15. ANOVA: analysis of variance

Summary

In practice, comparing means (through statistical tests) usually involves more than two categories, groups and/or variables. Consider, for example, an experiment with 6 different treatments: the outcome of each will vary to a greater or lesser extent, and the researcher will try to understand that variation, concluding whether one or more of them really differ from the rest in their effects. The analysis of that variation is precisely what is known as ANOVA (analysis of variance).

The technique is very simple in essence. It compares the variation within each group with the variation between groups. If the latter is considerably large (and the former considerably small), it can be said that the groups are different. That said, depending on whether different variables (factors) that affect the outcome are to be tested, ANOVA as such can become quite complex. Whatever the scenario, it is very important to keep in mind what introducing more components into the analysis implies, a key consideration for drawing valid conclusions. An important case in this respect is the use of blocking factors, which are deliberate arrangements in the experiment made to control the variation of the response.

After the conclusions of the ANOVA, other procedures are usually carried out to determine specifically which treatment / group differs from the rest. This is what is known as post hoc analysis.

ANOVA, being a statistical test, is a procedure widely used in scientific research; but it is worth stressing that it should not be seen simply as a test, but as a way of estimating a continuous variable as a function of one or more categorical variables, which is broadly called a regression (a topic covered in the next chapter). It is also important to note that analysis of variance, as presented here, is a parametric method, with a nonparametric counterpart for cases where the assumptions are not met (or where a parameter other than the mean is investigated).

Further reading