User:WereSpielChequers/Going off the boil?

To create a crowdsourced encyclopaedia we enlisted a crowd, and by working together that crowd became a community. The Encyclopaedia grew and became widely used, and as the number of readers rose so did the number of editors. But recent years have seen a disconnect between our ever growing readership and a stable or slightly declining community. This is a summary of the various theories as to why the Wikipedia community appeared to go off the boil some time around 2007. Collaboration is welcome, especially if anyone can link any of these theories to when they were first proposed, and indeed who proposed them.

Most commentators seem to agree that the community changed circa 2007. Editing levels peaked in April 2007, article creation peaked, editor retention started to fall, requests for adminship's best year was 2007, before it fell off a cliff in early 2008 (though its best quarter was late 2005). Readership continues to increase, arguably so does quality. Raw editor numbers waned from 2007 to 2014 (though they rallied in 2015, and the whole 2015-2020 era has on at least one measure had a larger community than in late 2014) and the community is becoming more closed and harder for new editors to become "core" members. The proportion of our readers who edit is steadily falling. The proportion of editors who stay for their 100th or 1000th edit is falling. There are various theories as to why this is happening, some dispute as to whether this means we are in decline, and a couple of things that would reinvigorate the community anyway.

This is a cost of quality

Wikipedia gets better every year: readers are less likely to encounter vandalism or typos, and more of our information is well cited. Articles that were featured in our early days are now likely to have been defeatured even if they have substantially improved since they were promoted. The downsides of this are that some of our existing editors haven't made the transition to the new standards, and some of our entry paths for new editors have been all but lost.

From citation needed to revert unsourced

More and more Wikipedia content is verified with inline citations to reliable sources, and people who don't source their edits are being driven off by the increasing trend away from tagging unsourced edits with [citation needed] to simply reverting them. To some the loss of editors who don't cite their work is an acceptable price for the improvement in quality, others worry that this deters editors from less academic backgrounds. Some formerly trusted editors had the AutoPatroller user right revoked.

Weaknesses of this theory include lack of definitive research to show that people who cite reliable sources are less likely to leave than other editors, or indeed that the percentage of Wikipedia's content that is actually well referenced has been rising. Nor do we know if the proportion of our unreferenced old information that is actually untrue is greater than recent well cited info. The strengths of this theory are that it chimes well with the experience of some of our stalwarts, and that if it is true we needn't worry about the gentle decline in the size of the community. It may even be a temporary decline with a short term decline in the number of editors who don't reference their edits masking an underlying steady increase in the number of editors who do supply sources.

We fix typos too quickly

As quality rises, it becomes less likely that the average reader will spot a typo; we have plenty of typos on new articles on obscure subjects – but those articles have few readers. AutoWikiBrowser has semi-automated typo fixing, and typos are getting rarer. So intermittent editors who just fix what they find when they read are editing less, and fewer readers will become Wikipedians by fixing a typo and making their crucial first edit.

We fix vandalism too quickly

The downside of edit filters and vandalfighting bots is that we no longer recruit many new vandalfighters, and those who fight vandalism by patrolling their watchlist are likely to have less to do. Many editors who started in our early days did so when they saw some vandalism and found that they could easily fix it. As the speed and efficiency of our filters, tools and bots has increased so the number of readers who see vandalism and remove it has fallen. Rapid removal of vandalism is of course a good thing, and the loss of this entry level for new editors is an acceptable price to pay for a less vandalised Wikipedia. Much as with the lack of typos, this explanation sees us currently checked as one part of the community gently fades away, but with an underlying growth rate from the other sources of new editors.

Mitigating the cost of quality

With fewer typos and less visible vandalism we need new entry level activities such as image adding and bot assisted typo hunting. The challenge is how to connect enthusiastic readers who are willing to give us some time, with things that they can constructively do in a series of steps that start as simply as it used to be to fix a typo or remove vandalism.

More contentiously we need to decide whether unsourced new information is still acceptable. Currently our editing screens welcome edits even if unsourced, but many editors revert unsourced edits on sight. This is a recipe for newbie biting. Either we need to guide some of our patrollers and admins into tagging uncontentious but unsourced new edits with [citation needed], or we need to change policies and editing screens to make it clear that unsourced information is unwelcome. The current situation where our unwritten rules are stricter than our written ones is untenable.

All the good stuff has already been done

On this theory all the really interesting articles have been started and there is nothing comparable to work on. Of course this varies by subject area: sports fans will always have new stars to write about as will astronomers, and there are likely to be volcanic eruptions, earthquakes, politicians and newly discovered species. But new chemical elements are a rarity, the solar system has actually lost a planet and we are unlikely to see any more battleship launches. Of course there are whole areas of human knowledge where Wikipedia is barely started, but for people whose interests are already covered it is less easy to join in. A good way to test this would be to look at editing trends by topic; if the 2016 US elections get comparable activity to 2012 and 2008, but computing and the second world war are declining faster than the overall trend then we could be relaxed about the easy parts of some topics being complete.

