Why do well received questions often score early down votes?

Question

I've noticed something interesting in voting trends. Quite often the more interesting questions receive early down-votes (fairly quickly) before subsequently receiving 15+ up votes. This is in contrast to some truly poor questions which are ignored and sit on zero (0).

I find this behaviour intriguing. It seems that the more precise and narrow people are with their question, the more chance there is of early down votes which is almost the exact opposite of the intended purpose of voting on questions. I am, personally, very cautious with a down-vote on a question (preferring to comment instead) and very liberal with an up vote. So it seems odd to me that people down-vote well bounded questions which encapsulate a single point the OP doesn't understand.

The only explanations I can come up with are that either:

• Some people are a bit arrogant and down-vote "you idiot, why don't you know this?"
• Some people freak out when they read a question they feel in unanswerable; not just they don't know the answer but feel that an answer cannot be found.

I recognise everyone has their reasons for voting, everyone is entitled to their opinion, and we cannot and should not "police" the voting system. But I do wonder what we can do as a community to be more graceful particularly towards those asking [ultimately well received] questions.

Answer 1

I'm just speculating, since of course an actual answer could only come from... a committee of downvoters. And I may be thinking about the wrong examples, since you didn't provide any in your question.

In my experience, the kind of questions you are referring to receives at most one or two downvotes. As you note, even if they appear to show some pattern, a few downvotes are probably not significant of a collective attitude or culture, likely being the legitimately free choice of a handful of individuals.

That said, I see plausible reasons for downvoting narrow, well scoped questions: when we feel they don't show research effort, we can be reasonably sure it is actually the effort that is lacking. (Of course I'm referring to the often quoted downvote arrow's tooltip: "This question does not show any research effort; it is unclear or not useful").

I am no big downvoter at all and I agree that commenting should be the preferred way to engage with a question that needs to be improved. But it seems to me that narrow and clear questions are actually a case where downvoting can be useful, because it has some chance to carry a clear message: "You don't seem inexperienced, I feel it would be fair to ask you a bit more of effort when you post a question".
This kind of downvote may be felt by some users as useful to the site: they try to discourage scarcely researched questions in contexts in which it seems to be room for discouraging in a (hopefully) constructive way — that is usually not the case when questions are unclear and seemingly come from inexperienced users.

---

Beside speculating, we can also take a look at some data.

The following table comes from this SEDE query, which selects the number of questions that had a net score of at least 15 on their first day, grouped by year and by the number of downvotes they received on their first day.

Columns:

• tot(#q): total number of questions in the year (15+ score on first day);
• day1 DV:: downvotes on first day;
  • #q: number of questions;
  • %q: percentage of questions over the yearly total;
  • avg: average score on first day.

_{Data retrieved on 2019-03-13.}

                |day1 DV: 0         |day1 DV: 1         |day1 DV: 2         |day1 DV: 3         |day1 DV: 4         |day1 DV: 5+
year    tot(#q) |#q %q      avg     |#q %q      avg     |#q %q      avg     |#q %q      avg     |#q %q      avg     |#q %q      avg
-------------------------------------------------------------------------------------------------------------------------------------
2010    1       |1  100.00  31.00   |   0.00            |   0.00            |   0.00            |   0.00            |   0.00    
2011    7       |6  85.71   20.00   |1  14.29   16.00   |   0.00            |   0.00            |   0.00            |   0.00    
2012    11      |10 90.91   19.60   |1  9.09    41.00   |   0.00            |   0.00            |   0.00            |   0.00    
2013    23      |17 73.91   23.35   |5  21.74   25.40   |1  4.35    31.00   |   0.00            |   0.00            |   0.00    
2014    35      |31 88.57   19.45   |2  5.71    18.50   |   0.00            |1  2.86    26.00   |   0.00            |1  2.86    17.00
2015    23      |20 86.96   19.30   |2  8.70    17.00   |   0.00            |1  4.35    15.00   |   0.00            |   0.00    
2016    39      |26 66.67   21.23   |6  15.38   20.33   |6  15.38   19.50   |1  2.56    22.00   |   0.00            |   0.00    
2017    29      |17 58.62   21.06   |8  27.59   20.38   |1  3.45    25.00   |1  3.45    18.00   |2  6.90    16.50   |   0.00    
2018    35      |25 71.43   20.00   |8  22.86   22.63   |1  2.86    23.00   |1  2.86    18.00   |   0.00            |   0.00    
2019    11      |7  63.64   19.00   |4  36.36   18.50   |   0.00            |   0.00            |   0.00            |   0.00    

