print_wilcoxon_rs: Print the results of a Wilcoxon rank sum test (Mann-Whitney-U...

In m-Py/prmisc: Miscellaneous Printing of Numeric and Statistical Output in R Markdown and Quarto Documents

Print the results of a Wilcoxon rank sum test (Mann-Whitney-U test)

Description

Print the results of a Wilcoxon rank sum test (Mann-Whitney-U test)

Usage

print_wilcoxon_rs(
  wc_object,
  decimals_p = 3,
  consistent = NULL,
  group1 = NULL,
  group2 = NULL,
  groupvar = NULL,
  effsize = NULL,
  neg = FALSE,
  N = NULL,
  decimals_eff = NULL
)

Arguments

wc_object	an object returned by wilcox.test
decimals_p	how many decimals should be printed for the p-value (defaults to 3)
consistent	an optional parameter determining for which group the test statistic U should be reported. Can be set to 'min' or 'max'. See details.
group1	a vector containing the cases of the first group
group2	a vector containing the cases of the second group
groupvar	a vector containing a grouping variable
effsize	a character indicating which effect size should be reported, if any. Possible values are: 'r', 'rsqu' and 'd'. By default, no effect size will be reported. See details.
neg	a logical indicating whether the effect size should be reported with a negative sign. Defaults to FALSE. See details.
N	an integer passing information about the total N of the sample. Needed when effect sizes should be calculated, but N cannot be inferred because neither of group1, group2 or groupvar have been provided. Should groups or a grouping variable have been provided, N must not be used.
decimals_eff	an integer specifying how many digits the effect size should be printed with, if an effect size is desired. Defaults to 2. r and r squared will be printed without a leading zero, d will be printed with a leading zero because it is possible for d to take values larger than one.

Details

In order to calculate a Wilcoxon rank sum test, the argument paired in wilcox.test needs to be FALSE. Otherwise, a Wilcoxon signed rank test will be computed instead and the statistics printed by print_wilcoxon_rs will be misleading.

Note that the test statistic W calculated in wilcox.test that is printed as test statistic U in print_wilcoxon_rs will correspond to the U of the first of the two groups compared in wilcox.test. In the default method of wilcox.test, this is the vector assigned to x. In the formula method, this is the first group as identified by the grouping variable. Some software, like SPSS, consistently reports the smaller or larger U. If you wish to mimic this, you can specify the desired behaviour by providing the consistent argument. Setting consistent to 'min' will print the smaller of the two U, setting it to 'max' will print the larger U. In order to do so, you need to provide the n for both groups.

You can either do that by passing the data of both groups to the arguments group1 and group2, respectively. From those, print_wilcoxon_rs will calculate the group sizes. Any vector with the length of the corresponding group size will produce the desired result. This is the recommended option if your wc_object has been created using the default version of wilcox.test, i.e. if groups were passed as x and y. The order in which you pass group1 and group2 to print_wilcoxon_rs does not have to correspond to the order in which x and y were passed to wilcox.test.

Alternatively, you can pass a grouping variable to the argument groupvar. From this, print_wilcoxon_rs will calculate group sizes. This is the recommended option if your wc_object has been created using the formula version of wilcox.test, i.e. if a grouping variable was passed after the ~. To prevent mistakes, you can either use group1 and group2 or groupvar.

By default, when consistent is not provided, print_wilcoxon_rs will print U using W as taken from wilcox.test.

There are three options available for calculating an effect size via the argument effsize:

• the point biserial correlation r (effsize = 'r'), which is calculated by dividing Z-score by the square root of N. According to Cohen (1988) a small, medium and large effect correspond to r = .1, .3 and .5, respectively. Currently, print_wilcoxon_rs infers the Z-score from the p-value.

• r squared (identical to eta squared; effsize = 'rsqu'), which is the ratio of variability associated with an effect compared to the ratio of overall variance

• d (effsize = 'd'), which is calculated from r as follows:

2*r/(sqrt(1-r^2))

According to Cohen (1988), a small, medium and large effect correspond to r = .2, .5 and .8, respectively.

Information on the direction of the effect (as indicated in the sign of the Z-score) or in which order the groups have been compared is not included in the output of wilcox.test. Hence, only absolute effect sizes can be calculated. It is advised to report absolute effect sizes unless there is a meaningful for the two groups tested (e.g. Fritz et al, 2012). In some cases, it might be desired to specify the direction of an effect by including the sign of the effect. To that end, it is possible to print the negative effect size with neg = TRUE, but caution is advised: You should always check if the sign of the effect size you report is the correct one, especially in the case of one-sided testing.

Value

A string describing the results of the Wilcoxon test; to be included in an R markdown document.

Author(s)

Juliane Nagel [e-mail]

References

Cohen, J. (1988). Statistical power analysis for the behavioral sciences (2nd ed.). Hillsdale, NJ: Erlbaum.

Fritz, C. O., Morris, P. E., & Richler, J. J. (2012). Effect size estimates: Current use, calculations, and interpretation. Journal of Experimental Psychology: General, 141(1), 2-18. http://dx.doi.org/10.1037/a0024338

Examples

data(iris)
dat <- subset(iris, Species %in% c("setosa", "versicolor"))
wc_iris <- wilcox.test(dat$Sepal.Length ~ dat$Species, correct = FALSE)
# include this call in Rmd inline code
print_wilcoxon_rs(wc_iris)

wc_iris2 <- wilcox.test(dat$Sepal.Width ~ dat$Species, correct = FALSE)
                        
print_wilcoxon_rs(wc_iris2, consistent = "min", 
                  group1 = dat$Sepal.Width[dat$Species == "setosa"], 
                  group2 = dat$Sepal.Width[dat$Species == "versicolor"])

print_wilcoxon_rs(wc_iris2, consistent = "max", 
                  groupvar = dat$Species)

print_wilcoxon_rs(wc_iris2, consistent = "max", 
                  groupvar = dat$Species, effsize = "rsqu")
                  

m-Py/prmisc documentation built on March 29, 2025, 2:30 p.m.