Let's Remove The Auto Downvote On Validated VLQ Flags (On Questions)

Question

Currently, flagging a post as very low quality causes two things to happen:

1. The post is flagged for moderator review
2. The community user casts a downvote on the post if the flag is validated

Optionally, the flagger can also cast a downvote. Some strange magic happens if the post is subsequently improved, but the whole process is a kludgey work-around that was put into place when we had an issue surrounding folks neglecting to vote (primarily on questions).

In order for voting to remain a useful indicator of a post's relevancy and on-topic standing, we need to try to avoid pile-on signal where possible. VLQ flags, effectively, let users cast two downvotes on a question, and that can be pretty brutal to new users. Since we've introduced rolling rate limits and better anti-recidivism systems, we don't really need auto-voting on suspected low quality posts; the organic votes cast by carbon life forms do a fine job.

Too frequently, the only discernible difference between a question with a score of -2 and another with a score of -12 is the absence of 10 people that didn't pile on, if you compare lots of similar posts with wildly varying scores. This constant source of strangeness actively inhibits research, or even visually telling the difference between not good and really bad.

It goes without saying that the pile-on effect is pretty crushing to new users, too; it's mean, it's unnecessary, and it's actively unhelpful.

We'll still auto-downvote SPAM, there won't be any major changes there, what we want to stop doing is automatically tacking on a down vote that was designed to solve a problem that we pretty frankly no longer have, and it's not because auto-downvoting worked so well :)

We're putting this out early for discussion, in case there's an argument compelling enough to reevaluate how useful the system might remain, but we're pretty confident that it can fade off into robo-history pretty uneventfully.

Voting is always going to be an important part of how the site works, but we've got to look at a way where we can encourage more consistent application - reeling in the robots is one way to start, so we'd like to start there.

Questions? Concerns? Should we print commemorative bad robot stickers? Let us know.

---

Update (6/13/2018)

This has been completed for questions only. We're going to look at the answers part of this soon, which is going to require some more work. Title updated to reflect that, too.

Answer 1

Since it's pretty much my fault that VLQ comes with an automatic downvote to begin with, I feel like I should weigh in on this.

I feel very strongly that we should not be removing posts we aren't even comfortable slapping a downvote on.

For all the hand-wringing about downvotes over the years, at least they're transparent: if your question or answer gets downvoted, you know it. Everyone knows it. There's no hidden tribunal deciding the score of your post without your knowledge; every vote is visible.

Deletion is not so transparent. We've made some improvements to that over the years - you can at least view your deleted posts now - but it still feels like gaslighting sometimes; you thought you wrote something, but it's not there on your profile...

None of this is an argument for automatic downvotes though; as long as folks are good about only flagging and deleting things that they're also ok with downvoting, we don't need this system. So... Are they?

Downvotes on VLQ-flagged posts

Once upon a time, we had automatic downvotes for certain close reasons too. We removed those when they were no longer necessary - when > 70% of closed questions got downvoted the old fashioned way.

So let's take a quick look at the numbers for VLQ:

PostType	Helpful VLQ Posts	% Helpful Downvoted	Deleted	% Deleted Downvoted
Question	2297	89.20	1767	92.30
Answer	3394	45.96	2845	51.35

That's data for Stack Overflow, taken from a 30-day period starting 30 days ago (to allow the dust to settle). I'm not gonna do a fancy graph thing, but... Back when I originally proposed this solution, the numbers were about 60% for questions and 40% for answers. Folks have gotten a LOT better about downvoting questions. They have NOT gotten much better about downvoting answers.

The good news is... We don't need this for questions. We can turn it off; less unnecessary pile-on, essentially the same signal.

The bad news is... We still haven't fixed the problem when it comes to answers.

But never fear! I have two potential solutions: a cheap simple one, and a... much more elaborate one.

Cheap solution: just make the flagger downvote

Funny thing is, I didn't originally propose having the system downvote posts. That was a concession because waffles thought my original proposal was too ridiculous to be serious. But I still like my original proposal:

Silently convert them to down-votes. Complete with the normal rep-deduction for the flagger/voter.

We should listen to 2011 me. That guy was quoting poetry & stuff; he sounds like a pretty savvy dude, not at all broken down and sick of life and drunk on cheap whisky. We have real-time rep recalcs now, which we didn't in 2011 - so you get the rep back for an answer downvote as soon as the answer is deleted; your downvote on a VLQ answer is a wager that it is, indeed, nasty enough to be removed.