External things

The web is a fast-changing environment and perhaps we've reacted better to things that might affect our readership figures than our editorship figures.

Competition

Nowadays there are lots of crowd-sourcing opportunities; people can categorise paintings or write recipes and how to guides. With lots more sites that aren't competing with us to create an encyclopaedia, but are competing for the same sort of volunteers we should expect to see some drift in the proportion of our readers who become editors.

Web monitoring

The first decade of Wikipedia saw the growth use of web monitoring software by employers as they sought to reduce non-work activities taking place during working hours. It would be interesting to analyse how much editing patterns have changed by day of the week and by hour of the day. But on this theory a much higher proportion of the pre-2007 editing was done by people at work than in more recent years. Measuring this is of course complicated by the 24 hour clock; it would be much easier to test in single-time-zone Wikipedias such as German, Italian, Korean and Japanese, although in some of these countries web monitoring by employers is seen as more of a civil liberty issue than in the biggest country for edits to English Wikipedia. One positive thing about this theory is that it accounts for a drop in editing without the need for the community to be overly concerned. However much "editing at work" declines, "editing in your own time" is unaffected, and under this theory the community is far from being in a "death spiral". It may even benefit from 2020 onwards by the move to telecommuting.

Mobile surfing

The big contrast is between our rapidly growing readership (up until Jan 2013) and our broadly stable editorship. The Devs have done a great job at making Wikipedia readable on mobile phones, and if we include mirrors our readership may still be growing faster than the Internet. But we haven't yet cracked editing on a mobile, hence editing levels are not increasing in line with readership, and we are beginning to see the greying of the pedia as we become more dependent on the generation who are less mobile phone oriented.

Tablet editing is much more practical than smartphone editing, even though tablets are a less efficient editing device than PCs, so as tablets gain market share from laptops and PCs we can expect to see a drop in editing. When estimating the effect of the rise in mobile, it is important that we differentiate between smartphones and tablets.

According to the Times of India, in 2013 editorship fell 15% and non-mobile page views 20%, whilst Mobile views rose 50%. The perceived drop in pageviews during 2013 may merely reflect increased use of mirrors such as Google knowledge. This would imply that the shift to mobile and mirrors may on its own more than account for the perceived editor decline.

The combination of improvements to Mobile Wikipedia, improvements to smartphones as editing devices, falling mobile bandwidth costs, greater consumer experience of using smartphones as computers, and the process of diffusion of newish tec to the global population, will likely lead to increased mobile editing in future years. Editing Wikipedia is not an entry level task for a new Smartphone user any more than it is for a new PC user, so as with PCs we can expect a time lag between the technology becoming ubiquitous and peak mobile editing.

Internal things – IT

Edits are only part of the picture

Editing on the English language Wikipedia fell significantly between the 2007 peak and the 2014 nadir (37 days for ten million edits in April 07, 58 days in Jan/Feb 2013, 67 days in Jan/Mar 2014 (followed of course by a brief spike caused by the change to the way interwiki links were done and a large number of edits moving them to Wikidata), and the 2015-20 rally and plateau. But that's only part of the picture. Since 2009 our edit filters have prevented more and more vandalism from going live as edits, so the recent raw edit figures omit much of the vandalism that would otherwise now occur, and we also "lose" the associated vandalism reversions, warnings and block messages. It isn't possible to work out how many edits are prevented by the edit filters as vandals who encounter them behave differently. Also as the edit filters only started in 2009, there does seem to have been a "real" drop in editing between 2007 and 2009, and probably again post 2012. Though of course there will also be some effect from the WikiLove feature, as some of the thanks that are delivered through that process would previously have been delivered as edits. Without the filters the post 2009 - 2012 decline would be shallower or even an increase. But the ratio between readers and edits would still be changing as readership rapidly increases.

Our dated software

In the age of Facebook our editing interface sucks. It is still attractive to those nostalgic for the experience of 1980s programming, or who are sufficiently tech savvy to resolve an edit conflict. But we probably need a WYSIWYG editor and we certainly need the software to resolve more edit conflicts without losing edits. There are various ways in which edit conflicts could be reduced, some as simple as treating the addition of a new thread, template or category as not conflicting with the addition of text. The WMF has a major investment going on to deliver WYSIWYG, but they could also easily tweak the code to prevent most edit conflicts (end August 2013 note. We still need a WYSIWYG editor, and perhaps the current development can eventually be salvaged. But with at least one of the serious bugs that was found in the Beta testing not yet fixed, there has to be a concern as to whether the WYSIWYG editor will ever be suitable for general use, and no progress on the more important issue of reducing edit conflicts - update in 2020 the visual editor has now been out for several years, and while it still has problems, it has for some time been claimed to have a better onboarding rate than the classic editor).

