distdiffr: The distdiffR two-sample tests of bivariate distributional...

In EricMcKinney77/distdiffR: Two-Sample Tests of Distributional Equality for Bivariate Data

The distdiffR two-sample tests of bivariate distributional equality

Description

The distdiffr() function conducts two-sample permutation tests of distributional equality based on differences in the bivariate empirical cumulative density functions (BECDFs). The differences in BECDFs are computed across a series of rotations, toroidal shifts, or both rotations and toroidal shifts of the combined data (specified via testType). The number of rotations and toroidal shifts may be specified (via numRot or numShifts, respectively). The number of toroidal shifts may also be determined by a proportion of the combined sample size (via propPnts). However, \insertCitemckinney2022extensions;textualdistdiffR has shown that limiting the number of toroidal shifts to ease the computational load of the test will still provide stable results. Simulations have shown the combined rotational and toroidal shift test to be the most powerful yet appropriately conservative test. For more information, see \insertCitemckinney2022extensions;textualdistdiffR and \insertCitemckinney2021extensions;textualdistdiffR.

Usage

distdiffr(
  data1,
  data2,
  testType = "combined",
  numRot = 8,
  propPnts = NULL,
  numShifts = NULL,
  shiftThrshld = 25,
  numPerms = 999,
  psiFun = CalcPsiCWS,
  seedNum = NULL
)

Arguments

data1	A two column matrix of bivariate observations from one sample.
data2	A two column matrix of bivariate observations from another sample.
testType	A string indicating the type of test to be used. Must be one of c("rotational", "toroidal", "combined").
numRot	An integer number of rotational shifts of the pooled samples.
propPnts	A numeric proportion of points to be used as toroidal shift origins. Cannot provide both propPnts and numShifts. If neither are provided, shiftThrshld is used.
numShifts	A numeric integer. The number of points to be used as toroidal shift origins. Must be less than the pooled sample size. Cannot provide both propPnts and numShifts. If neither are provided, shiftThrshld is used.
shiftThrshld	A numeric integer. Used if neither propPnts or numShifts are provided. If the pooled sample size is less than shiftThrshld, every point will be used as a toroidal shift origin. Otherwise, only a random sample of shiftThrshld points will be used.
numPerms	An integer number of permutations of the original data.
psiFun	A function specifying the Psi statistic calculation.
seedNum	An integer random seed value.

Value

A list including three objects: (1) the Psi statistic computed on the original data (2) a vector of Psi statistics computed on the permuted data (3) the p-value for the test

References

\insertRef

mckinney2022extensionsdistdiffR

\insertRef

mckinney2021extensionsdistdiffR

Examples

# Randomly assign all three species to two samples
seedNum <- 123
set.seed(seedNum)

data(iris)
# Randomly assign all three species to two samples
irisPermuted <- iris[sample.int(nrow(iris)), ]
sample1 <- as.matrix(irisPermuted[1:75, 1:2])
sample2 <- as.matrix(irisPermuted[76:150, 1:2])
pooled_data <- rbind(cbind(sample1, 1), cbind(sample2, 2))

# Rotational test
output <- distdiffr(sample1, # Note: Data inputs must be matrices
                    sample2,
                    testType = "rotational",
                    numRot = 8, # Default value
                    seedNum = seedNum)
output$pval

# Toroidal shift test with proportions of points
output <- distdiffr(sample1,
                    sample2,
                    testType = "toroidal",
                    propPnts = 0.1,
                    seedNum = seedNum)
output$pval

# Toroidal shift test with a threshold below pooled sample size
output <- distdiffr(sample1,
                    sample2,
                    testType = "toroidal",
                    shiftThrshld = 25, # Default
                    seedNum = seedNum)
output$pval

# Toroidal shift test with a threshold above pooled sample size
output <- distdiffr(sample1,
                    sample2,
                    testType = "toroidal",
                    shiftThrshld = 200,
                    seedNum = seedNum)
output$pval

# Toroidal shift test with a number of shifts
output <- distdiffr(sample1,
                    sample2,
                    testType = "toroidal",
                    numShifts = 8,
                    seedNum = seedNum)
output$pval

# Combined rotational and toroidal shift test
output <- distdiffr(sample1,
                    sample2,
                    testType = "combined", # Default
                    numRot = 8,            # Default
                    shiftThrshld = 25,     # Default
                    seedNum = seedNum)
output$pval

# Also see browseVignettes(package = "distdiffR")

EricMcKinney77/distdiffR documentation built on April 24, 2022, 9:03 p.m.