# Is there an actual “pity” or “sympathy” upvote problem?

There have been a number of discussions about "pity" or "sympathy" upvotes.

http://meta.stackexchange.com/search?q=pity+sympathy+upvote

That is, the idea that once a post is voted down to -1, some kind-hearted user will come by and upvote the post, no matter how terrible or wrong it may be, to return it back to zero.

Does the actual voting data actually support this theory -- or is it all in peoples' heads?

-
Even if there is, it is being more than offset by crowd mentality. The majority of people aren't too kind-hearted. –  Kobi Jun 1 '11 at 4:09
+1, even though this question sucks –  Mike Pennington Jun 1 '11 at 4:21
I've seen users admit that they regularly pity-vote, but I don't have concrete data beyond that. –  Anna Lear Jun 1 '11 at 4:27
–  waffles Jun 1 '11 at 4:27
I see it on the PHP tags. Terrible questions/answers are upvoted to become neutral. But weirdly, it's only if it has -1. Anything below that doesn't garner much in the way of pity votes. –  JohnP Jun 1 '11 at 4:43
Subjective title is subjective. The voting pattern exists. But it's hardly a problem. -- And yes, I've recently picked up that habit. While I was very reluctant to offset someone elses downvote in the past, I feel obliged to give pity upvotes with some of the recently caused downvote frenzies. (separate topic..) –  mario Jun 1 '11 at 4:47
@mario: WTF? Pity upvotes should never be given no matter what the reasons are. I'd say it actually is a big problem if people start feeling obliged to do it... –  fretje Jun 1 '11 at 8:59
I've upvoted some answers with negative votes motivated out of pity but only in situations when the answer was reasonable and the down votes didn't describe the reason for the downvote. –  jpierson Jun 1 '11 at 9:59
Pity, sympathy? What are these things? –  Grant Thomas Jun 1 '11 at 12:15
@Jon: probably nothing - there are plenty of good reasons to up-vote something that's already been down-voted, so without some magical ability to judge the intent of the voter... But really, this applies to all votes - past discussions have shown that folks hold radically different opinions of when and why up- and down-votes should be cast. IMHO, the only thing you can reasonably do - should a problem exist - is try to mitigate any unintended consequences of it, as we've slowly been doing (consider a certain user who at one time had ~10K from asking hundreds of awful questions...) –  Shog9 Jun 1 '11 at 17:49
Pity votes are far less of a problem than the sheer number of votes given to questions that don't actually help anyone. Every vote on fascinoma questions like this one are a "pity" vote. –  user7116 Jun 1 '11 at 21:35
@mario: You're not offsetting anyone's vote. You're rewarding someone. An upvote is worth +10, whereas a downvote is worth a mere -2. Each time you "pity upvote," you're not undoing something; you're increasing that user's reputation by 8. Ultimately, that's the real problem: pity upvotes increase reputation, rather than nullifying the loss. –  Borror0 Jun 3 '11 at 17:02
@mario: Let me get this straight: you purposefully increase the reputation of bad users? What makes you think your voting pattern is somehow defensible? –  Borror0 Jun 3 '11 at 17:43
Is there a problem? I don't know if I'd call it a "problem" but so far, every single question that I've downvoted for being poor "questions" and was the first vote on was promptly followed up by an upvote within a minute. And most of them that I've noticed never received any other votes afterwards, up or down. I'd say it is an annoyance however as it's the type of question that you would ever upvote otherwise, and the lack of additional upvotes show that. So much for doing my part in trying to bring those questions down. –  Jeff Mercado Jun 5 '11 at 6:09
@JeffA: Perhaps and you have a point. However considering the actual volume of questions and comparing the poor questions that have been pity voted versus legitimate questions, the data will always look insignificant. I've watched this reposted question go all the way down to -6 within about 30 minutes. Yet somehow, two people (as of now) felt it necessary to upvote it. Those are pity upvotes if I ever saw them and this particular situation is not counted in your statistics. –  Jeff Mercado Jun 6 '11 at 21:20

Some baseline statistics.

