InfluentialPoints.com
Biology, images, analysis, design...
Use/Abuse Principles How To Related
"It has long been an axiom of mine that the little things are infinitely the most important" (Sherlock Holmes)

 

 

 

Unequal variance ANOVA

Background

Although conventional parametric (and non-parametric) ANOVA require the assumption of equal variances, Welch (1951) developed an unequal variance one-way analysis of variance. This test has been little used to date, although its provision in R (as oneway.test) probably means it will be more heavily used in future.

Subsequent multiple comparison of means should be done using pairwise.t.test with a non-pooled standard deviation. This calculates pairwise comparisons between group levels with Bonferroni-type corrections for multiple testing.

How to do it

Worked example

We will analyse the data from Johnston et al. (2001) using R:

One-way unequal variance ANOVA


Gives something like this:

One-way analysis of means (not assuming equal variances)

data:  con and trt
F = 5.9763, num df = 2.000, denom df = 42.893, p-value = 0.005134

Pairwise comparisons using t tests with non-pooled SD

data:  con and trt

  A      B
B 0.0596 -
C 0.0044 0.1280

P value adjustment method: holm