There's been some confusion regarding the statistics used in the question and its answers, so I'll take a moment to clear those up. In the bottom right of the "summary" subtab of a user's Activity tab, there is a Votes Cast section. Those counts do include votes on deleted posts; relevant MSE. On your own "votes" subtab, you don't see up/downvotes on deleted posts. (You can see closure and deletion votes on now-deleted posts.) The Data Explorer's UpVotes and DownVotes columns in the Users table are the same as the Votes Cast section. I'm not sure that there's a right way or wrong way to measure voting using these figures, but we should be clear about what we're counting.
Voting statistics are inherently difficult to measure because of the necessary anonymity on who cast what, but let's look at some SEDE numbers just for fun. I made this query to show some info on the most active voters. It includes the 50 users with the most votes who have been seen in the last month. The UpPerDown column is the user's upvotes divided by the number of downvotes. It's not possible to get "negative" ratios here, since one cannot cast a negative number of votes. To get some visible symmetry between "more downvotes" and "more upvotes," the PositivityScore column is a logarithm of UpPerDown. For example, someone who upvotes twice as often as they downvote gets a score of ln 2 = about 0.69, while someone who downvotes twice as often as they upvote gets a score of -0.69. Someone who votes up and down equally often gets zero. The CommPerMaybeDv column is my attempt to measure mentoring, but since SEDE doesn't have a lot of info on deleted posts, and because votes are anonymous, there are so many pitfalls here that you shouldn't trust it too much. It's supposed to show the number of comments made by the user on posts they might have downvoted divided by their total downvotes, but again, there's not enough public information to be certain, and deletion continues to skew results. Think of it as an arbitrary mentoring score with a ton of noise.
All data aside, I agree that it's a Good Thing to comment when downvoting. Forcing people to comment when downvoting has been requested many times, but there isn't a good way to make that work. Relevant MSE, one of many.
Voting in both directions is very important, for multiple reasons. Besides rewarding/disciplining post authors, it also signals to future readers what is good and worth reading. A negative score is a much stronger signal in that way than a zero score. Comments with downvotes would help in that department (as warnings of suboptimal practices), but unfortunately, post authors frequently become very upset at commenters that they suspect cast downvotes. Several times, I have been accused of downvoting things I didn't because I tried to help the user. High-rep users who have been around a while and subjected to that a lot could be understandably weary of it.
There are some rare cases where commenting is actively undesirable. I once dealt with a user (on a different site) who repeatedly posted non-answers. Each time, I gently reminded him of the site's requirements, but he became increasingly upset. Eventually I realized that he would not listen, and when I removed his last post with no comment (and therefore no notification), he left of his own accord.
The system as it stands allows people to vote as they please, so long as the votes aren't targeted (e.g. revenge downvoting, sockpuppet upvoting). The site is so large that one voter with unusual preferences will on the whole be counteracted by another with the opposite patterns.
You will be happy to know that, in general, there are far more upvotes in play than downvotes. This tiny query tells us that Super User has more than 10 times as many upvotes as downvotes (not counting votes from deleted users or the automatic downvotes on red-flagged posts). This other query counts the votes on posts by new-ish users, those with less than 500 reputation. The upvote proportion is even more extreme here: almost 12x as many positive votes as negative ones. Of course, that doesn't include deleted posts, but it's still interesting.
More relevant MSEs: