# Observation on user voting statistics

The question of who downvotes / upvotes seems to come up a fair bit in meta discussions. I recently noticed mention of a script that Anindo Ghosh wrote that identifies a list of users with a high downvote percentage and of course when you view a user's profile the same statistics are available as raw numbers:

Extreme Downvoters

My rating of 33% downvotes seemed about right taking into account all votes, but only if it included permanently deleted posts. Over a few days I took a careful note of what I'd downvoted and one morning found out of 12 downvotes the night before:

• 10 had been deleted for 'reasons of moderation' and accounts deleted (troll posts)

• 1 had been closed as not a real question but remained on site

• 1 remained open

On a second day I got similar results, while when you downvote answers that are later deleted you get back the -1 rep but it doesn't change the voting statistics on your account. Not that I think it's any sort of problem that should be addressed I'm pointing out that using these statistics to judge user behaviour is largely useless.

Comments are welcome I'm just pointing out what I do personally (and presumably others) can skew results a lot on these overall numbers. Maybe I'm wrong but I think heavily downvoting trolls quickly reduces their chance to further spam the site because they'll get a ban on posting faster?

• @W5VO, lol, actually I thought when I read community 'owns' downvotes on bad posts it might get removed from the accounts that downvoted but I'm guessing it only owns the downvotes on the original post? – PeterJ Jun 10 '13 at 14:05
• Those statistics don't show the percentage of up/down votes broken up between questions and answers. The aggregate is therefore of little meaning and it is hard to understand site dynamics because too much individual information is lumped into these gross totals. Most of my downvotes are for crappy questions, since we get a lot of them here, and it often doesn't occur to me to upvote questions when they are good. On the other hand, I downvote answers only when they are outright wrong, misleading, gibberish, or spam. The aggregates don't tell you these things. – Olin Lathrop Jun 10 '13 at 14:40
• If we would have access to every up and downvote (linked to both a user and a post), we could strip out the deleted questions. Too bad that isn't possible, but that's also a privacy issue of course. Nice work, btw! :) – Keelan Jun 10 '13 at 17:21
• Another problem with that script is that it shows users with low total votes. Until you have cast at least 100 votes or so, I think it is difficult to draw any conclusions about your voting patterns. Users with less than 100 total votes should be eliminated from the list. – Olin Lathrop Jun 12 '13 at 13:05
• @OlinLathrop I played with the data explorer a bit; it seems impossible to separate question votes from answer votes while maintaining user identity. The provide up/down vote totals for a user but don't give the post, and up/down votes per post but don't give the user (and by the post, we can know if it's a question or an answer). I suspect this is by design, to keep up/down votes anonymous. – Phil Frost Jun 22 '13 at 11:50

Regardless of what statistical measures you may apply, and what content gets lumped into the aggregate scores, you can still very easily apply a little logic.

Community is the Mods proxy "bad guy" account. I think it's safe to assume that that account could be held as a low water mark. So if you're "out-performing" community in total DV's and almost never up vote, I would say that you're just a troll who spam votes.

Even the semi-hallucinatory variably lucid resident troll (I assume the stats come from before the account lock out) has a better record than a number of people.

• Community is not a bad guy proxy account, do not use that as any sort of grade of what others should be done, it absorbs votes in many many cases. – Kortuk Jun 10 '13 at 19:43
• @Kortuk All spam and offensive flags result in downvotes attributed to community. Its safe to assume that the percentage could be used as a measure. – asheeshr Jun 11 '13 at 3:17
• @AsheeshR I thought it pulled some from account deletion and such also, I could be wrong. – Kortuk Jun 11 '13 at 13:35
• @Kortuk Yes. That is included too, but the downvotes are >= spam posts + offensive posts + dvs by deleted users whereas upvotes = uvs by deleted users. Hence, using it can be used as a minimum reference percentage. – asheeshr Jun 11 '13 at 13:40
• If I've read what AsheeshR is saying correctly, if the totals did include those other elements the bar would be lowered even further, i.e. it would take more of an effort to scrape underneath. If anything that strengthens the argument. – placeholder Jun 11 '13 at 13:47

Olin points out that there are some users that haven't voted much, so we don't have an accurate picture of their habits. We can correct for this by creating a confidence interval for the downvote rate, that is, instead of giving the rate, we give a range of values within which the actual value is likely to fall, based on the votes they have made so far. If they have made a lot of votes, the range is small. Few votes, the range is large.

This is called a binomial proportion confidence interval.

I made this query:

Extreme downvoters with a 95% confidence interval

This is sorted by the minimum of the interval, so if being at the top of the list is "bad", then we are assuming good faith. Of course, it only is fair to do the same for upvotes:

Extreme upvoters with a 95% confidence interval

Again, sorted by the minimum.

Unfortunately, it doesn't seem possible to break this down by question and answer votes. The schema will describe what votes were done on what posts, and we can know if a post was a question or an answer. But, up/down votes are not associated with a user (close, accepted, etc are); we can only get a user's upvotes and downvotes as an aggregate count. I bet this is by design, to keep votes anonymous.

My interesting observation: the confidence intervals are tighter for extreme upvoters than downvoters.

• These queries are clearly more relevant, statistically balanced, and hence valid than my query, that started off this whole topic... What they do underline, though, is that clearly there are some people overflowing with negativity, pretty much the same "extreme" people as in the original list. Yes, there apparently also are some people who are all appreciation and encouragement. That being said, I think it is clear which group one would appreciate more as decent fellow human beings. – Anindo Ghosh Jun 23 '13 at 13:37