You can think of the Jerk Threshold problem as one of a prisoner's dilemma with a (sufficiently large) extra term, which you can think of as some combination of personal guilt, societal punishment, and whatever positive prosocial instincts humans have.
In the classic prisoner's dilemma, your payoff is +3 if you defect and the other prisoner doesn't, +2 if neither of you defects, +1 if both of you defect, and 0 if you don't defect and they do. There's no Jerk Threshold in this problem - regardless of your opponent's behavior, in isolation, you are always better off defecting.
But let's add the extra Antijerk Term - call it A - that you pay to defect. This could be you feeling bad about defecting, or it could represent the chance that you face retaliation for defecting later, whatever - there's some slight cost to being a jerk. We can see how the Jerk Threshold changes.
Your payoff matrix is now [[2, 3-A], [0, 1-A]]. If your opponent has a probability p of defecting, you can compute:
E[cooperate] = p * 0 + (1-p) * 2 = 2 - 2p, with the former term corresponding to you getting screwed and the latter corresponding to cooperation.
E[defect] = p * (1-A) + (1-p) * (3-A) = 3 - 2p - A, with the former term corresponding to the defect-defect state and the latter one being you screwing them.
For A 1, the problem abruptly shifts into a cooperation game, because 2 - 2p > 3 - 2p - A for A > 1.
-----
In this case, the Jerk Threshold abruptly shifts from 0 to 1 (that is, nothing could make you cooperate -> nothing could make you defect). But in the real world, the A term varies depending on context. The A term with a friend is very high, because you have a lot of opportunity for retaliation and they'll feel especially bad screwing you over. The A term with a stranger you're somewhat hostile to is low (maybe even negative). And A varies from actor to actor - some of us have stronger consciences than others. So you end up with bubbles where there's a stable local equilibrium (because the A values are high internally and that maintains stable cooperate-cooperate equilibria) despite differences with the outside world.
Properly modeling this probably looks something like the Ising model [1] on some complicated social graph. Which explains why we see these kinds of phase transitions - most real graphs are dense enough to have them. The bubbles we just described correspond to magnetic domains, and the incentives not to cooperate while in contact with a defect-bubble (or vice-versa) correspond to the high potential energy of domain walls.
[1] https://en.wikipedia.org/wiki/Ising_model