An experiment with a control group is indeed needed. And yes, it should both evaluate the probability of a divorce per participant, and per married participant.

My estimate... well, it would depend on how much of the Game one knew "naturally" before the seminar. It probably wouldn't work for both extremes -- probably even more for the extreme that couldn't ever get married (and therefore divorced) without the training. :D

For an average person, I would guess that taking the "red pill" does decrease the probability of a divorce by maybe 5%... but I have nothing to support this guess.