Stack Overflow

• Questions: 1,681,746
• Questions with zero score: 778,511 (46%)
• Questions with at least one downvote: 88,874 (5.3%)
• Questions with zero score and at least one downvote: 26,441 (3.4%)
• Questions with zero score having exactly two votes of +1, -1: 16,730 (1%)
• Questions with zero score having exactly two votes in order of -1 then +1: 9,873 (0.6%)
• Answers with zero score: 1,459,773 (37%)
• Answers with at least one downvote: 250,477 (6.5%)
• Answers with zero score and at least one downvote: 61,654 (1.6%)
• Answers with zero score having exactly two votes of +1, -1: 44,238 (1.1%)
• Answers with zero score having exactly two votes in order of -1 then +1: 17,005 (0.4%)

Server Fault

• Questions: 78,768
• Questions with zero score: 37,522 (47%)
• Questions with at least one downvote: 3,090 (3.9%)
• Questions with zero score and at least one downvote: 1,015 (2.7%)
• Questions with zero score having exactly two votes of +1, -1: 566 (0.7%)
• Questions with zero score having exactly two votes in order of -1 then +1: 313 (0.4%)

Super User

• Questions: 82,315
• Questions with zero score: 33,121 (40%)
• Questions with at least one downvote: 4,260 (5.1%)
• Questions with zero score and at least one downvote: 1,267 (3.8%)
• Questions with zero score having exactly two votes of +1, -1: 564 (0.7%)
• Questions with zero score having exactly two votes in order of -1 then +1: 245 (0.4%)

I think the most telling statistic is the questions with exactly two votes that are at score zero. How many of these are in the order downvote, then upvote?

• SO: 59% Q / 38% A
• SU: 43%
• SF: 55%

Where "flip a coin" is 50%.

Based on this data, "sympathy" or "pity" upvoting is not an actual problem -- if it was, I would expect to see much higher rates of an upvote following a downvote.

-
Does this statistic include closed questions? –  Paul McMillan Jun 1 '11 at 4:09
@paul this is all non-deleted questions. –  Jeff Atwood Jun 1 '11 at 4:48
I'd say the 59% down-then-up is a telling statistic. You might be able to throw out SU and SF, as the sample sizes are considerably smaller, but 59% on SO seems to indicate that there is, in fact, some pity voting happening. It's almost a 20-point spread. If it was not happening in great numbers, you'd expect the percentage to gravitate towards 50% with a larger sample, would you not? Problem or not, it appears to lend credence towards the anecdotal observations. –  Anthony Pegram Jun 1 '11 at 5:01
But how many of those are -1, edit, +1? Surely upvoting due to an improved question doesn't count as pity, does it? –  Gabe Jun 1 '11 at 5:12
Could you also post stats on closed questions and break them out by percentages of upvotes / downvotes? –  Mike Pennington Jun 1 '11 at 5:12
@Gabe good point, but since the data suggests it's not a problem even with counting the edited posts, then it certainly won't be a problem when they are removed from the stats. –  Adam Davis Jun 1 '11 at 5:34
Well, this won't include the sort of question where everyone agrees the question is bad and it gets downvoted to -5, yet when you split the vote you see someone upvoting it, for some reason. Obviously it could be someone working towards the electorate badge, but I think the criteria can be expanded to include these too. –  Yi Jiang Jun 1 '11 at 9:14
@Adam: Anthony is quite right that 59% is a lot more than 50%, in particular since the sample size is more than 16,000. To make it more obvious: 59% is 44 percent more than 41%! Gabe's point was that this might be explained by the sequence "downvote - edit - upvote". So it would be interesting to see only those 1-1=0 questions that have no edits (or no edits between the votes). –  Hendrik Vogt Jun 1 '11 at 12:40
@YiJiang The Electorate badge doesn't require upvotes. Although it is a common misconception that it does so you may be on to something there. –  Anna Lear Jun 1 '11 at 13:16
@Jeff What did you use to gather these stats? I'd love to see the spread for Programmers. –  Anna Lear Jun 1 '11 at 13:16
Notes on the significance of these figures (as compared to a null hypothesis of randomness, and assuming counting statistics) added as an answer. –  dmckee Jun 1 '11 at 14:02
@Jeff - something else to consider is the opposite effect of "corrective downvote". Sometimes I do this as well - I see an answer that isn't right but isn't wrong enough to DV. If it gets upvoted, though, I will DV to cancel it out. –  JNK Jun 1 '11 at 14:12
This is good stuff, but... I'm not sure I see the need for a "0 score" criteria. In these stats, if a post gets 2 down-votes and then one up-vote, it's not counted; it probably should be. –  Shog9 Jun 1 '11 at 17:58
Yeah, I'm failing to see the reason for finding posts with a 0 score only. Surely it would make more sense for the criteria to be no votes followed by one or more down votes followed by an up vote? This is one such case; -5 followed by +2. I'm not sure pity voting is a real problem, but it's certainly annoying when you see it happen. –  Andy E Jun 1 '11 at 22:39
Jeff, comparing these statistics may not be as good indicator for the non-existence of counter-voting as they seem. In contrast one can interpret these numbers in the reverse direction: I don't have the distribution of the votes, but it seems that the number of counter-votes is considerable higher than the probability of them (I think the probability of a question receiving a -1,+1 should be way smaller than the probability of receiving no votes). In addition these numbers seem to imply that there is significant number of anti-sympathy counter-votes: -1,+1. –  Kaveh Jan 22 '13 at 2:20

