6. Tests for Continuous Data - Comparing Two Groups When Data Are Not Normal

Basic Statistics: Simple Group Comparisons
With Dr Helen Brown, Senior Statistician at The Roslin Institute, December 2015
*Recommended Youtube playback settings for the best viewing experience: 1080p HD
************************************************
Content:
Comparing two groups when data are not normal
-Example histograms for non-normal data:
-Could try a transformation of data
--eg log(x) or log(x+1) if skewed to the left (often suitable for count data)
--recheck the histogram

-Mann-Whitney U-Test:
--Suitable for non-normal data
--Sometimes alternatively called a Wilcoxon Rank Sum Test
--A ‘non-parametric’ test

Mann-Whitney U-Test: Ranking the data
Tests if groups differ based on ranks of data
p=0.06
not quite statistically significant
-Typical output (Minitab)

Mann-Whitney U-Test: General points

Suitable for:
--any continuous data even if non-normal
--comparing data from two independent groups

Ranks take no account of size of differences on original scale
--less sensitive than t-test if data are normally distributed
--t-test preferred if data are suitable

Paired non-normal data
Use Wilcoxon signed rank test
--based on signed ranks of differences between pairs of observations