interaction effect

In an ANOVA the interaction effect is the measured effect where two or more of the factors have a complex behaviour over and above the main effect of each.

For example, suppose we have measured the following error rates (averaged over many replications, and assuming equal numbers of replication in each cell):

	system A	system B
novice	70	50
expert	10	30

There is an interaction effect of magnitude +/10, whereby novices on system A and experts on system B perform 10 points higher than one might expect based on the main effects for the expertise and system, and novices on system B and experts on sysem A perform 10 points lower than expected.