Wikipedia:Wikipedia Signpost/2015-08-26/In focus

• It's an interesting trend. Do you know why the August figures will take so long? All the best: Rich Farmbrough, 02:07, 28 August 2015 (UTC).[reply]
  
  • I'm assuming it will be a month after July came out, though that may have been delayed for extra scrutiny as several people were cautious about the upturn. ϢereSpielChequers 10:10, 28 August 2015 (UTC)[reply]
• This is great! ~Liance^{talk/contribs} 03:27, 28 August 2015 (UTC)[reply]

• There are many ways to criticize this metric, and clearly the 100+ edits a month statistic is an imperfect way to measure the count of "really useful editors", paraphrasing Thomas the Tank Engine. But it is a useful metric, and I see no evidence that the ranks of disruptive or obsessive editors racking up 100 edits or more a month is increasing at the expense of genuine content creators. This data calls the "Wikipedia editor count is plunging!!" meme into question. Cullen³²⁸ Let's discuss it 06:24, 28 August 2015 (UTC)[reply]
  • It hasn't just been a meme. The raw figures had been falling for some time, and while part of the decline was due to the edit filters, and part to increasing speed of vandalism reversion, a lot of people believe that most or all of the decline was "real". Yes there will be a time in the distant future when edit count on individual language projects will be as useful an indicator as tonnage of horse shit removed now is of London traffic. If and when editing of infoboxes moves to Wikidata then inevitably the edit count on individual language wikipedias will dip, just as the change to a hub and spoke system for intrawikis lost us lots of perfectly good bot edits. If we ever reduced the default from four warnings to three before we blocked vandalism only accounts, then we would not only lose more good edits than bad (one rollback and one userpage warning for each lost vandalism) but our cherished and longstanding metric of editors doing 5 or more edits in the month would take a knock. By the mid twentieth century the AIs could have taken over this site and be generating each language version of an article from a common core except only where the local language community had opted for local human control of an article of special cultural significance to their language. I'd like to see us measure something more nuanced such as unique hours in which a volunteer saves an edit. But as long as we treat it as an indicator, and remember that what it indicate does change over time, then I think these statistics are worth collecting and monitoring. ϢereSpielChequers 10:10, 28 August 2015 (UTC)[reply]

A nice piece: the downward trend had been showing signs of reversing for a while.

One thing to bear in mind though is that the number of edits per article has clearly declined. In March 2007 there were 4.8 million edits spread across 1.6 million articles (about 3 edits per article). In March 2015, there were 3.1 million edits spread across 4.8 million articles (about 0.6 edits per article).

So articles as a whole are stabilising. In some cases, this may be because they have matured and become really good (I believe featured articles for example generally see fewer edits) or "good enough", in other cases it may just be a question of fatigue on the part of editors who used to fight over content – then you get articles that look like abandoned battlefields (I can think of a few).

In addition, new articles added see fewer edits than they used to in the past – possibly because they are in niche topics, with shorter content and fewer people interested in them.

What has also sharply declined, of course, is the number of highly active Wikipedians per article. This has potential implications for quality control. For example, hundreds of thousands of articles are on no active editor's watchlist. Andreas JN466 14:32, 28 August 2015 (UTC)[reply]

Hi Andreas, I'd agree there has been a fall in edits per article, though I'd point out that part of that will be the things that contributed to the overall fall in the number of edits. Edit filters, faster vandalism reversion, the revised intrawiki system etc. To some extent improved edit filters and pending changes can also compensate for the decline in watchlisting, my own watchlisting picks up rather less vandalism than it once did. I suspect we are also filling in the gaps with lots of uncontentious articles that rarely change, it would be interesting to do stats on the proportion of contentious articles at any time - my suspicion is that the number of disputed articles grows far more slowly than the number of articles. But I do agree that there is a problem, and my preference is that at some point we introduce flagged revisions as it works on DE wiki. That way we can be sure that every edit by an IP or newbie is looked at by at least one regular. ϢereSpielChequers 16:48, 28 August 2015 (UTC)[reply]

I interpret this low-edits-per-article-overall aspect of the statistics very pessimistically. And indeed, I interpret the recent unmistakeably-upward trend in active editors extremely pessimistically.

