# perf_wilcox: Compute column-wise statistics in a performance matrix In b2slab/diffuStats: Diffusion scores on biological networks

## Description

Function perf_wilcox compares all the columns of a matrix through a wilcox.test. The columns are assumed to be performance measures (e.g. AUROC) whereas the rows are instances.

## Usage

 1 2 3 4 5 6 7 8 perf_wilcox( perf_mat, adjust = function(p) stats::p.adjust(p, method = "fdr"), ci = 0.95, digits_ci = 2, digits_p = 3, ... )

## Arguments

 perf_mat Numeric matrix whose columns contain performance metrics of different methods. adjust Function to adjust the p-values for multiple testing. By default, p.adjust with its default parameters is used. ci Numeric, confidence interval (defaults to 0.95) digits_ci Integer, digits to display in the confidence interval digits_p Integer, digits to display in the p-value ... further arguments for format

## Details

The statistical comparison of the columns is intended to ease comparisons between methods in a rigorous way. Methods are compared pairwise and a p-value for difference in performance. The function perf_wilcox returns a character matrix so that (1) the upper triangular matrix contains confidence intervals on the estimate of the difference between performances, and (2) the lower triangular matrix contains the two-tailed p-value that tests difference in performance, with multiple testing correction. The comparison takes place between row and column in that precise order: a positive difference favours the row and a negative one, the column.

## Value

Character matrix. The upper triangular matrix contains a confidence interval and the estimate of the pairwise difference in performance. The lower triangular matrix shows the associated two-tailed p-value, with multiple testing correction.

## Examples

 1 2 3 4 5 6 7 8 9 10 11 12 13 # Dummy data frame to test n <- 100 perf_mat <- cbind( good = runif(n = n, min = 0.5, max = 1), so_so = runif(n = n, min = 0.2, max = 0.7), bad = runif(n = n, min = 0, max = 0.5) ) wilcox_mat <- perf_wilcox(perf_mat) # See how the methods in the rows compare to those # in the columns, confidence interval # (upper) and p-value (lower) wilcox_mat

b2slab/diffuStats documentation built on Feb. 26, 2021, 2 p.m.