From the Stack Overflow list:

Answers with zero score having exactly two votes of +1, -1: 44,238 (1.1%)

Answers with zero score having exactly two votes in order of -1 then +1: 17,005 (0.4%)

is statistically significant if we simply assume counting statistics

sqrt(44237) = 210
sqrt(17005) = 130

(44237 \pm 210)/2 - (17005 \pm 130) = 5113 \pm 167


Of course it is still 5000/3.8e6 so it is a small correction, and it go against the direction of pity voting.

Following that model for all the above data

                             Sign   Significance
------------------------------------------------
Stack Overflow   Questions    -      13
Server Fault     Questions    -       0.7
Super User       Questions    +       0.5


The interpretation of the table is as follows:

• a pattern of pity voting would generate a negative value for sign
• significance less than ~3 isn't evidence for anything
• significance more than ~5 is evidence for something, but doesn't prove what.

Both stack overflow results show large significance, but of different signs. The Server Fault and Super User data show no significance.

Not sure what to make of that.

I have developed a half-assed theory (pure speculation, really) that could account for the difference between questions and answer. It may be the question-posters are (a) likely to vote for any answer they received and (b) likely to do so soon after it appears.

If there is anything to that it would mean that some fraction of the (+1,-1) answers are the result of people trying to de-emphasize so-so answers that the OP voted for.

-
This is all significance against 50/50 anyway, which assumes that randomness actually has something to with voting decisions. I'm not so sure that's a good assumption. –  NickC Jun 1 '11 at 14:13
@Renesis: As I am only looking at those events that are (+1,-1) the only thing at issue is the order those votes arrive in. Why wouldn't random be the correct null hypothesis? –  dmckee Jun 1 '11 at 14:56
Because the set is limited to questions with exactly 2 votes but all questions are not fixed at 2 votes, it assumes that +1 and -1 questions are equally likely to get a second vote, and equally likely to not get a third vote. –  NickC Jun 1 '11 at 17:10
The analysis set is not limited to questions with two votes: it is limited to question with one up and one down. There is no a priori reason to expect those to be dominated by up first or down first, so the null hypothesis is that the situation is random. –  dmckee Jun 1 '11 at 19:41
Well... I usually only downvote if the answer already has a vote (- or +) because of the known problem of sympathy upvotes! Anyway, I agree with the 50/50 null hypothesis, but I think it is impossible to interpret the reasons that it fails from this data alone. I would argue, however, that this shows that previous votes influence behaviour. Is this desirable? I don't think so but I cannot think of a good way around it. –  z7sg Ѫ Jun 8 '11 at 17:15
@z7sg: Correlation data certainly can't tell you what the cause was, but a lack of correlation is pretty strong evidence of a lack of causation. The really interesting thing about this data is the different highly significant signs for questions and answers on Stack Overflow. What is up with that? –  dmckee Jun 8 '11 at 18:00
@dmckee Hard to say of course! One difference is that it doesn't cost any rep to vote down questions. But for me at least, that doesn't influence my behaviour, it is more related to what other people are voting. It would be interesting to see how people voted if they only knew the order and not the number of votes for each answer. –  z7sg Ѫ Jun 8 '11 at 18:11
From personal experience, I know that I give very quick up votes on answers to my questions that are at all helpful. Since the asker is almost always the first person to see an answer and understand the context, I think your theory passes the smell test. Later answers can transform mildly helpful answers into distracting answers. –  Jon Ericson Jun 5 '12 at 16:10

The wide spread belief that “pity upvoting” is common changes how some people votes, therefore the belief that “pity upvoting” is common is a problem regardless of it’s truth.