Now, for reasons methinks are obvious, I would very much like to be proven wrong, about the pessimistic conclusion being the more plausible one. But, since I see zero mentions of "paid*" nor "disclos*" nor "sock*" in the statistics being offered here, nobody seems to be *trying* to disprove me, and I'm not sure whether that is because my fears are implausible and my anecdotal evidence flawed, or because other people are more optimistic about humanity than myself, and have not considered the alternative explanation for the uptrend. Questions: what percentage of the uptrend is paid editing by disclosed paid editors? What is the estimated ratio of disclosed-paid-editor-count, to undisclosed-paid-editor-count? Have those figures changed since 2013, when the current uptrend kinda-sorta-began? Those are my pessimistic questions. I realize that the wiki-stats available won't have my answers, since to get exacting answers would require analysis of every single edit by unbiased neutral humans. (Catch-22! those are the people we want *making* new edits ... not spending time analyzing *past* edits.)

But I do think that some of the figures we have could be analyzed, to detect agenda-and-promotional-and-battleground-type editing, in a specific fashion. For instance, of the July-uptick, how much or how little of that uptick is related to the 381-strong sockfarm just revealed? We know the sock-list now, so this should be a straightforward SQL query, for somebody with backend-privs. As another idea, more likely to give us rough answers of the kind we're looking for: what if we picked a random sampling of 1000 edits from the June'15 dataset, and analyzed those alone (e.g. by publishing the list of the 1000 randomly-selected edits here on-wiki then crowd-source analyzing them all in divide-and-conquer-fashion). We could categorize them into sourced-improvement, quasi-sourced-quasi-improvement, neutral-grammar-or-similar-tweak, unsourced-addition-factual-looking, unsourced-addition-promotional-looking, quasi-sourced-addition-agenda-pushing, blatantly-vandalism, and maybe a few others?

Then, do the same for 1000 random edits from June'14, from June'13, from June'12, and so on back to 2005 or something, to give us a nice long ten-year-trend. I'd rather see those type of statistics, grouped by type of edit, than the generic anonymized statistics grouped by edit-count-itis-standing. That would give us some insight into whether the uptrend is an increase in good-faith contributors... or instead, to quote a memorable phrase used above out-of-context, a metric related to the tonnage of horseshit that needs to be removed. Gracias, 75.108.94.227 (talk) 16:17, 1 September 2015 (UTC)[reply]

Some of these things are difficult to test, and just because no one is volunteering to test them doesn't mean they agree or disagree with them. I have tested the direct effect of Orangemoody on the >100 a month metric. I've checked a dozen of those accounts at random and only one was even close to 100 edits in any one month, and while many now have more deleted edits than live, the pattern includes gnomish edits as well as article creation, so many of them will have live if reverted edits. So I doubt there will be much direct effect on the >100 edit metric, as unless my sample is ridiculously skewed the great majority of those editors won't have reached the 100 edit total in any one month. Indirectly of course there should be some effect, deleting a bunch of articles created over the last few months will take out all the edits to those articles, and there could be categorisers or other gnomes who drop below 100 live edits in one or more recent months as a result of this. Whether that exceeds normal attrition will be hard to spot, but sometimes AFD deletes contentious articles with a lot of contributors and if that happens it could eclipse the effect of these deletions. I would not be at all surprised if the July uptick disappears when we get the August figures, many articles created in July won't have been deleted until August, and these will be skewed towards one that were categorised, wikified and deletion tagged in July. So when August comes out if July 2015 has dropped below July 2013 but above July 2014 I won't be surprised but I'd suspect it is more likely normal deletions than Orangemoody - not least because any effect of Orangemoody will be spread across the several months that those edits were made. As for the rest of your comments, I don't know how to estimate how much undisclosed paid editing is going on, and I doubt anyone else does either. My suspicion is that spam generally increases in line with perceived value to marketers of an entry in Wikipedia, and that probably lags one or two years behind our total readership, but how do you measure covert activity? ϢereSpielChequers 13:05, 3 September 2015 (UTC)[reply]

Where I hope Orangemoody will have an effect (a negative one in terms of number of edits and number of articles, but a positive effect on the much less easily measurable factor of quality) is inducing us to remove the similarly promotional articles inserted by other promotional editors, and especially the many thousands added in the past when our standards were lower. The only real way to prevent such abuse is by more careful scrutiny, and I hope the effect of this concentrated effort to remove some of this material will be to encourage people to work much harder at detecting and removing others. DGG ( talk ) 23:25, 3 September 2015 (UTC)[reply]

