The question implies that you can't see a refutation, but it doesn't presume there is one to see in the first place.
So, the question really becomes: If I can't see a refutation, what are the chances my opponent will find one?
Well, clearly, if there isn't one, he won't find it. But there are several ways he could find one if it's there.
- He could see it as soon as you play your move.
- He could find it as the position evolves, and you play out your combination.
Of course, #2 is much easier, since the pieces are moving into position, and if there is a refutation, it may become much more obvious as it approaches.
So, #2 represents the real risk. If you can visualize the position after you make what you believe is the last move of the combination (mating, or winning material), then sit still and figure out all of the moves he could make that might affect you in any way. For each one, you need to know that you have an adequate reply.
If you can't visualize that position clearly enough to identify those moves, then you can't determine whether the move is safe to play. In that case, I would advise that you not play it unless your position is desperate.
By all means try to confound a superior opponent; for one thing, it will burn time on his clock.
If you are a creative and resourceful player, it may be worthwhile to play a sacrifice that offers you some positional advantages, even if it may have a hidden refutation. Whether you should do this or not depends on how well you generally convert such advantages into wins.
Mikhail Tal was famous for his tendency to attempt sacrificial combinations that were too complex to calculate fully; he depended on his ability to improvise as the position evolved, and find new resources.
Karpov, on the other hand, while a very competent tactician, would never take such a risk. He would only play the combination if he could visualize the final position and determine that the opponent had no meaningful threats.
So, it's a matter of style, to a large degree. Build your tactical vision by playing through games, 4-5 moves per side at a time, without moving the pieces. Visualize the board, and try to predict the next move. Then, bring the position on the board up to the current move, and check how well you visualized the position. Finally, see if you predicted the next move correctly.
If you can't visualize the board reliably 90% of the time, reduce the number of moves and try again. If you can, try stretching yourself to increase it. If you can handle 6-7 moves per side, you're doing pretty well.