[notahacker]

The theory is that a cohort whose uncertainty over a decision is so close to 50/50 that a coin toss has a large statistical effect on whether they make the change or not[1] ought to be close to 50/50 on whether it has a desirable outcome in the event they change (allowing for a degree of risk aversion, and making the fairly standard economic assumptions that humans primarily optimise for an analogue of [future] satisfaction, and their errors in predicting their future satisfaction are random rather than systematically skewed towards underestimating or overestimating their future outcome)

In practice, it's likely the choice of reported happiness over a six month followup period as the indicator of whether it worked is the major factor here; many negative aspects of a decision take much longer to take effect (get even more bored/frustrated with the new job, miss their ex despite the shortcomings in their relationship, realise that going back to school is costing a lot of money and not advancing them in any way)

It's possible that the larger fraction of people who are happy six months after making the change they agonised over is entirely cancelled out if you ask them after 24.

[1]the paper seems to suggest the coin had a statistically significant impact on the sample group's decision making even if their ex ante predictions of their probability of making the change were greater or less than 50%