22

I've recently come back and posted a few answers, and was surprised at the lack of interaction. In fact, this site seems quite dead. Thus I played with the Stackexchange Data Explorer a bit, it's a neat toy. This begs a question. What happened around mid-October 2018?

enter image description here

X-Axis is question creation date. Y-Axis is average number of views for questions created at this date. You can display the query here. What is this huge drop in views after October? Same query on The Workplace SE data does not exhibit this trend. Feel free to use the "switch site" feature to try it on your favorite exchanges.

I wondered if this was due to a bug/feature in the pageview counting system, so I probed using another metric, user creation. The curve shows the average number of users being created per day in orange, and the last access of users in blue. This one is from Electrical Engineering.

enter image description here

EE gets a steady stream of new users (50-100 per day) and its last access dates are distributed in the usual hockey stick fashion. If you don't know how to read this, point at "Nov 2017" and you see a value of about 50, this means 50 users visited the site on this particular day and never came back since. The rising part of the curve at the right means many users have their last access date in the last month or so, these users are visiting the site often.

Now for IPS:

enter image description here

This is completely abnormal. A huge number of users saw something on the site around mid to end october 2018, and they never came back, so their last access date stayed the same until now. Or were they group-banned or deactivated? I plotted user creation on the same graph to be certain that this was not due to a sudden influx of new users. On the contrary, users creation dropped massively.

What did they see that made them not want to come back?

This isn't what you want to see:

enter image description here

Again, October 2018 seems to suffer from a curse. There was a time questions would get 7-8 answers on average! And now...

enter image description here

The obvious reason why answers are not posted is that people are no longer posting them. This is the number of unique users posting answers. You want that to be high, not dropping like a rock.

enter image description here

Again, something happened around September or October 2018.

Let's have a look at the whales, ie the users who make the site work. I've cut up the timeline in 28 days slots (4 weeks avoid weekend-dependent data bias) and plotted users who posted 5 or 10 answers in every 28 day slot.

enter image description here

Around September 2017, 20 distinct users posted more than 10 answers each in a 28 day period. The last two blue points on the right are just astralbee all alone. All the other heavy posters ran away. In fact the "most answers in a 28 day period" trophy went to astralbee 6 times and apaul 3 times.

enter image description here

This is a number of answers vs time plot (not reputation). I really gotta give credit to the guy in the upper right corner. Does apaul even sleep? If you're reading that, I must say I respect your determination... or obstination XD

Feel free to use the "switch site" feature to try the queries on your favorite exchanges and compare. Some other exchanges are slowly dying, like EE. Others are going well.

THE QUESTION AT LAST

What happened around mid-October?

(if the data is accurate, of course).

7
  • 7
    I believe that you might find this link insightful
    – Rainbacon
    Jan 3, 2019 at 20:42
  • One thing that I think is really interesting is how smooth the 14 day average got in the first graph after we left HNQ. What does it look like if scaled for only November-December?
    – sphennings
    Jan 3, 2019 at 20:53
  • 2
    @sphennings smooth enough to remove the averaging data.stackexchange.com/interpersonal/query/958378#graph
    – user2135
    Jan 3, 2019 at 21:02
  • 1
    Those are very interesting graphs. Would it be possible to have a link to the query for your last graph? (you posted for all the others but not this one)
    – Ael
    Jan 3, 2019 at 21:05
  • @Noon ah sorry, the last one was generated by a python script, I just made queries to download the users and questions in csv form.
    – user2135
    Jan 3, 2019 at 21:09
  • Alright, thank you anyway, you did an impressive work with those queries
    – Ael
    Jan 3, 2019 at 21:10
  • Possible duplicate of What caused this site to be excluded from Hot Network Questions?
    – gparyani
    Jan 15, 2019 at 22:10

1 Answer 1

19

What happened in October was the exclusion of this site from hot network questions. This was also discussed on meta.stackexchange. Since then there has been some discussion about whether IPS even wants to be back on HNQ, with the sentiment leaning towards "Let's stay off for now.".

A lot of our traffic was driven by HNQ. No longer having questions featured lead to fewer people being attracted to this site. In short people are more likely to follow an attention getting posts and nothing attracts attention like a controversial post.

As you can see from the top graph of your question there used to be a lot of variability in the 14 day rolling average in number of views per question before October. I'd hazard a guess that you can find a contentious question associated with every spike in traffic.

Before our removal there were already discussions about how to reduce the negative effects of HNQ on this site. All of ideas that were considered were about mitigation rather than preventing the problem in the first place.

5
  • Yeah. IPS had stopped growing before the tritter drama occured, though.
    – user2135
    Jan 3, 2019 at 21:48
  • 5
    @peufeu "What happened before October?" is a very different question than "What happened in October?". The former is much harder to quantify given the smaller effect and larger time scale.
    – sphennings
    Jan 3, 2019 at 21:55
  • 2
    No, it's quite simple. Do you remember when I said IPS was doomed and got banned for it? That was a long time ago, and all the information required to make that diagnosiswas already there, plain for all to see. Probably around the peak on this curve from the original question. It's been going down since. The hot question thingy is only the last nail in the coffin, but there were a lot more nails before that. It's quite sad IMO.
    – user2135
    Jan 4, 2019 at 0:42
  • 1
    Well, we've substantially reduced the number of off-topic questions and answers. Cutting traffic by 80% is such an elegant solution, they should roll that out to the other exchanges. Maybe we can try to get de-indexed by google, as well? Jan 5, 2019 at 0:22
  • @kbelder no need, just require an increasing rep threshold to post, the older the question is. Log10(int($DAYS_OLD-2)) x 1000 should git r dun. By the time Google indexes it, drive-by posters can't. Jan 15, 2019 at 0:21

You must log in to answer this question.