I had really slowed down my content editing up until about 1 month ago, largely because I am working for the Foundation on

Anyways, I still have a bunch of wikiproject pages on my watchlist, but they barely ever come up anymore. WProject comics used to have maybe 10 comments a day. Now it's a few a week. Same with the rest. Things are way less active than they used to be. But, I'm back, doing a little bit here and there, though. Maybe we hit rock bottom and will start moving up. I think wikinews never came back, though, as an example, so who knows. - Peregrine Fisher (talk) 03:07, 3 September 2015 (UTC)[reply]

Welcome back! I find the active pages, editors and even wikiProjects change over time, so looking at a bunch of pages that were active several years ago will likely include more of the ones that have become less active than the newly active ones. My experience of WikiProjects is that many have watchers who return if there is a query and respond if it interests them. So while I agree that overall we probably are less active than at peak, it is easy to get a skewed impression and difficult to really measure how the community has changed. ϢereSpielChequers 12:33, 3 September 2015 (UTC)[reply]

August stats are in and the trend continues. I'm starting to monitor this at User:WereSpielChequers/100+ editors ϢereSpielChequers 17:36, 8 September 2015 (UTC)[reply]

There is an interesting discussion at JimboTalk about picking ten random articles, and then comparing their article-quality today, versus their article-quality one year ago in 2014, versus their article-quality 5 years ago in 2010, versus their article-quality ten years ago in 2015. Although it is somewhat biased by the eye-of-the-beholder phenomenon, it does seem to be a useful metric.

    Is there a way to run a similar test on the usernames with 100+ edits per month? Which is to say, randomly pick ten of those usernames, and then examine their contrib-quality for September 2015, compared to September 2014, September 2010, and September 2005? That might give us a rough idea of what sort of editors, doing what sort of work, are causing the uptrend. 75.108.94.227 (talk) 05:47, 11 September 2015 (UTC)[reply]

I don't know how to get that data myself, but I fear it would have the same drawback as ten random articles in that it wouldn't be statistically valid; Especially if you were looking for sub trends such as FA contributions. Ten editors might give you a rough idea between gnomes and content writers, but in the unlikely event that a single FA writer happened to be among the ten that would skew things hugely. It would also be incredibly skewed if you limited it to ten editors who have been active all the way through as opposed to ten random active editors for each year. Yes there are editors who have been consistently active since 2005 - I've heard that until this year the 2006 cohort was the most productive in every year since 2006. But those editors have themselves changed over time, some may have gone from vandalfighting adolescents to article writing postgrads. Others may have written the articles they were interested in and since hung around maybe doing an hour or two on AWB or Huggle once a fortnight. I've only been here since 2007 so I wouldn't make your sample, but I guess I'm in the group who found some niches, in my case categorisation and typofixing, and much of my editing by volume is still in those niches. But my earliest edits definitely won't have included citations, and while I add little content nowadays, when I do I cite it. So if you were looking at it from that angle I would suggest a much bigger sample size, and decide if you were looking at how longterm editors change over time, or at how the currently very active community changes from year to year. Personally I'm taking a very different approach, looking for anomalies and patterns, hypothesizing various things that could have changed the stats over time and then trying to quantify that effect. For example there was a suggestion that the Orangemoody related deletions would have a measurable effect on this by deleting hundreds of articles created in the first half of this year. User:WereSpielChequers/100+ editors is my evidence that the effect of Orangemoody on the >100 edits stats was very probably trivial, or just conceivably, big but almost exactly balanced out by an otherwise unobserved and unexpected move in the opposite direction. ϢereSpielChequers 11:07, 11 September 2015 (UTC)[reply]

We have years of decline, then one indicator goes contrary. If it is reflective of anything it is the bad actors beginning their steady march. As another editor mentions above WikiProjects have mostly failed. Barnstars and WikiLove are nice and all but they have nothing to do with the integrity of the project. - Shiftchange (talk) 07:21, 5 October 2017 (UTC)[reply]

It isn't just one indicator now, though we held back the publication of the article for months while people double checked the stats on this one. At the other end of the scale look at Wikipedia:Time Between Edits. For the last year we have been pretty consistent at about 60 days per ten million edits, in 2014 we had three consecutive ten millions that took over 70 days. As for Wikilove, thanks are very nice, but they aren't logged as edits and I'm sure that sometimes they replace them. ϢereSpielChequers 23:35, 24 October 2017 (UTC)[reply]