A sample of some tens of questions is probably too small to be significant (or, more precisely, a sample of 1 to 13 early downvoted questions is probably too small).
Hence, here is the same table for questions having a net score of 5 or more on their first day:

                |day1 DV: 0             |day1 DV: 1         |day1 DV: 2         |day1 DV: 3         |day1 DV: 4         |day1 DV: 5+
year    tot(#q) |#q     %q      avg     |#q %q      avg     |#q %q      avg     |#q %q      avg     |#q %q      avg     |#q %q      avg
-----------------------------------------------------------------------------------------------------------------------------------------
2009    1       |1      100.00  7.00    |   0.00            |   0.00            |   0.00            |   0.00            |   0.00    
2010    73      |72     98.63   6.60    |   0.00            |1  1.37    5.00    |   0.00            |   0.00            |   0.00    
2011    353     |346    98.02   6.52    |7  1.98    7.71    |   0.00            |   0.00            |   0.00            |   0.00    
2012    407     |398    97.79   6.63    |9  2.21    10.78   |   0.00            |   0.00            |   0.00            |   0.00    
2013    510     |482    94.51   7.05    |24 4.71    10.50   |4  0.78    13.25   |   0.00            |   0.00            |   0.00    
2014    783     |733    93.61   7.08    |45 5.75    7.36    |3  0.38    7.33    |1  0.13    26.00   |   0.00            |1  0.13    17.00
2015    525     |482    91.81   7.31    |34 6.48    8.00    |7  1.33    7.43    |1  0.19    15.00   |   0.00            |1  0.19    7.00
2016    523     |432    82.60   7.82    |63 12.05   8.49    |24 4.59    10.63   |3  0.57    11.67   |1  0.19    10.00   |   0.00    
2017    546     |454    83.15   7.44    |69 12.64   8.81    |17 3.11    8.71    |2  0.37    13.50   |3  0.55    14.33   |1  0.18    13.00
2018    505     |392    77.62   7.80    |93 18.42   8.49    |12 2.38    8.42    |7  1.39    9.29    |1  0.20    7.00    |   0.00    
2019    101     |75     74.26   8.04    |23 22.77   9.30    |3  2.97    9.33    |   0.00            |   0.00            |   0.00    

I'm not sure of what we can infer, though, apart from noting a rising trend in positively received but early downvoted questions. This is probably something we should address, but maybe not so relevant to our current issue.

Finally, It may be interesting to compare these statistics to those of the whole question base on U&L. In the following table all questions are considered, independently of their score on their first day:

                |day1 DV: 0             |day1 DV: 1             |day1 DV: 2             |day1 DV: 3             |day1 DV: 4         |day1 DV: 5+
year    tot(#q) |#q     %q      avg     |#q     %q      avg     |#q     %q      avg     |#q     %q      avg     |#q %q      avg     |#q %q      avg
-----------------------------------------------------------------------------------------------------------------------------------------------------
2008    5       |5      100.00  1.20    |       0.00            |       0.00            |       0.00            |   0.00            |   0.00    
2009    26      |26     100.00  0.81    |       0.00            |       0.00            |       0.00            |   0.00            |   0.00    
2010    1,521   |1,495  98.29   1.60    |22     1.45    0.41    |4      0.26    0.25    |       0.00            |   0.00            |   0.00    
2011    7,111   |6,928  97.43   1.67    |155    2.18    0.70    |16     0.23    -0.69   |8      0.11    -2.38   |4  0.06    -3.25   |   0.00    
2012    9,682   |9,422  97.31   1.59    |200    2.07    0.98    |31     0.32    -0.45   |19     0.20    -2.00   |8  0.08    -3.00   |2  0.02    -4.00
2013    14,259  |13,612 95.46   1.38    |489    3.43    0.80    |93     0.65    -0.26   |43     0.30    -1.81   |14 0.10    -3.43   |8  0.06    -4.75
2014    20,204  |18,956 93.82   1.30    |937    4.64    0.73    |198    0.98    -0.71   |61     0.30    -1.41   |30 0.15    -3.20   |22 0.11    -4.50
2015    24,107  |22,535 93.48   0.95    |1,184  4.91    0.28    |248    1.03    -0.93   |87     0.36    -2.02   |27 0.11    -3.44   |26 0.11    -4.62
2016    25,396  |23,071 90.85   0.89    |1,644  6.47    0.28    |398    1.57    -0.50   |152    0.60    -1.97   |78 0.31    -3.12   |53 0.21    -5.28
2017    25,130  |22,732 90.46   0.82    |1,636  6.51    0.34    |451    1.79    -0.95   |169    0.67    -2.25   |86 0.34    -2.93   |56 0.22    -4.55
2018    24,613  |21,132 85.86   0.63    |2,433  9.89    0.24    |624    2.54    -0.92   |253    1.03    -1.94   |99 0.40    -3.20   |72 0.29    -5.10
2019    5,221   |4,411  84.49   0.59    |600    11.49   0.13    |139    2.66    -1.06   |46     0.88    -2.11   |15 0.29    -3.67   |10 0.19    -4.50

