Shapiro-Wilk test to asses multivariate normality. This is a slightly modified copy of the mshapiro.test function of the package mvnormtest, for internal convenience.

Usage

mshapiro.test(x)

Arguments

Value

Author(s)

Maxime HERVE <maxime.herve@univ-rennes1.fr> from the work of Slawomir Jarek

See Also

shapiro.test, mshapiro.test

Examples

x <- 1:30+rnorm(30)
y <- 1:30+rnorm(30,1,3)
mshapiro.test(cbind(x,y))

Pairwise comparisons after an exact multinomial test

Description

Performs pairwise comparisons after a global exact multinomial test. These comparisons are performed using exact binomial tests.

Usage

multinomial.multcomp(x, p.method = "fdr")

Arguments

Details

Since chi-squared and G tests are approximate tests, exact tests are preferable when the number of individuals is small (200 is a reasonable minimum).

An exact multinomial test with two groups is strictly the same than an exact binomial test.

Value

Author(s)

Maxime HERVE <maxime.herve@univ-rennes1.fr>

See Also

multinomial.test, binom.test

Examples

counts <- c(5,15,23)
multinomial.test(counts)
multinomial.multcomp(counts)

Exact multinomial test

Description

Perfoms an exact multinomial test on a vector of counts.

Usage

multinomial.test(x, p = rep(1/length(x), length(x)))

Arguments

Details

The function works as chisq.test or G.test :

- if theoretical proportions are not given, equality of counts is tested

- if theoretical proportions are given, equality of counts to theoretical counts (given by theoretical proportions) is tested.

Since chi-squared and G tests are approximate tests, exact tests are preferable when the number of individuals is small (200 is a reasonable minimum).

Be aware that the calculation time increases with the number of individuals (i.e. the sum of x) and the number of groups (i.e. the length of x).

An exact multinomial test with two groups is strictly the same as an exact binomial test.

Value

Author(s)

Maxime HERVE <maxime.herve@univ-rennes1.fr> based on multinomial.test

See Also

chisq.test, G.test, binom.test, multinomial.multcomp, multinomial.theo.multcomp

Examples

counts <- c(5,15,23)
multinomial.test(counts)

Pairwise comparisons after an exact multinomial test for given probabilities

Description

Performs pairwise comparisons after a global exact multinomial test for given probabilities. These comparisons are performed using exact binomial tests.

Usage

multinomial.theo.multcomp(x, p = rep(1/length(x), length(x)), prop = FALSE,
  p.method = "fdr")

Arguments

Details

Since chi-squared and G tests are approximate tests, exact tests are preferable when the number of individuals is small (200 is a reasonable minimum).

An exact multinomial test with two groups is strictly the same than an exact binomial test.

Value

Author(s)

Maxime HERVE <maxime.herve@univ-rennes1.fr>

See Also

multinomial.test, binom.test

Examples

counts <- c(5,15,23)
p.theo <- c(0.2,0.5,0.3)
multinomial.test(counts,p=p.theo)
multinomial.theo.multcomp(counts,p=p.theo)

Univariate correlation test for multiple variables

Description

Performs correlation tests between one variable and a series of other variables, and corrects p-values.

Usage

multtest.cor(mult.var, uni.var, method = "pearson", p.method = "fdr",
  ordered = TRUE)

## S3 method for class 'multtest.cor'
plot(x, arrows = TRUE, main = NULL, pch = 16,
  cex = 1, col = c("red", "orange", "black"), labels = NULL, ...)

Arguments

Author(s)

Maxime HERVE <maxime.herve@univ-rennes1.fr>

See Also

cor.test

Examples

data(iris)

# Original coordinates
plot(Petal.Length~Sepal.Length,pch=16,col=as.numeric(iris$Species),data=iris)

# New axis
abline(-6,1.6)

# Coordinates on new axis
new.coord <- coord.proj(iris[,c("Sepal.Length","Petal.Length")],1.6)

# Correlation between the whole dataset and new coordinates
mult.cor <- multtest.cor(iris[,1:4],new.coord)
plot(mult.cor)

Univariate comparison of groups for multiple variables

Description

Performs group comparisons for multiple variables using parametric, permutational or rank tests, and corrects p-values. Gives also group means and standards errors for each variable.

Usage

multtest.gp(tab, fac, test = c("param", "perm", "rank"),
  transform = c("none", "sqrt", "4rt", "log"), add = 0, p.method = "fdr",
  ordered = TRUE, ...)

## S3 method for class 'multtest.gp'
plot(x, signif = FALSE, alpha = 0.05,
  vars = NULL, xlab = "Group", ylab = "Mean (+/- SE) value",
  titles = NULL, groups = NULL, ...)

Arguments

Details

In case of parametric tests, t-tests or ANOVAs are used depending on the number of groups (2 or more, respectively). In case of permutational tests: permutational t-tests or permutational ANOVAs. In case of rank-based tests: Mann-Whitney-Wilcoxon or Kruskal-Wallis tests.

