Skip to contents

Test Statistics for the Two-Sample Problem

Source: R/two-sample-stats.R
two-sample-stats.Rd

This is a collection of functions that provide test statistics to be used into the permutation scheme for performing two-sample testing. These test statistics can be divided into two categories: traditional statistics that use empirical moments and inter-point statistics that only rely on pairwise dissimilarities between data points.

Usage

stat_welch(data, indices1, ...)

stat_student(data, indices1, ...)

stat_t(data, indices1, ...)

stat_fisher(data, indices1, ...)

stat_f(data, indices1, ...)

stat_mean(data, indices1, ...)

stat_hotelling(data, indices1, ...)

stat_bs(data, indices1, ...)

stat_student_ip(data, indices1, ...)

stat_t_ip(data, indices1, ...)

stat_fisher_ip(data, indices1, ...)

stat_f_ip(data, indices1, ...)

stat_bg_ip(data, indices1, ...)

stat_energy_ip(data, indices1, alpha = 1L, ...)

stat_cq_ip(data, indices1, ...)

stat_mod_ip(data, indices1, ...)

stat_dom_ip(data, indices1, standardize = TRUE, ...)

Arguments

data

Either a list of the n1 + n2 concatenated observations with the original n1 observations from the first sample on top and the original n2 observations from the second sample below. Or a dissimilarity matrix stored as a dist object for all inter-point statistics whose function name should end with _ip().

indices1

An integer vector specifying the indices in data that are considered to belong to the first sample.

...

Extra parameters specific to some statistics.

alpha

A scalar value specifying the power to which the dissimilarities should be elevated in the computation of the inter-point energy statistic. Default is 1L.

standardize

A boolean specifying whether the distance between medoids in the stat_dom_ip function should be normalized by the pooled corresponding variances. Default is TRUE.

Value

A real scalar giving the value of test statistic for the permutation specified by the integer vector indices.

Traditional Test Statistics

  • stat_hotelling implements Hotelling's \(T^2\) statistic for multivariate data with \(p < n\).

  • stat_student or stat_t implements Student's statistic (originally assuming equal variances and thus using the pooled empirical variance estimator). See t.test for details.

  • stat_welch implements Student-Welch statistic which is essentially a modification of Student's statistic accounting for unequal variances. See t.test for details.

  • stat_fisher or stat_f implements Fisher's variance ratio statistic. See var.test for details.

  • stat_mean implements a statistic that computes the difference between the means.

  • stat_bs implements the statistic proposed by Bai & Saranadasa (1996) for high-dimensional multivariate data.

Inter-Point Test Statistics

  • stat_student_ip or stat_t_ip implements a Student-like test statistic based on inter-point distances only as described in Lovato et al. (2020).

  • stat_fisher_ip or stat_f_ip implements a Fisher-like test statistic based on inter-point distances only as described in Lovato et al. (2020).

  • stat_bg_ip implements the statistic proposed by Biswas & Ghosh (2014).

  • stat_energy_ip implements the class of energy-based statistics as described in Székely & Rizzo (2013);

  • stat_cq_ip implements the statistic proposed by Chen & Qin (2010).

  • stat_mod_ip implements a statistic that computes the mean of inter-point distances.

  • stat_dom_ip implements a statistic that computes the distance between the medoids of the two samples, possibly standardized by the pooled corresponding variances.

References

Bai, Z., & Saranadasa, H. (1996). Effect of high dimension: by an example of a two sample problem. Statistica Sinica, 311-329.

Lovato, I., Pini, A., Stamm, A., & Vantini, S. (2020). Model-free two-sample test for network-valued data. Computational Statistics & Data Analysis, 144, 106896.

Biswas, M., & Ghosh, A. K. (2014). A nonparametric two-sample test applicable to high dimensional data. Journal of Multivariate Analysis, 123, 160-171.

Székely, G. J., & Rizzo, M. L. (2013). Energy statistics: A class of statistics based on distances. Journal of statistical planning and inference, 143(8), 1249-1272.

Chen, S. X., & Qin, Y. L. (2010). A two-sample test for high-dimensional data with applications to gene-set testing. The Annals of Statistics, 38(2), 808-835.

Examples

n <- 10L
mx <- 0
sigma <- 1
delta <- 10
my <- mx + delta
x <- rnorm(n = n, mean = mx, sd = sigma)
y <- rnorm(n = n, mean = my, sd = sigma)
D <- dist(c(x, y))

x <- as.list(x)
y <- as.list(y)

stat_welch(c(x, y), 1:n)
#> [1] -22.40635
stat_t(c(x, y), 1:n)
#> [1] 22.40635
stat_f(c(x, y), 1:n)
#> [1] 1.625238
stat_mean(c(x, y), 1:n)
#> [1] -10.54055
stat_hotelling(c(x, y), 1:n)
#> [1] 100.4089
stat_bs(c(x, y), 1:n)
#> [1] 110.904

stat_t_ip(D, 1:n)
#> [1] 501.0447
stat_f_ip(D, 1:n)
#> [1] 1.625238
stat_bg_ip(D, 1:n)
#> [1] 174.8656
stat_energy_ip(D, 1:n)
#> [1] 9.469125
stat_cq_ip(D, 1:n)
#> [1] -18.70016
stat_mod_ip(D, 1:n)
#> [1] 10.54055
stat_dom_ip(D, 1:n)
#> [1] 10.12451