modified.gtest: Zero-Tolerant G-Test for Contingency Tables

In Upsilon: Another Test of Association for Count Data

Zero-Tolerant G-Test for Contingency Tables

Description

Performs G-test \insertCiteWOOLF:1957aaUpsilon on contingency tables, slightly modified to handle rows or columns of all zeros.

Usage

modified.gtest(x, log.p = FALSE)

Arguments

x	a matrix or data frame of floating or integer numbers to specify a contingency table. Entries must be non-negative.
log.p	a logical. If TRUE, the p-value is calculated in closed form to natural logarithm of p-value to improve numerical precision when p-value approaches zero. Defaults to FALSE.

Details

This test is useful if a p-value must be returned on a contingency table with valid non-negative counts, where other implementations of G-test could return NA as the p-value, regardless of a pattern being strong or weak.

This function handles tables with empty rows or columns (where expected values are 0) by calculating the test statistic over non-zero entries only. This prevents the result from becoming NA, while giving meaningful p-values.

Value

A list with class "htest" containing:

statistic	the G statistic (log-likelihood ratio).
parameter	the degrees of freedom.
p.value	the p-value of the test.
estimate	the value of mutual information.
method	a character string indicating the method used.
data.name	a character string, name of the input data.
observed	the observed counts.
expected	the expected counts under the null hypothesis.

References

\insertRef

WOOLF:1957aaUpsilon

Examples

library("Upsilon")

# Create a sparse table with empty rows/cols
x <- matrix(
  c(0, 3, 0, 
    3, 0, 0), 
  nrow = 2, byrow = TRUE
)
print(x)
# Perform the modified G-test
modified.gtest(x)

Upsilon documentation built on March 7, 2026, 5:07 p.m.