So here's what I'm thinking, for answers only:

• If the flagger hasn't already downvoted the answer, the system casts a vote for them. They can retract it if they want, subject to normal vote-locking rules.
• If the flagger has already downvoted, then-extraneous text about downvotes is removed.
• If the flagger has used all of their votes for the day, the option is disabled.
• If the flagger doesn't have the downvote privilege (< 125 rep), then anonymous feedback is recorded instead.

Advantages of this solution:

• It's cheap to implement
• Directly couples the thing we want to encourage instead of patching around it like the current behavior.
• Removes the double-downvote pile-on bonus for folks who downvote and flag.

Disadvantages:

• Some folks were probably using this flag as a cheap way to get around downvoting.
  ...I don't really consider that a disadvantage, but some folks'll probably see it that way.

Elaborate solution: remove VLQ flags on answers entirely, replace with fake "wrong answer" flag

We don't actually need VLQ flags on answers; they're not used that much, and for "cat on keyboard" type stuff I've been encouraging the use of abusive flags on those for years now anyway.

Yes, it'd be a pain to strip out, and there's not a lot to be gained from doing so just for the heck of it... But what if we took this opportunity to kill two birds with one stone?

^{(Hat-tip to the inimitable jmac for that mockup - I miss you!)}

This is a bit less forceful than the cheap option above, but it still offers immediate encouragement to downvote - just now instead of flagging for stuff that's not horribad. And, leaves NAA for stuff that isn't even recognizable as an answer, and abusive for stuff that's straight-up eyeball-searing garbage.

Advantages of this solution

• Doesn't add further complexity to how VLQ behaves
• Potentially helps reduce spurious Not an Answer flags by providing an alternative for folks who are just upset about wrong answers.
• Potentially reduces the number of extra "in need of moderator attention" flags generated by the removal of VLQ.
• Encourages downvoting in a slightly less forceful manner than the cheap solution.

Disadvantages

• There's... Kind of a lot of stuff that assumes VLQ exists on answers right now. Audits, VLQ queue, help center, tons and tons of meta posts... It'd take some time and work to clean all that up.
• Folks may not like the idea of an always-disabled flag, sitting there, mocking them.
• The rude/abusive flag comes with a MUCH bigger penalty than VLQ does.
  ...I don't actually consider that last one a disadvantage, but I'm sure someone will.

That other thing that I don't really want to talk about but deserves mention

There's an automated flag that's raised on posts which trip various quality heuristics. It behaves pretty much identically to VLQ, right down to the automated votes. Just FYI.

Answer 2

This is particularly "mean" to the new user because s/he normally assumes that the downvote came from a person. Thus the system's behavior is both confusing to the new user and makes the community look "meaner" than it actually is.

Generally I challenge the presumption that downvotes are "mean." But when they are applied automatically by the system, often times without the intention of the flaggers or the flag validators, it seems to me that the system, if not "mean," could at least be nicer.

Answer 3

Further, the auto-downvote can do the wrong thing sometimes. When I was starting out as a moderator a general piece of advice I received was: mark flags helpful if they were helpful, even if you didn't take the implied action. I remember handling a VLQ flag where the case was borderline; it was legitimate for the user to ask moderators to take a look. So I marked it helpful but didn't delete the post. Imagine my surprise when the post got an automatic downvote right after I validated the flag. I know, because it's come up in TL, that I'm not the only moderator who's been surprised by that. The automatic downvote is unexpected, unnecessary, and occasionally unwanted. Let's reserve votes for human beings.

The automatic downvote for spam or offensive posts serves a different purpose (for questions); when a question reaches -4 it drops off of the front page. We want to limit the visibility of posts that have been identified as probable spam. What we really want is for posts with a few red flags to drop out of view; the only implementation we have right now is votes. If we can alter visibility without creating non-human votes, that would be better. It's possible that now, with Charcoal, we delete these posts quickly enough that this concern is no longer a big deal; I'd want to see a separate meta question where we could discuss it before getting rid of spam downvotes as part of this question about VLQ downvotes.

Answer 4

Well, on a healthy site we have quite a few other mechanisms that handle low quality posts - and I think leaving voting to carbon based lifeforms feels... less hacky. We should be guiding not punishing.

I've often told folks the point of a flag is to have someone look at a potential problem - and if there's folks doing review or even seeing it via the mod queue, the ability/choice to vote or better yet guide probably does more good than a extra downvote, unless we want a quick deletion by our users.

The downvote doesn't help in any way, especially if there's a new user unused to the more... byzantine intricacies of SE - they won't even know its automatic

I'd actually go a little further and wonder if we need to auto-downvote spam. The existing mechanisms of hitting them with a ton of flags, organic and mechanised, and hitting them with a rep penalty (and hopefully nuking the user) does the same.