MCPtest {MCPtests}

Multiple comparison procedures to the means.

Description

MCPtest applies several tests of multiple comparisons present in the literature. The tests chosen are based on the evaluation of the researcher to decide the choice of test for analysis in the experiment.

Usage

MCPtest(
  y,
  trt = NULL,
  dferror = NULL,
  mserror = NULL,
  replication = NULL,
  alpha = 0.05,
  main = NULL,
  MCP = "all",
  ismean = FALSE,
  parallel = FALSE
)

Arguments

y

Model (aov or lm), numeric vector containing the response variable or the mean of the treatments.

trt

Constant (y = model) or a vector containing the treatments.

dferror

Degrees of freedom of the Mean Square Error.

mserror

Mean Square Error.

replication

Number de repetitions of the treatments in the experiment. For unbalanced data should be informed the harmonic mean of repetitions. This argument should be informed only if ismean = TRUE.

alpha

Significant level. The default is alpha = 0.05.

main

Title of the analysis.

MCP

Allows choosing the multiple comparison test; the defaut is "all". This option will go perform all tests. However, the options are: the Skott-Knott midrange test ("MGM"), the Skott-Knott Range test ("MGR"), the Tukey midrange test ("TM"), the Scott-Knott's test ("SK").

ismean

Logic. If FALSE (default), the y argument represents a model (aov or lm) or a numeric vector containing the response variable. If TRUE the y argument represents the mean of treatments.

parallel

Logic. If FALSE (default), the MCPtest function runs without parallelization. Otherwise, the code is executed with parallelization. Note that the parallelization is not always more efficient.

Details

The MCP argument allows you to choose various tests of multiple comparisons at once. For example, MCP = c("MGM", "MGR"), and so on.

Value

MCPtest returns the print of a list of results. First, the summary of y. Second, the statistics of the test chosen. And finally, the mean group results for each test. If MCPtest function is stored in an object, the results will be printed and also stored in the object.

References

BATISTA, Ben Deivide de Oliveira. Testes de comparacoes multiplas baseados na distribuicao da midrange estudentizada externamente. 2016. 194f. Tese (Doutorado em Estatistica e Experimentacao Agropecuaria) - Universidade Federal de Lavras, 2016. (Portuguese, Brazil)

SCOTT, A. J.; KNOTT, M. A cluster analysis method for grouping means in the analysis of variance. Biometrics, International Biometric Society, v. 30, n. 3, p. 507-512, 1974.

DUNCAN, D. B. Multiple range and multiple F Tests. Biometrics, International Biometric Society, v. 11, n. 1,p. 1-42, 1955.

Examples

# Simulated data (completely randomized design)

# Response variable
rv <- c(100.08, 105.66, 97.64, 100.11, 102.60, 121.29, 100.80,
        99.11, 104.43, 122.18, 119.49, 124.37, 123.19, 134.16,
        125.67, 128.88, 148.07, 134.27, 151.53, 127.31)

# Treatments
treat <- factor(rep(LETTERS[1:5], each = 4))

# Anova
res     <- anova(aov(rv~treat))
DFerror <- res$Df[2]
MSerror <- res$`Mean Sq`[2]

# Loading the MCPtests package
library(MCPtests)

# applying the tests
results <- MCPtest(y = rv,
                  trt = treat,
                  dferror = DFerror,
                  mserror = MSerror,
                  alpha = 0.05,
                  main = "Multiple Comparison Procedure: MGM test",
                  MCP = c("MGM"))

# Other option for the MCP argument is "all". All tests are used.

results$Groups     # Results of the tests
results$Statistics # Main arguments of the tests
results$Summary    # Summary of the response variable

# Using the y argument as aov or lm model
res  <- aov(rv~treat)

MCPtest(y = res, trt = "treat", alpha = 0.05,
       main = "Multiple Comparison Procedure: MGM test",
       MCP = c("MGM"))

# For unbalanced data: It will be used the harmonic mean of
#                       the number of experiment replicates

# Using the previous example
rv <- rv[-1]
treat <- treat[-1]

res  <- lm(rv~treat) # Linear model

# Multiple comparison procedure: MGR test
MCPtest(y = res, trt = "treat", alpha = 0.05,
       main = "Multiple Comparison Procedure: MGR test",
       MCP = c("MGR"))

# Assuming that the available data are the averages
#  of the treatments and the analysis of variance

# Analysis of Variance Table

# Response: rv
#            Df Sum Sq Mean Sq F value    Pr(>F)
# treat      4 4135.2 1033.80  14.669 4.562e-05 ***
# Residuals 15 1057.1   70.47

mean.treat <- c(100.87, 105.95, 117.62, 127.97, 140.30)
treat <- factor(LETTERS[1:5])
DFerror <- 15
MSerror <- 70.47488
replic <- 4

MCPtest(y = mean.treat,
       trt = treat,
       dferror = DFerror,
       mserror = MSerror,
       replication = replic,
       alpha = 0.05,
       main = "Multiple Comparison Procedure: MGM test",
       MCP = c("MGM"),
       ismean = TRUE)