I've been thinking about semiprotection lately, and it's actual effect. It's undoubtedly true that semiprotection vastly reduces the amount of vandalism to any given article. The question, though, is does it reduce overall vandalism. Is semiprotection just like The Club, keeping your car from getting stolen because the thieves move on to the next one, or is it like more cops on the street, actually reducing the incidence of car theft/vandalism.

Put another way, if someone comes to Wikipedia and attempts to vandalize an sprotected article, do they just decide it was a stupid idea anyway or do they find another, unprotected article to vandalize?

Probably not. But if we can prevent BLP vandalism by forcing them to vandalize Pokemon articles for instance, then that's a good thing.

But that's a defeatist attitude. If you semiprotect everything but the "autoconfirm playground," it may be that the would-be vandals will leave Wikipedia completely, and move to defacing library books or tagging buildings with spraypaint. Or throwing rocks through the windows of empty houses now owned by banks who evicted deadbeat owners without wondering if they might be cheaper than 24-hour property watchmen....

Economically, there are always equivalent alternatives to any activity.

Thanks, you encouraged me to read that article and find a free image for it.

I'm sure there's a level of cannibalism, but I can't imagine that it's 100%. I suspect that, in the case of really malicious BLP vandalism, where the vandal's intent is specifically to harm a specific person, the cannibalism rate would be closer to 0%. And even where the cannibalism does exist, Kevin's point above is bang-on.

What would be really great is to try substantially increasing our level of semi-protection - maybe by applying it to an entire class of articles, numbering somewhere in the six figures - so we could collect some data to answer your question.

Great idea. Did you have a particular class of articles in mind? Say a 3 letter category starting with ... B ???

Having involved myself with the vandalism from both sides (perpetrator and RC patroller) I can tell you that most vandalism is a crime of opportunity, meaning that someone reads an article, says "hey cool, what if I add this little bit of nonsense," and does so. The more mainstream the article, such as countries or common fruits and vegetables, the stronger the temptation, and the more people will be drawn to that article. Lower profile articles get much less vandalism. Therefore it is necessary only to semiprotect the most high profile articles in order to accomplish a significant reduction in the rate of vandalism.

While editing under the "Crystal whacker" alias I focused almost entirely on chemistry articles. By far the most frequent vandalism was to element articles, such as hydrogen and helium. I watched (using "recentchangeslinked" on list pages) hundreds of chemistry articles, but mostly the lower profile articles such as hydrochloric acid or acetic acid -- chemicals that are plenty common, but don't attract the attention of immature teenagers -- got left alone. I saw a proposal in February 2009 on Wikipedia talk:Wikiproject Elements to semiprotect all 120-ish chemical element articles in order to root out the vandalism. I had "retired" Crystal whacker, but I would have supported that move. I strongly believe that semiprotection should be used more freely, and not just on BLPs.

However, some people are motivated to vandalize whatever article they can get their hands on, and will not be stopped no matter what, unless Wikipedia finally grows up and makes it harder to create throwaway accounts. One variant is an individual who simply jumps off the deep end and wishes to cause mayhem. Another variant is what someone observed months ago on an administrators noticeboard, that vandals attack the first article linked in Barack Obama's article because they cannot vandalize the Obama article itself.

Overall, I would say semiprotection deters some vandalism but not all; but on balance, I think the deterrent effect is significant and should be recognized as helpful to maintenance of clean articles.

I think semi-ing works pretty well - it is all about economy of limited resources (time), and numerous articles just copping a caning over and over. I pretty much semi- anything which is copping greater than 1 vandal edit a day that I come across, especially large articles. If someone wants to lift it, I am happy to go with the flow too and discuss
Cas

There are 3 million articles and 14 million other pages on wikipedia. There are 916 active admins, of which only 21 have lifetime protection totals greater than 1,000, who have protected pages 178,976 times since December 23, 2004.

All things being equal, it would take 17,000 active admins monitoring 1,000 unique pages each in order to properly be able to administer semi-protection based on the criteria you described above.