User skins

In 2010 the English Wikipedia's default appearance changed from one optimised for editors to one optimised for readers. This has probably contributed to the combination of rapidly growing readership and reduced recruitment of new editors.

Wikipedia offers logged in editors a choice of "skins", ways to see the encyclopaedia, but most editors including all IP editors use the default. In 2010 the default skin was switched from one "optimized for editor usage" to Vector. Vector is a "clean" uncluttered user interface with many editing options hidden in dropdown menus. This is better for readers, but options in dropdown menus are an extra click away from the user, and so the shift in 2010 to Vector as the default skin could partially explain why the last few years have seen a stable or falling editorship despite a rapidly rising readership.

We know that new editors in the Vector era have been less likely to become active editors than new editors did in the Monobook era. But many other things have changed, and we have no true control group, though it is a racing certainty that our core of active longterm editors overlaps strongly with our remaining Monobook users. Research here could be useful, or we could simply default new registered accounts to Monobook and IPs to Vector.

Loss of session data

One of the changes that seemed to come in during 2013 was the increase in loss of session data incidents. As the proportion of edits lost to the inexplicable Sorry! We could not process your edit due to a loss of session data. Please try again. If it still does not work, try logging out and logging back in. rises so there is a treble whammy of lost edits. Directly there are edits lost because of whatever causes these lost edits. Indirectly edits are lost because of the greater time required for a given number of successful edits - some editors may compensate for this by spending more time editing, but those whose time is limited will just commit fewer edits. Other edits will be lost because it just degrades the user experience and thereby makes other hobbies relatively more attractive.

Internal things – The community

In 2007 the longest serving editors had been here six years; by 2020 we have some veterans who've been here for nearly two decades, and most of our admins have been around for well over a decade. That time difference changes the structure of the community, giving time for cliques to form and perhaps for arrogance to creep in. Conscious and unconscious changes made by the community include:

Template bombing

2007 was a key year in the transition of Wikipedia from the wp:SoFixIt culture that was a core ingredient of our success to the template bombing culture of WP:SoTemplateItForSomeHypotheticalOtherToFix that has soured the community. On this theory collaboration helped build the community. Pre 2007 new editors contributed work and then saw their work improved with wikilinks, headings and categories. Today's newbies are more likely to see their work disfigured with templates calling for someone to add wikilinks, headings and categories, that's if it isn't simply deleted. This has changed us into a community that new editors find unwelcoming and unattractive, especially as they often experience the interaction via an edit conflict that loses their next edit in favour of the templating.

The advantage of this theory is that it is fairly easily reversed – many templates could easily be replaced with unobtrusive hidden categories that would enable other editors to find orphans, uncategorised articles and so forth, but in a way that came across as collaborative rather than snarky.

There are conflicting theories; Both that templates attract new editors by identifying things that they can do, and that templates deter new editors from editing templated articles; And also that templates need either to be added as quickly as possible to new articles in order to be seen by the editor before they leave, or that the quicker a new article is templated the more likely it is that the author will be driven away.

A useful research project would be to test these conflicting theories.

The greying of the pedia

We are a much-loved source for the mobile generation, but not one they can readily edit. As any regular UK meetup attendee can confirm, the community, or at least parts of the community, is rapidly aging. The days when the typical new editor was a 14-year-old are long gone. If it's true that teenagers typically have an 18-month wiki-career whilst silver surfers stick around until death, senility or banning, then the community is transitioning rather than uniformly shrinking. Personally I like to see the energy and fresh perspective of the young, but the silver surfers give us a viable future, and move the age profile of our community more into line with other large volunteer communities.

Another concern is that older more experienced editors may be less tolerant of newbie mistakes, either because the age gap is too great, or because the wiki age gap has grown and the experienced editors forget what it is to be new here.

Snark

There are various conflicting theories as to what constitutes snarkiness or incivility and more pertinently whether it is increasing or even decreasing over time. One study based on editors' initial experiences showed an increase in negative as opposed to positive early experiences between 2007 and 2011, I suspect that much of this could be explained by the rise of templating and reversion of unsourced edits – behaviour that many editors consider perfectly civil and reasonable. Other sources of snarkiness include the drama boards (WP:ANI, Arbcom etc) and the often heated discussions that emerge about almost any subject that could be controversial in real life. I'm not convinced that the drama boards involve more than a small minority of editors, and anyone coming to a global site should not really be surprised to find that a contentious real life subject is contentious here. So my suspicion is that our edit conflicts when two people try to edit the same section at the same time account for more of our perceived snarkiness than all the drama boards combined. But it would be interesting to see someone research this. However despite our policy of no personal attacks, we have had editors comparing each other to excreta and various other gross incivilities that would be considered juvenile and unacceptable in most real life organisations. Whether that is more or less common in recent years would be a fascinating topic to study.