-
I think people remember a pity upvote like they remember a slap in the face. It stings and it will always be remembered as an offense, an affront, a slight against human dignity when a crap post at -1 suddenly goes to 0. Perhaps that's why they perceive it as common, but the data does not support this. –  Jeff Atwood Jun 1 '11 at 8:32
What is "wide spread"? Wide spread amongst voters on Stack Overflow? Or wide spread amongst vocal users on Meta Stack Overflow? –  Jan Fabry Jun 1 '11 at 9:22
@Jan, we don't know that! –  Ian Ringrose Jun 1 '11 at 10:52

Regarding the -1 then +1 pattern: I will sometimes upvote a mediocre but not harmful/wrong answer if it is at -1, where I wouldn't have upvoted it if it were at zero.

However, I do not consider this "pity" voting - I'm not concerned about the user's rep. I'm just trying to sort the posts into order of usefulness. For me, a negative score means the content should be disregarded or taken with copious salt. An "ok" post at -1 is therefore just incorrectly classified, and I want to fix it.

So, my point is that you cannot ascribe "poor user, have some rep" motivation to all these votes.

-
"I will sometimes upvote a mediocre but not harmful/wrong answer if it is at -1, where I wouldn't have upvoted it if it were at zero." ==> if this isn't pity voting, then what is it exactly? IMO, this is almost the definition of pity voting... –  fretje Jun 1 '11 at 14:11
@fretje I'd think pity upvoting is more, like Jeff's italic point, voting up "no matter how terrible or wrong it may be". –  Grace Note Jun 1 '11 at 14:14
@fretje it's not pity upvoting because it's not motivated by a feeling of pity. –  Blorgbeard Jun 1 '11 at 14:17
@Grace: I guess we'll first have to settle on the definition then... IMO it's already pity voting when the person wouldn't do it if the score wasn't negative. –  fretje Jun 1 '11 at 14:17
@fretje My personal thought pattern is, I don't care what motivates the person to vote, as long as their spirit in voting is upvoting what they find useful, and downvoting what they believe is not useful. The nature of how score works makes this happen in both directions - users refusing to upvote highly voted answers because they feel "It's high enough", or users refusing to downvote negative elements (but not upvoting) because "It's low enough". Voting to "balance" works much the same, though I hope that the majority who do it are still believing that the question deserves the upvote. –  Grace Note Jun 1 '11 at 14:21

Here is my non-answer based on the ideas I got from the stats answer.

We could conduct some very basic A/B testing.

Divide all our users into 2 groups:

1. Group A (odd user ids) see all questions they did not vote on which are scored -1 as 0. So they will not see -1 anywhere on the site for a bit instead it will be 0.
2. Group B (even user ids) see the "truth".

Run this for a day, then look.

If we notice the first group is less likely to upvote, we know it may be a problem. This will also gauge the size of it.

Of course it is a very elaborate experiment that would mean we need lots of changes just to perform it, personally I do not think it is worth the effort.

-

I have sympathy voted in the past.

If it is indeed a problem, perhaps single downvote questions should see a red zero, instead of a black one. That should offset the psychological aspect of things.

-
So that would be a negative zero, as opposed to a positive zero. *8') –  Mark Booth Jun 1 '11 at 13:52
@markBooth precisely. –  Moshe Jun 1 '11 at 14:32

I will upvote a marginal answer from someone with a low reputation to encourage them.

I expect good answers from someone with a high reputation.

I don't think I'm the only person that feels this way.

The voting on Meta Stack Overflow is more opinion based than the voting on Stack Overflow and Programmers.

The problem I see more often is good answers in obscure topics with zero or one upvote, while fair answers in popular topics get 10+ upvotes. I don't think you can remove opinion and popularity from voting.

