13

So, I'll confess, I'm addicted to SEDE. To date, I've spent most of my time crafting queries to help with Spring Cleaning over on Chem.SE.

However, alongside the site cleanup efforts, it appears that the impending approach of finals, standardized exams, etc. is resulting in an uptick in lower quality, "HOW I DOZ THE PORBLM" sorts of questions on the site. This got me curious as to whether Chem.SE has seen similar inrushes of lower-quality questions around this time every year.

So, naturally, I wrote a query. Disappointingly(?), no such trend is evident.

However, out of curiosity I ran the query on some other sites. The results on SuperUser were striking (click to enlarge):

SU monthly posted questions chart

The yellow trace is the total number of questions posted to the site per month, including questions that have since been deleted. The blue curve includes only questions that remain non-deleted; the red curve is the difference between these two. The green counts questions posted that are currently closed, but not deleted.

What did the Stack Exchange network do in May 2016 that reduced the number of low-quality questions so dramatically? I looked around Meta.SE a bit to see if I could find anything relevant, but without success. This Meta.SE answer talking about the hoops new askers have to go through is from 2014, and so is an unlikely explanation for the change. The second iteration of the SEQP was under heavy discussion in Oct 2016, well after this drop occurred, there doesn't appear to be anything around that time in , and a search of Meta.SE over the first half of 2016 yields nothing particularly illuminating.

FWIW, a similar drop is evident for the other major SO properties (click thumbnails to enlarge):

Stack OverflowSO questions plot

Server FaultSF questions plot

Ask UbuntuAskUbuntu questions plot

However, the drop is absent for most of the other sub-sites I checked—viz., Software Engineering, AskDifferent, Arqade, Physics, Math and Chemistry. Unix & Linux did show a slight drop around that same time, but it looks like more of a correction to an aberrantly high number of bad questions that the site received in April 2016.

So, again: What did Stack Exchange do in May 2016 that has had such success in discouraging poor-quality questions on the major sub-sites?

2
  • The perpetual "recent changes" question meta.stackexchange.com/a/272544/133299 - "2016-05-03: The new-user ask limit is now one question every 40 minutes, network-wide." seems pretty promising?
    – Cascabel
    Commented Apr 6, 2017 at 19:32
  • @Jefromi Well, that didn't take long. Yes, that absolutely makes sense. <favorites 'recent changes' post>
    – hBy2Py
    Commented Apr 6, 2017 at 19:33

1 Answer 1

26

I'm afraid the explanation here is kinda boring.

There's a job that runs weekly which deletes any question more than 365 days old which...

  • has a score of 0 or less, or a score of 1 or less in case of deleted owner
  • has no answers
  • is not locked
  • has view count <= the age of the question in days times 1.5
  • has 1 or 0 comments

On Super User, there are currently 10,484 questions which qualify to be deleted under those criteria except that they're less than 365 days old - so in 30 days, around 900 of them will be deleted.

Same goes for most sites, with obviously a lot more questions qualifying on most of the larger sites. As such, you'll always see a significant drop in deleted questions extending roughly 1 year in the past (with smaller drops for the last 30 days and last 9 days).

4
  • 1
    Wait... if this were the reason, though -- wouldn't there be a corresponding jump in the 'non-deleted questions' curve of equal magnitude at the same time? Also, this drop is positioned less than 365 days ago -- none of the May 2016 questions, and basically none of the Apr 2016 questions (SEDE updated 4/2/17) would be subject to this automated garbage collection.
    – hBy2Py
    Commented Apr 6, 2017 at 20:09
  • 3
    There is a jump in non-deleted questions, @hBy2Py; it doesn't match the drop in deleted, but there's an overall drop in the number of questions created so it wouldn't. You're grouping these questions by month, and assigning the date for each group as the end of the month - so the drop shows up on April 30th, but that group includes all of April. The relevant scheduled task last ran on April 1st, so only a few hours worth of eligible posts from April of 2016 would've been included in the last SEDE update - thus the lack of deleted questions in the April group in your query.
    – Shog9
    Commented Apr 6, 2017 at 21:04
  • 1
    Ahhhh, of course, MAX(pwd.CreationDate) lumps everything in arrears. I'll change the query to MIN(pwd.CreationDate). Thanks, and sorry for the error.
    – hBy2Py
    Commented Apr 6, 2017 at 21:09
  • 5
    Here's a simpler (and slightly more accurate) query that'll group by the first of the month and exclude the last, partial month: data.stackexchange.com/superuser/query/654621#graph
    – Shog9
    Commented Apr 6, 2017 at 21:11

You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .