R vs SAS vs StatXact - Wilcoxon signed-rank test

Wilcoxon signed-rank test in R, SAS and StatXact

Introduction

This section compares the implementation of Wilcoxon signed-rank test in R, SAS and StatXact.

General Comparison Table

The following table provides an overview of the methods support comparability between R, SAS and StatXact for the new analysis point.

Analysis Supported in R Supported in SAS Supported in StatXact Notes
Wilcoxon signed-rank test with p value only YES YES YES Available in all, but results match only between R and StatXact. See details on p value on each software page.
Hodges-Lehmann estimator YES NO YES Available in R and StatXact only. In SAS needs to be derived manually.

Exact/non-exact method

Non-exact method with/without continuity correction

YES

YES

NO

NO

YES

NO

Stats package in R and StatXact support both options. SAS applies a default one depending on N.

Only Stats package in R support both options.
Dataset with “0” differences YES NO YES SAS ignores 0s. In R only Coin package supports.
Dataset with ties YES YES YES Supported in SAS and StatXact. In R only in Coin.

Example Data

Analysis will be conducted on the example of anonymized data from 2-period, cross-over study comparing treatments A and B in patients with asthma and acute airway obstruction induced by repeated mannitol challenges.

For the purpose of the results comparison we will consider a specific case where the dataset has no ties and N (number of observations) = 240.

Wilcoxon signed rank test was applied to analyse the time to return to baseline FEV1 post-mannitol challenge 2. Median difference, p value and 95% CI were provided using the Hodges-Lehmann estimator.

blood_p <- read.csv("../data/WilcoxonSignedRank_TTR.csv", dec = ".")
head(blood_p)
  SUBJID   TRT_A   TRT_B
1      1 143.670 153.316
2      2 163.082 170.576
3      3 153.393 168.599
4      4 153.082 142.358
5      5 146.720 141.193
6      6 150.668 147.204

Results Comparison

Summary and Recommendation

  • If you need a flexibility of choosing between methods (exact, non-exact etc) in one software, go for R or StatXact.
  • If you need a detailed documentation explaining applied methods, go for StatXact.
  • If you only need p value and don’t mind a default exact or t-Student distribution depending on N, you can choose SAS.
  • If you need 0s to contribute to the analysis, go for Coin package in R or StatXact.
  • If your dataset has ties, go for Coin package in R or StatXact.

Additional References

Details of how to implement the methods in the discussed software are available below:

R:

  • Check how to perform this analysis in R here

SAS & StatXact:

  • Check how to perform this analysis in SAS & StatXact here