1a) The problem with early [shared] clue is that it allows scum to adapt their claims to them, instead of allowing gotchas after true/false claims.
1b) The problem with late clues [unshared] is that they are individual, and will only drive suspicions (which will have to be proven through other means). And, to some extend, may serve as unjustified accusations ("trust me he's scum, i have reasons to say this, let's lynch him, oops, my bad, misinterpretation but i cannot tell you of what, tee hee").
2a) The problem with early [shared] likes/dislikes is that they may be tools for interpretation of hints both for scum and town (in probably asymetrical ways - potential loss for mafia is greater, so so may be usefulness immediacy).
2b) The problem with later [unshared] likes/dislikes is that they are probably the kind of hints that work when combined together, and not in isolation.
3) And the problem with all that is that unshared infos are not used at all by dead people.
Generally speaking, I don't think that the worse issue is that the info may be used privately. Clumsy early display could benefit mafia just as well (by allowing the construction of compatible lies, by allowing potential mapping with an info headstart), while private hints are still used by players. The worse issue is death, absolute loss of info that hasn't been put in the bank of collective knowledge, and is then not used by anyone at all.
So, there are a lot of distinct, and inter-dependant, parameters to take in account. and I feel that most stances just select one and discard the others. These parameters should be clarified, and people's stances towards each of them individually stated.
I try to map them out in the most readable way. I visualise them as dragbars that make each other move back and forth when we touch one of them. We have to prioritize some aspects, sacrifice others, maximise and minimise combinations. This doesn't necessarily mean dragging one bar all the way up or down.