CHAPTER 11 Comparing Average Values between Groups 157
Sometimes you may not know which post-hoc test to select for your ANOVA. If
you have an advisor or you are on a research team, you should discuss which one
is best. However, if you run the one you select and do not trust the results, it’s not
a bad idea to run the other ones and keep track of the results. The p values will
always come out different, but if the interpretation changes — meaning different
comparisons are statistically significant, depending upon what test you choose —
you may want to rethink doing post-hoc tests. This means that the results you are
getting are unstable.
Running nonparametric tests
As a reminder, the Wilcoxon Sum-of-Ranks test is the nonparametric alternative
to the t test, which you can use if your data do not follow a normal distribution.
Like with the t test, you can run a Wilcoxon Sum-of-Ranks test in R with options
that gives you results if you are doing a paired t test. But to simply repeat the
independent t test we did earlier comparing mean fasting glucose in married
NHANES participants compared to all other marital statuses, you would run this
code: wilcox.test(NHANES$LBXGLU ~ NHANES$MARRIED).
The Kruskal-Wallis test is a nonparametric ANOVA alternative. Like the ANOVA,
you can use the Kruskal-Wallis to test whether the mean fasting glucose is equal
in the three-level marital status variable MARITAL. The R code for the Kruskal-
Wallis test is different from the ANOVA code because it does not require you to
produce an object for the summary statistics. The following code prints the results
to the output: kruskal.test(LBXGLU ~ MARITAL, data = NHANES).
Nonparametric tests don’t compare group means or test for a nonzero mean dif-
ference. Rather, they compare group medians, or they deal with ranking the order
of variables and analyze those ranks. Because it this, the output from R and other
programs will likely focus on reporting the p value of the test.
Only use a nonparametric test if you are absolutely sure your data do not qualify
for a parametric test (meaning t test, ANOVA, and others that require a particular
distribution). Parametric tests are more powerful. In the NHANES example, the
data would qualify for a parametric test; we only showed you the code for non-
parametric tests as an example.