|Statistical tests for the Extended option using the Open R API: Welch, Wilcoxon, Limma, Mann-Whitney|
|Which statistical test should I select?|
|The selection of test involves several parameters. Below is a short introduction to the tests that are provided as part of the Extended statistical functionality.|
The Welch t-test is more flexible than the regular (Student's) t-test since the former does not assume equal variances in the two compared groups. It typically performs better than the Student's t-test for normally distributed data with unequal variances in the two groups.
The Wilcoxon rank sum test (also known as the Mann-Whitney U test) is a non-parametric, rank-based test of the null hypothesis that two compared populations are identical. The alternative hypothesis is that one population tends to have larger values than the other. For paired data, the Wilcoxon signed-rank test can be used to test the hypothesis that the median difference between paired values is zero. Since these tests are based only on ranks, they are typically quite robust against outlier values. They also do not assume normality, and hence they may be more efficient than the t-test when data are not normally distributed.
Limma (Smyth, 2004) implements a extension of the linear model, adapted to high-dimensional data sets (developed especially with gene expression data in mind). The method "borrows" information across the large number of variables in order to obtain more reliable estimates of the gene-wise standard errors. Since the test statistic used by Limma is not independent of the gene-wise variances, the data should not be variance filtered before Limma is applied.
|Back to Search Results|