ANOVA
ANOVA is ANalysis Of VAriance
- it’s a set of statistical models
- they are used to analyze differences between group means and their associated procedures
- e.g. variation among and between groups
Types
One-Way ANOVA F-Test
Goal: compare many means in a single hypothesis
- instead of doing pairwise $t$-test, do ANOVA
- but you can still perform $t$-test or Tukey HSD Test as post-ANOVA analysis
- also, a good way of checking independence between two variables: numerical and categorical
Some authors argue that the ANOVA step is in fact unnecessary and we could perform the Tukey HSD test alone. Nevertheless, the ANOVA + Tukey approach is considered standard is most books.
Not Normal?
If not Normal, use these non-parametric tests
- Wilcoxon-Mann-Whiney Test if the class variable is binary
- Kruskal-Wallis Test for any nominal variable
Links
- http://www.marketingdistillery.com/2014/08/10/multiple-abn-tests-in-marketing-with-anova-and-r/
- Book with chapters about ANOVA [http://vassarstats.net/textbook/toc.html]
- Conceptual Introduction to the Analysis of Variance [http://vassarstats.net/textbook/ch13pt1.html]
- ONE-way analysis for independent samples [http://vassarstats.net/textbook/ch14pt2.html
- The Kruskal-Wallis Test for 3 or More Independent Samples [http://vassarstats.net/textbook/ch14a.html]
- http://www.personality-project.org/r/r.anova.html
Sources
- Data Mining (UFRT)
- OpenIntro Statistics (book)
- http://en.wikipedia.org/wiki/Analysis_of_variance