I imagine what they do is take people like you who are random, then sample that over an extended period and determine seasonal averages and deviations for the figures, to allow them to generate a best guess as to how many people will cancel for 'random' reasons.

Then everyone else is directly attributed to external influencer's, such as the release of WAR, or internal ones such as expansions and patches and so they can gain approximate data as to the migratory patterns of their customers.