# Incentive for downvoting?

I'm fairly new to this site, so I'm somewhat confused by the idea behind downvoting, especially considering that one gives up some of one's own reputation to downvote something else. What incentive does one have to downvote something instead of simply flagging the post and moving on?

• "This answer is not useful." is a reason for downvoting, but not for flagging. – Michael Greinecker Jun 11 '14 at 13:16
• IIRC you don't lose reputation when downvoting a question. (Only for answers.) – Martin Sleziak Jun 11 '14 at 13:20
• To expand somewhat on @MichaelGreinecker's comment, by flagging a post that shouldn't be flagged, you are wasting other peoples' time. (Quite probably mine!) – user642796 Jun 11 '14 at 13:37
• In addition, I like being able to reliably tell which answers and questions are well writte, good, and interesting. This relies on active up voting and downvoting by the community. – davidlowryduda Jun 11 '14 at 13:48
• math.stackexchange.com/help/why-vote – user147263 Jun 11 '14 at 14:51
• Some of the rhetoric at the why-vote page might be familiar from the old country. Never mind the comedy, though. I was wondering, did SE ever explain the reason for 1k reputation barrier to seeing the downvotes? If it is so important that everyone cast them, why hide them from the majority of users? @wordsthatendinGRY – zyx Jun 11 '14 at 16:25
• @zyx I'm only speculating here, but one of the reasons might be to reduce the amount of long arguments with people who haven't yet shown themselves to be serious contributors. – Michael Greinecker Jun 11 '14 at 18:58
• Because you too could enjoy your 15 minutes of fame, before changing your name and going on to become a rather useful meta-person? – user1729 Jun 12 '14 at 8:30
• Note that the -2 reputation you get for downvoting is really a very tiny amount. I suppose it matters more when you have close to 0 reputation, but that's a good thing, it means your incentivized not to downvote until you're more experienced. – Jack M Jun 19 '14 at 17:44
• ... to make the world a better place? – Hagen von Eitzen Jun 25 '14 at 16:53

I'm leaving aside the "simply flagging" part, which probably stems from unfamiliarity with how flagging works.

Downvotes are major contributors to both automatic deletion of posts, and to their manual deletion. Apart from software-forced deletions, downvoting gives authors an incentive to remove the post: partly to recover the loss of reputation, but mostly to get rid of the negative number. This applies to both questions and answers.

Some statistics on deletion: During February 2013-February 2014 period, $\textbf{21419}$ questions were deleted on this site. Of them,

• $10663$ were deleted by their authors (usually, after being downvoted and/or closed),
• $10028$ were deleted automatically (based on low score among other things),
• $378$ were deleted by moderators
• $218$ by $10$K users
• $102$ by spam/offensive flags

So, as far as deletions are concerned, downvotes are by far more effective than flags or explicit votes to delete.

I also have some statistics on downvotes that did not lead to deletion. More than 40% of them had visible effect, dropping the question score to negative. Specifically:

• there were $36804$ downvotes on non-deleted questions, $<0.13$ per question. Of these downvotes,
• $41.38\%$ were on questions that currently have negative score
• $21.52\%$ ..... zero score
• $12.38\%$ ..... score $1$
• $24.71\%$ ..... score $2$ or higher.

Let's look at the answers next.

• there were $20597$ downvotes on non-deleted answers, $<0.05$ per answer. Of these downvotes,
• $31.67\%$ were on answers that currently have negative score
• $18.07\%$ ..... zero score
• $13.32\%$ ..... score $1$
• $36.94\%$ ..... score $2$ or higher.

Thus, nearly $50\%$ of answer downvotes had visible effect, dropping the score to either negative, or zero (a zero-score answer allows the question to remain in Unanswered queue).

• A transmission whose bits are randomly corrupted more than 50 percent of the time is indistinguishable from noise. There will be segments of the message that by pure chance are sent correctly because the bit errors happen to accidentally miss them. In an environment with small number of votes and page views, that would explain a lot of your data. Is there a search that can disclose the distribution of (N_up, N_down) pairs so as to test whether the "effectiveness" of downvoting is different from what is predicted by a random noise model? – zyx Jun 12 '14 at 1:49

Downvotes have some practical implication for which questions appear on the "front page" and for automatic deletion of negative-score unanswered questions.

