mrpp: Multiresponse permutation procedures

Description

Multiresponse permutation procedures (MRPP) are used for univariate and multivariate analyses of grouped data in a completely randomized one-way design. MRPP are used for comparing equality of treatment groups analogous to one-way analysis of variance (or t-test) for univariate data, or multivariate analysis of variance (Hotelling's T^2) for multivariate data.

Usage

mrpp(variables, group, data, expon = 1, c.form = 1, hotelling = FALSE, commens = TRUE, interv = 0, number.perms, exact = FALSE, has.excess = FALSE, excess.value, max.dist, save.test)

Arguments

variables

the names of response variables to be used in the analysis. If more than one is used these are specified using the form c(var1,var2,...).

group

the name of the grouping variable to be used in the analysis.

data

the data.frame or matrix containing columns with names matching all values supplied in the variables and group arguments. Alternatively, if neither variables and group are supplied, it is assumed that the first column is the grouping column, and all remaining columns are variables to be used in the analysis.

expon

allows selection of alternative exponents in distance calculations. Default uses 1 corresponding to Euclidean distance. Use of 2 is squared Euclidean distance, corresponding to many conventional parametric tests on means.

c.form

has four options that control how the groups are weighted:

c.form = 1: C(I) = n(I)/sum(n(I))
c.form = 2: C(I) = (n(I)-1)/sum(n(I)-1)
c.form = 3: C(I) = 1/sum(1)
c.form = 4: C(I) = (n(I)*(n(I)-1))/sum(n(I)*(n(I)-1))

hotelling

a logical indicating Hotelling's variance/covariance standardization of the multiple dependent variables.

commens

a logical value indicating whether to perform average Euclidian distance commensuration of multiple response variables. Commensuration can only be done when there is more than one variable.

interv

allows an analysis to be conducted on univariate circular data such as time or compass orientation. This analysis recognizes that there are no endpoints to the measurement scale. interv should be set to the number of units in the circular measure.

number.perms

if specified a Monte Carlo resampling procedure with number.perms permutations is to be used rather than a Pearson III approximation.

exact

a logical value indicating whether to perform an exact test. This is computationally intensive for >30 observations.

has.excess

a logical indicating whether there is an excess group.

excess.value

the value of the excess group, if not specified it is assumed that the largest grouping value indicates the excess group.

max.dist

specifying a numeric value causes the MRPP analysis to replace interobject distances delta_(i,j) greater than the truncation value with the truncation value.

save.test

a logical indicating to store the permutation values of the test statistic. This is only a valid option when number.perms is set.

Value

mrpp returns an object of either class MRPPObj or EMRPPObj.The functions summary as well as print can be used to obtain a summary of the test.Generic accessor functions pvalue and ResampVals (for MRPPObj) can be used to obtain the p-value and Monte Carlo resampled test statistic values respectively.

Details

The default Euclidean distance function in MRPP provides an omnibus test of distributional equivalence among groups or a test for common medians if the assumption of equal dispersions is applicable. Options allow MRPP to perform permutation (randomization) versions of t-tests, one-way analysis of variance, Kruskal-Wallis tests (for ranked data), Mann-Whitney Wilcoxon tests (for ranked data), and one-way multivariate analysis of variance. Options in MRPP also allow you to truncate distances to evaluate multiple clumping of data, establish an excess group, and select arc distances to compare circular distributions of grouped data. Multivariate data are commensurated (standardized) to a common scale but an option allows you to turn off commensuration. Commensuration can be done by using average Euclidean distance (default) or the variance/covariance matrix for the dependent variables. Multivariate medians and distance quantiles (MEDQ) are provided as estimates to be used in describing distributional changes detected by MRPP analyses.

References

Mielke, P.W., Jr., and K.J. Berry. 2001. Permutation methods: A distance function approach. Springer-Verlag.

Examples

Run this code

out <- mrpp(variables = c(distance,elev),group = sex,data = bgrouse,
 exact = TRUE)
summary(out)

Run the code above in your browser using DataLab