-
The whole essence of voting is opinion and popularity, to be frank - it's what it is meant to represent. A highly voted answer only means something "good" because it represents something that is in popular opinion. We cannot enforce what leads a person to decide whether a post qualifies for an upvote or a downvote - but we still encourage that you focus on the merits of the post, and vote on the content. Whether some fraction of individuals express bias on account of "sympathy", it appears to be fine. Statistics show that "pity voting" is not really a problem. –  Grace Note Jun 1 '11 at 12:32
I agree with your statements, but I don't see what they have to do with the question at hand... –  fretje Jun 1 '11 at 12:33
@fretje: The question in the title is "Is there an actual “pity” or “sympathy” upvote problem?" My answer attempts to illuminate one aspect of sympathy voting, and remind everyone of other vote biases. It's up to you and everyone else to determine if my answer has any merit. –  Gilbert Le Blanc Jun 1 '11 at 12:39
@Grace: I'm not yet convinced that statistics shows there's no problem. See Anthony's comment to Jeff's answer. –  Hendrik Vogt Jun 1 '11 at 12:44
@Hendrik It's 59% of only those with exactly two votes. Which is only 1% of the content. Certainly, a large number, quite large over 50%. Not a significant number in the grand scheme, though. I'm unconvinced that this problem is actually severe and not, in part, only 59% on account of so many people being so terribly afraid of "pity" voting that they just don't bother to downvote. –  Grace Note Jun 1 '11 at 12:49
@Gilbert: I'm not saying it doesn't have any merit. But you're talking about voting based on the reputation of the poster. While this question is about voting based on the presence of a negative score on a post (being pity or sympathy voting). 2 totally different things in my opinion. Although you shouldn't do either (you should always vote based on the content of a post), I don't have problems with the former... the latter on the other hand... –  fretje Jun 1 '11 at 12:52
@Grace: OK, I should have been more explicit. My main point is that Jeff found the 59% "most telling", and Anthony's point was that 59% is a lot more than 50%. And I'm with Anthony here. (You're right, one could say this is marginal since we're talking only about a small portion of all the questions here.) –  Hendrik Vogt Jun 1 '11 at 12:55
@Hendrik What I find most "telling" is that with questions costing 0 to downvote, and answers costing -1 to downvote, that the ratio of potential "pity" voting on answers is less than the inverse of that on questions. If pity voting was a real problem, I'm with Jeff - I'd definitely expect to see a far greater ratio, especially on answers. That'd be where people would be most reluctant to add the second downvote, and where there's so much more content to allow such voting practices. –  Grace Note Jun 1 '11 at 13:00
@Grace: It's really interesting that the numbers for answers are so different. But don't forget that the free downvotes on questions are rather new, so they can't play a big role in those numbers. –  Hendrik Vogt Jun 1 '11 at 13:06

Just found a nice example. After posting a series of duplicate (and barely comprehensible) questions, which got down-votes and close-votes, the user edits his last one to include "guess everyone hates me". Presto - two up-votes within 30 seconds.

And another: some lovely chap posts a question saying "I know my posts suck, but who cares, you shouldn't ban me from asking" and refuses to listen to anything anyone is telling him. Goes to -8 and then 3 upvotes materialize despite his continuing childish behavior. (oh, and this doesn't appear to be his first account that's been question banned either.)

Contestant number three posted an angry, conspiracy-filled rant accusing someone of threatening to delete his account, but wouldn't even disclose which site it occurred on, much less provide any evidence. Naturally he was shocked that we wouldn't all just take his word for it and do something to someone. Then he wasted hours of everyone's time arguing in circles without actually explaining anything. Four people thought this was worthy of an upvote, meaning he gained reputation for his vague yet belligerent meta-help vampire diatribe.

-

Damnit people! That was too many downvotes for this question and not enough upvotes. Can't I just have a score of zero?

-
Harder to equalize now, is it? –  Lorem Ipsum Jun 1 '11 at 21:43
Dunno. I still don't have enough rep to see up vs. down votes. If I change the plea, will it help you think? –  Paul McMillan Jun 1 '11 at 22:23
You're at -3/0 right now. You can use this timeline to get the spread easier even when you're at sub-1k. –  Grace Note Jun 1 '11 at 22:28
+1 because I hadn't come across the timeline feature (mentioned in a comment to this answer) before. Not a pity upvote! –  Andrew Grimm Oct 28 '11 at 7:33