Youtube seems to have a thing for removing likes or dislikes added on a video by people that didn't watch the most of it (something that is being tossed around as "invalid views"). May be related. More so, YouTube is super vague about its inner workings, so we may have a hard time knowing for sure.

@T.Sar It seems like a somewhat reasonable policy. Definitely not great behaviour, as it gives Youtube pretty extreme control over perceived opinion. At the same time, I can understand wanting to prevent dislike bombing from people who aren't even watching the video and just disliking on principle. Hard to say you really disliked the video, and not just trying to skew opinions, when there's evidence you didn't watch it.

I'm sure you can purchase dislikes and low retention time from click farms as part of a negative SEO attack, just as easily as you can purchase likes and full views from them. If youtube detect such bot spam it's going to moderate them.

@JMac Indeed. I agree that isn't a very nice police, but when you put it on context that youtube is a pretty common target for botting (as RatchedFreak points out), it is a police that I would be glad to have in place were I a content creator.

@Zibbobz I think you might be right if Youtube was claiming they DID NOT remove likes/dislikes ... However since YT is acknowledging the claim as something they DO do I think it's fine.

@Zibbobz: what's the incentive for YouTube to claim they adjust the like/dislike count if they were not doing that? The specifics of their policies are another matter, but the plain answer here just says "yes, they do".

@JMac - If you have seen the video from other sources, you are already informed enough to vote "legitimately" (by YouTube's (arbitrary) standards). No need to watch it. Besides, one can dislike things in general at first glance - meaning there are films which I just need to watch for 2 seconds in order to know I dislike them. It does not matter if my dislike is objective enough - it's similar to the normal discrimination of our choices - if we choose to buy A, we also choose not to buy B, C, D, E, etc.

@Battle You might be informed enough by Youtube's standards; but they would have no way of knowing it, so they don't really have any reason to count the vote by their own methods. It makes some sense to prevent botting and people trying to game the reputation of things for someone else's benefit. I don't really think it's a great system, but it makes sense. A lot of people try to manipulate voting for many reasons, so prevention is also important. You'll likely never find a perfect balance between too relaxed or too strict.

This answer answers the question (so gets my upvote) but doesn't answer the controversial but unasked question of whether YouTube deletes dislikes for political reasons.

It stands to reason they're admitting to removing (dis)likes which, inferring from their Views Policy, are essentially artificial likes, not to censorship (as in: legimate likes they don't approve of). If the question is whether they ever remove any like for any reason whatsoever, then the answer is yes. Although a few of the quotes in the question imply more nefarious intentions, which may indeed exist, but are in no way verified by this quote. You might want to edit to make explicit what exactly this quote is admitting.

(as well as what this quote is not admitting to).

@T.Sar - That seems problematic. If I get about 10% in and hit something offensive to me (perhaps it merely insults my intelligence, which doesn't really justify a flag), I'm damn well not going to sit through 5 more minutes of it to see if it gets better or worse.

@T.E.D. I fully understand you. I'm not saying I like this method of policing likes and dislikes, but I certainly understand why such a policy would be in place.

@T.Sar - Fair enough.

I don't get it. If there exist a concept of "valid" like/dislike according to parameters such as percentage of footage seen they could easily enforce it. Just disable the ability to like/dislike the video before reaching the requirements for validity. No need to allow everyone to cast their like/dislike and then periodically check which one to remove...

@Bakuriu The reason you wouldn't want to do that is the same reason Stackexchange doesn't make it public what will cause you to get question/answer bans. If everyone knows which criteria make one able to cast a vote, or if it is trivially obvious when your vote will be counted; it is far easier to abuse the system without having the abuse reversed. By having this happen in the background, bots and vote manipulators can be reversed without being able to connect any single action with the ability to like/dislike without consequence.

@JMac Apple & oranges. If the requirement is: if you have to have seen at least 50% of the video, I don't see how you can really abuse the system. Sure you can have your big farm of computers but you'll have to waste tons of bandwidth and time to download 50% of all videos before being able to cast the vote. This would hugely increase the cost of this kind of abuse and thus reduce it.

@Bakuriu It's really not apples and oranges. Both sites do not make it clear what is inappropriate in terms of exact formulas or numbers; because doing so would make the process gamable and thus easier to circumvent. What do you think is easier to program bots for, a system where you know that watching a video 50% will be accepted by the system, or a system where there are unknown variables that will determine what constitutes problematic behaviour? It's the same as SE not telling you the number of downvotes/deletions/etc. required before you trigger automatic bans.