As a form of communication, the current use of downvotes is hard to distinguish from random noise. There is no consistency between users in how frequently the votes are used, what they are intended to mean, how strong a message is meant to be conveyed, who is the target of the message, or the range of situations in which such messages would be sent. Increasing the confusion, the sample size is also small. Downvotes are used rarely compared to upvotes, the number on questions is low and on answers it is minimal (rarely more than 1 or 2 even for clearly wrong answers). The end result is that

effectively the only message sent by most downvotes is "one of the users who likes to downvote has visited and saw something (be it content or a username) that they did not like".

Most users may not receive any message at all. Users below 1000 reputation points cannot see the number of downvotes, and those above 1k have to investigate to get the information. Except for posts whose vote total becomes negative (which is difficult given the predominance of upvotes), downvotes are a mostly invisible game played between the more active users. The authors of the question/answer are notified of downvotes but the external quality metric that downvoting is supposed to provide, is concealed except through its effect on total scores.

I do not downvote on the main site so as not to increase the noise, among other reasons. If the system were modified to increase the volume, consistency and visibility of downvoting then it could become more useful (as it is on the meta site, where down votes are free and frequent and easily interpreted as "disagree"). As currently used and abused, the costs range from puzzled "why the downvote" comment discussions to covert voting wars between users, and some more blatant targeting of particular individuals. The benefits are random, with an expected value very close to zero.

• Upvotes are also quite random. About the only thing they seem to be correlated with is the speed of the answer, which is surely not the primary attribute that should be used to help rank good answers. The design of the SE rep/voting system is highly flawed (not only for these reasons). – Bill Dubuque Jun 11 '14 at 17:33
• @BillDubuque That upvotes are given to fast answers have nothing to do with the design of the SE reputation/voting system. It is rather due to the attitude of some users. – Hakim Jun 11 '14 at 18:13
• @حكيمالفيلسوفالضائع It certainly does have to do with the design of the system, because a better designed system could go a long way towards eliminating this flaw. – Bill Dubuque Jun 11 '14 at 18:16
• Upvotes have problems too, but the larger volume (and consequently, visibility) and zero cost turn them into something at least correlated with quality or popularity. Clearly a farcical number in many cases, but they say a vague something while downvotes are just pure noise in the current environment. @BillDubuque – zyx Jun 11 '14 at 18:49
• @zyx More often than not the best answers have fewer votes than the quickest, so votes are mostly useless for determining quality (in my experience). Cherry picking is a large problem, caused by the gamification principles at the heart of the SE model. It has caused some users to stop at nothing to accumulate "rep" (e.g. plagiarism). A better-designed system would, instead, motivate users to take their time to compose pedagogical gems - something that, alas, is rare on MSE, since the majority is hooked on the MSE game (as per design). – Bill Dubuque Jun 11 '14 at 19:19
• I agree with all that you say, and have said similar things many times here. My point is only that upvotes have limited meaning but downvotes are so random as to have almost no meaning at all. That's consistent with particular individuals being coherent, meaningful and nonrandom in their personal application of downvotes. The lack of inter-user correlation in the downvotes turns it into noise when combined with the rarity and invisibility, and the different thresholds the rep hunters may have for giving up 2 points, or the polite users for communicating negatively. @BillDubuque – zyx Jun 11 '14 at 19:24
• Interesting. I, by contrast, find downvotes on meta relatively more "noise like" than on main. Especially, since more often than not I find "I disagree" as a "noise like" feedback. Few posts are as clear cut as that there would be only one way or one direction in which one could disagree, and so I often found myself scratching my head in which way the disagreement was intended. And not knowing this, makes the interpretation pretty arbitrary. – quid Jun 11 '14 at 23:13
• @quid, on meta it is possible for a large enough number of downvotes to send information, as they are numerically comparable to, or outnumber, upvotes (and "7 people disagree in at least one of the 3 directions" is still something). The signal could be aggressive or political, it may be nothing more than an unpopularity vote against the author of the post, or a massive downvoting of brilliant proposals, but even in those cases one can guess OK what it means. On the main site, the random "error term" in amount of upvoting on a post greatly dominates the sum of the downvotes. – zyx Jun 12 '14 at 0:30
• Throwing away accuracy for the sake of precision is a peculiar method for dealing with noise.... – user14972 Jun 12 '14 at 5:11