Actually, a well received question seems to have roughly twice the chances of being downvoted early than a randomly picked one. But this may be just the predictable effect of being interesting — attract both upvotes and downvotes.

---

Now, we cannot ignore HNQs.

As Jeff Schaller noted, it is uncommon for questions on U&L to quickly receive a high number of votes (where "quickly" refers to the first day and "high number" may be 5, just to arbitrarily set a bar).
Unfortunately, knowing if a question has ever been on the "Hot Network Questions" list has become possible only recently, as announced on Meta Stack Exchange on March 11. An event is now added to a question's timeline and edit history when it appears on the HNQ list for the first time.

However, some voluntary projects allow us to know what has been HNQ on some specific time intervals. This answer to How to determine if a question has ever been a “hot network question”? on Meta Stack Exchange mentions a chat room that lists all the questions that have been picked as HNQ since 2019-02-28.
Considering the net score of questions on their first day (list retrieved with this SEDE query on 2019-03-15) and searching for them with that chat room's built-in search tool, it turns out that to date 71% (32 out of 45) of those that scored at least 5 have been HNQ (and 75% (15 out of 20) of those that scored at least 10, 77% (10 out of 13) of those that scored at least 15)¹.

¹_{Note that, as stated in the linked post, the question list is probably incomplete and thus we may have some false negatives.}

---

Putting it all together, the most likely answer I can come up with is: we are seeing a relatively small number of likely ordinary questions which get exposed in an uncommon way through the HNQ list and, therefore, have odds of receiving at least one downvote that are higher than, but comparable to, the odds of being downvoted of the average question.
I know this is not much exciting, but...

Answer 2

In my opinion, I think there's a semi-active user that is exercising their right to vote and is doing so on some criteria that obviously doesn't align with 16+ other voters. It's the disparity of that -1 (or more) followed by the +15 that sticks out, but it's likely that other posts are also being voted on but don't make it to +15.

I'm actually not sure anything needs to change; as you pointed out, those questions ended up being "well-received" (upvoted).

For another data point, the top nine questions (by overall score) here also have at least one downvote on them!

Answer 3

You didn't cite any examples, but I suppose many of those are basic questions that receive a few "you idiot, why don't you know this?" downvotes in the beginning, and then hit the Hot Network Questions list where most viewers can only upvote, so they do.

I believe that basic questions should be closed as duplicates rather than silently downvoted, but unfortunately this is rarely happening. I feel that has to do with the attitude of users who threat close votes as a personal insults, and the fact that dupe search is a thankless job of promoting someone else's answers with no rep gain. So instead of spending time finding duplicates, people prefer either downvote (hurting the user's morale), or repeat the same answers over and over again (hurting the overall quality of the site).

Answer 4

Here's another hypothesis: Questions that get lots of upvotes have necessarily gotten lots of views. It might happen due to the topic, or due to mysterious algorithms, who knows. But in any crowd there will be a few outliers who will casually downvote for no strong reason, especially when there is no clear trend yet. Once there are 10-15 upvotes, it takes more conviction to vote agaist the trend, so the casual downvotes die down. By comparison, questions with few views are also less likely to accumulate downvotes (unless they deserve them). So you're less likely to see a spate of casual ones.

If the topic is also somewhat controversial, in the sense that people have strong reactions to it for whatever reason, so much the better-- more people are prone to an impulsive downvote.