Author(s)

Maxime HERVE <maxime.herve@univ-rennes1.fr>

See Also

perm.anova, perm.t.test

Examples

data(iris)
mult <- multtest.gp(iris[,1:4],iris$Species)
plot(mult)

Univariate comparison of groups for multiple binary variables

Description

Performs group comparisons for multiple binary variables using a parametric or an exact test, and corrects p-values. Gives also group proportions and standards errors for each variable.

Usage

multtest.gp.bin(tab, fac, test = c("LRT", "Fisher"),
  p.method = "fdr", ordered = TRUE, ...)

## S3 method for class 'multtest.gp.bin'
plot(x, signif = FALSE, alpha = 0.05,
  vars = NULL, xlab = "Group", ylab = "Mean (+/- SE) proportion",
  titles = NULL, groups = NULL, ...)

Arguments

Author(s)

Maxime HERVE <maxime.herve@univ-rennes1.fr>

See Also

multtest.gp, Anova, fisher.test

Re-computation of an ordination using given row weights

Description

Re-computes an ordination using given row weights (possibly extracted from a correspondence analysis). The function is intended to be used prior to coinertia when row weights have to be equalized.

Usage

ord.rw(ord, CA = NULL, rw = NULL)

Arguments

Author(s)

Maxime HERVE <maxime.herve@univ-rennes1.fr>

Estimation of overdispersion with glmer models

Description

Estimates residual deviance and residual degrees of freedom to check for overdispersion with glmer models. This function is directly comming from http://glmm.wikidot.com/faq.

Usage

overdisp.glmer(model)

Arguments

Author(s)

Ben Bolker

See Also

glmer

Examples

require(lme4)

# Example from the 'glmer' function
gm1 <- glmer(cbind(incidence,size-incidence)~period+(1|herd),
 family="binomial",data=cbpp)
overdisp.glmer(gm1)

Rescaling of a [0,1] variable into the (0,1) interval (and vice-versa)

Description

The function uses the formula presented in Douma & Weedon (2019). It is primarily intended to be used in beta regression (regression for continuous proportions) when data contain zeroes and/or ones, but can be applied to any variable initially bounded in the [0,1] interval when rescaling is necessary. The function can also perform back-transformation.

Usage

p.beta(p, n = length(p), C = 2, back = FALSE)

Arguments

Author(s)

Maxime HERVE <maxime.herve@univ-rennes1.fr> from the following paper: Douma JC & Weedon JT (2019) Analysing continuous proportions in ecology and evolution: A practical introduction to beta and Dirichlet regression. Methods in Ecology and Evolution, 10: 1412-1430

Examples

# A fictive example with four animals performing a behavioral choice-test where time
#   can be spent in three branches (total time 20 min)
(tab <- data.frame(Individual=c("Ind1","Ind2","Ind3","Ind4"),Branch1=c(0,12,20,4),
  Branch2=c(8,4,0,6),Branch3=c(12,4,0,10)))
# Raw proportions of time spent in branch 1:
(p1 <- tab$Branch1/rowSums(tab[,-1]))
# Scaled proportions:
p.beta(p1,C=3)

Pairwise comparisons for CDA

Description

Performs pairwise comparisons between group levels with corrections for multiple testing, using CDA.test.

Usage

pairwise.CDA.test(X, fact, ncomp = NULL, p.method = "fdr", ...)

Arguments

Details

See CDA.test.

Value

Author(s)

Maxime HERVE <maxime.herve@univ-rennes1.fr>

See Also

CDA.test

Examples

require(ade4)
data(perthi02)

CDA.test(perthi02$tab,perthi02$cla)
pairwise.CDA.test(perthi02$tab,perthi02$cla)

Pairwise comparisons for proportions using G-tests

Description

Performs pairwise comparisons between pairs of proportions with correction for multiple testing.

Usage

pairwise.G.test(x, p.method = "fdr")

Arguments

Details

Since a G-test is an approximate test, an exact test is preferable when the number of individuals is small (200 is a reasonable minimum). See fisher.multcomp in that case.

Value

See Also

G.test, fisher.multcomp

Examples

x <- matrix(c(44,56,36,64,64,40),ncol=2,dimnames=list(c("Control","Treatment1","Treatment2"),
  c("Alive","Dead")),byrow=TRUE)
G.test(x)
pairwise.G.test(x)

Pairwise permutation tests based on cross (model) validation

Description

Performs pairwise comparisons between group levels with corrections for multiple testing, using MVA.test.

Usage

pairwise.MVA.test(X, fact, p.method = "fdr", cmv = FALSE, ncomp = 8,
  kout = 7, kinn = 6, model = c("PLS-DA", "PPLS-DA", "LDA", "QDA",
  "PLS-DA/LDA", "PLS-DA/QDA", "PPLS-DA/LDA","PPLS-DA/QDA"),
  nperm = 999, progress = TRUE, ...)

Arguments

Details

The function deals with the limitted floating point precision, which can bias calculation of p-values based on a discrete test statistic distribution.

Value