On a meta site, votes are intended to imply agreement with the question/answer. In contrast, votes are intended to imply quality, on the general site. Overall, when we cast a vote, we have a brief window to reverse or withdraw that vote. Once this window has passed, our votes are locked in, unless an edit is applied to that post.
This makes sense, in general. When we judge a post for it's quality, there is little reason1 to want to legitimately withdraw or reverse this vote, unless the post has had changes made to it. Without change, the quality of the answer is also unchanged.
In contrast, opinion does change. When I vote on meta, I am voicing agreement or disagreement with the post. I think it is a lot more plausible that this stance could change, without an edit to the original post.
Here is a good example of one such case. I asked a question regarding a particular tag that needed more clarity in it's usage. A user suggested an alternate to my own proposed solution, and I agreed that it was a better proposal. I deleted my original answer, and up-voted the new answer. Some time after, a user with more experience posted an alternate answer. With the clarity of the explanation provided in the new answer, I no longer agreed with the previous answer. However, I am still technically voicing agreement in the answer, as I can not undo my vote.
In some cases, this may appear as a "non-event". One vote should not make a difference. However, in cases where meta-usage is not as common, I feel it does make a significant impact on the overall view of the topic or suggestion.
I have previously been directed to simply perform an edit; I feel this is not a suitable option. Especially in certain cases where such edits would be semantic and questionable.
Can we have a bit more freedom in retracting and/or reversing votes in Meta?
It occurred to me that there might be further issues if this was implemented on the main meta site (i.e. this one), as reputation is awarded. I am asking specifically for site-specific meta, though I don't see why we couldn't discuss implementing this, here.