Cohort analysis seems really powerful, but the implementation is a bit confusing.
Case in point: I want to examine the behavior of my Facebook visitors from February and see if they come back at all over the next 5 months. So simple trigger, follow is just a "visit"
When I apply a visit-based segment (ref. domain = facebook), I get fewer original cohorts than when I make the same segment visitOR based. That seems to indicate that it's counting you in my visitOR group even if you didn't come from Facebook in February but did in one of the subsequent months. That's not what I want. When I use the visit-based version, are my future-month cohort revisits only counted when they, too, are from Facebook referrals? Also not what I want.
I want people who came to me from Facebook in February and returned -- with or without FB-referral -- in the subsequent months.
Here's a look at sample data showing the variance.
With my limited knowledge of Cohort Analysis I am attempting to answer your question.
Firstly between visit and visitor based segments, the behavior that you are seeing is expected because when you make the segment visitor based it includes all the pages the visitor has visited in that time frame. For example, if in one month I visit your site twice, first from facebook and the next time via a bookmark. If you are looking at referring domain = facebook with visit container it would take my data from only the first visit. If you apply visitor container it would take my data for both of the visits.
Now in a Cohort once you specify a group by the inclusion metric you will see the behavior for ONLY that group. Could you tell me your inclusion and return metrics so I could analyse the logic?
Both the inclusion and return metric is a "visit." I would love to make the inclusion metric a "Facebook referred visit" and the return metric a simple return visit from anywhere, but the interface does not allow calculated metrics for the inclusion trigger.
Hence the need for a segment. It seems odd to me that no one has yet found the need to see cohorts of this type/in this way.
It may be the only solution is to enable calculated metrics for inclusion, or at least allow segments to be applied to just the inclusion metric.
Or it may be that I'm missing something in the interface.