What seems like such a simple question unfortunately is probably quite difficult to answer. At best you could find someone who ran a high quality A-B test where the result showed something interesting. But you still have no idea whether that changed over time after the A-B test finished.

I think the main question is was the number of refund requests significant? If it was trivial then any marginal benefit is very low risk.

Patio11 has written quite a bit about his experience. Worth having a read of his blog if you haven't already.

https://www.kalzumeus.com/archive/