Jargon, cliques and policy bloat

Any community will in time develop its own jargon and cliques, a rules-based community risks having a bloated set of rules that become ever more impenetrable to the outsider. This is a natural part of the human condition, but we can consciously try to mitigate this, both by assuming good faith of all, especially those who we haven't dealt with before, and by simplifying our policies and procedures.

Gender

The community is very much aware that we are predominately male, significantly more male than our readership or the Internet in general. This has been a big topic of discussion for a couple of years now, without making much progress. One risk is that all we may have achieved so far is to deter potential women editors by hanging big signs up saying "women have problems here" whilst simultaneously deterring some male editors by discriminating or considering discriminating against them. My take on the gender issue is that it exists, and we haven't yet worked out why it exists. Asking those women editors who haven't been deterred from editing here to tell us what deters other women may not be the best solution, my preference would be to ask the sort of women who otherwise meet our core demographic of educated altruists to look at our site and tell us what makes it relatively uninviting to women. But my bet is that if we looked at the newbies who lose one of their first five edits to an edit conflict we'd find that proportionately more men would see that as a challenge to overcome whilst women would tend to see it as rejection and leave.

One thing we don't know is whether the gender skew is part of the reason for the community going off the boil in 2007 as we don't know what our gender skew was in that era. But it should be clear that resolving the skew and making the community more attractive to female editors would be one way of resuming growth, and well worthwhile for other reasons.

Ethnicity

If the gender skew is hard to measure or fix, at least it is widely recognised. We also have an ethnicity skew. It may well be a deeper and more complex skew than our gender one. It is harder to measure accurately, and is interwoven with our tolerance for less than perfect English.

Fun and learning styles

One of the changes that has come over recent years has been reduced tolerance for different learning styles. Despite a survey showing that "because its fun" is one of the biggest reasons for editing, the community has drifted towards seriousness. Notable examples include the deletion of "secret pages" and other userspace games which some new editors used to use to hone editing skills in userspace before experimenting in mainspace, and the end of the April Fools consensus to "keep it out of mainspace" with some admins trying to enforce "only in userspace".

"Cosmetic" edits

In 2010 policy was changed to "cosmetic" edits should not be done by bot. Confusingly, in the jargon of wikipedia cosmetic means an edit that doesn't change the appearance of the article, as opposed to edits that don't change the real meaning of an article but only change the appearance.... Aside from causing various heated and sometimes toxic arguments as to when an edit is cosmetic or actually changes the appearance of a page, just maybe not on the screensetting or browser that the critic uses; This policy has almost certainly lost us many minor edits that have to be saved up to be done at the same time as edits that are not cosmetic in wiki terms but which make at least a cosmetic difference in real terms.

Conclusion

Having listed so many factors that may have contributed to the change in the community it might seem surprising that while writing this I renamed it from Going off the boil to Going off the boil? But there is good reason for that, one of the most clearcut factors, the rise of the edit filters reducing our edit count is a measurement problem; Whilst the decreasing quantities of vandalism and typos are a huge success, but a success that in turn gives us a new problem. The continued success of our annual fundraisers shows that people are increasingly willing to help us, we just need to find different ways to recruit editors now that we can no longer rely on vandalism and typos to recruit them for us. Also some of the factors are limited and should not be extrapolated to a point where the pedia dies. In the cases of the rise of workplace internet filtering or the shift to reverting of unsourced edits much of the change has happened and crucially much of our remaining editorship will be unaffected by this. It would be tempting to try and weight these elements by their impact, ease of fixing or even their acceptability. My own preference would be to focus on trying to understand the ones that need research, and fix the ones that can easily be fixed, all the while remembering that lots of very well resourced hardware companies are trying to build products that are both portable and easy to edit with. And of course there is still the possibility that the visual editor might eventually be fit for purpose. So there is a possibility that the community could be reenergised, a fresh generation of Wikipedians be recruited and the community be very much back on the boil.

More generally I hope that this summary illustrates some of the complexity of the situation, and the perils of simplistic measurements and solutions. But we should take comfort that so many of our problems are the problems of success.

See also

• Wikipedia:Imminent death of Wikipedia predicted (humor essay)
• My presentation of this at Wikimania 2014 (starts 37 minutes in)

References

• User:Brews ohare/What happened to Wikipedia in 2006-2007? has various charts showing the decline in raw numbers of edits and so forth, however those charts do not take account of changes such as the edit filter and the increased speed of vandal reversion.