Of course, all this assumes that the greatest threat to high-profile articles is simple-minded vandalism. I would like to see someone dream up a system that would protect BLPs from malicious, POV-pushing admins (see "Citation needed."

Yep. A perfect example of somebody offering an unrealistic solution to a continuing problem.

The next level up is have a bot do it. If Cluebot reverts IP edits on an article more than 10 times a week, or day, or whatever, have it slap a sprotect on it for a month (or 6), and then notify somebody. Or pick your own outrageous parameters.

The REAL problem here is (again) the fact that there's a die-hard fraction of the community which wouldn't sprotect an article on ANY flatly-defined-beforehand parameters for IP vandalism. If you ask them what their limits are, they'll say "I don't know, but I know them when I see them." Which actually means, they'll agree to sprotect when they're bored and tired of reverting their own favorite watched article, but don't give a flying *&%$ about YOURS. And as long as there's enough free help to do the rest, and enough people who actually enjoy mindless reversion of scatology, like killing zombies in a video game, there's no help for this.

I have previously given some data on articles on chemical elements to show that they are heavily vandalized, and the number of them which is sprotected is going up in fraction every month. But a motion to sprotect all 100 odd elements didn't pass at Wikiproject elements. Duh. They're going to do it as they feel like it, until finally it's done. How stupid is that? Much the same thing is happening with U.S. presidents. But oddly, the same chemical element articles are NOT IP vandalized on the French Wikipedia, which happens to be one of the Wikis that the studies point at, to prove that IPs contribute more than they detract. Yes, the French have the same chemicals we do, but thanks for asking. The explanation is some other cultural thing.

Lar and I have both suggested we have a complicated test, whereby we look at all Cluebot reverts of all en.wiki articles starting with "Ba.." or something. That would be thousands. Then sprotect them all and look for the next month and compare. You could even unprotect and watch the reversions go up again. But it's actually a waste of time without consensus on what you would do with this data. No matter what you found, there's still be a fraction of die-hards who it wouldn't be enough for. It's not their problem, you see. Vandalism on WP is like pollution which affects everyone, but each incidence of which, doesn't spread much. It's practically impossible to get people to care.

Hmmm. bot idea is an interesting one. an auto monitoring of IP edit followed by a revert...
Cas

and you could do a study....


setup two wikis... with identical content... functionally similar domain names... subdomains that rhyme, perhaps?

semi protect every page on one....

and don't on the other....

then, wait......... and..........

Yes, but how do you factor in the IP who is the subject of the article added by the person with an account and a grudge?

Anyway, I've been working on an interesting experiment at Category:User-created public domain images from October 2006. It is a tracking category of about 25,000 images that are all around three years old. I've been patrolling the category, tagging copyright violations, shock images, etc and I'm up to letter J and have tagged over 1,000 images as violating some image policy. Now I accept that File:Katherine_E_Scharhon.jpg is a professional headshot and should be tagged as WP:CSD#F11 or even that File:Kalim.jpg is obviously a non-free screenshot, but when I ran into a three year old mis-tagged shock image earlier this evening (Google "tubgirl" if you have a strong stomach), I came to the conclusion that despite years of admins continuously enforcing image and copyright policies to a greater degree than NPOV or NOR have ever been enforced, our image compliance is still woefully deficient. Any solution, to any problem, BLP, images, nationalism, etc, that relies on education of the admin corp or individual enforcement of policy is bound to fall short by a wide margin simply due to the lack of manpower to cover all of the articles.

And I'm also not convinced that universal semi-protection is the "best" answer. I remember a comment from Kelly Martin [1]
I can't help but feel that a proper test with a control group would be required to determine whether similar things would occur with semi-protection and that some implementation of flagged revs would be superior to semi-protection in that it would eliminate the advantage repeat offenders could gain by registering an account and making enough edits to get through the semi-protection (remember, flagged revs is based on a human review of every edit, semi-protection has no such review process). Well, more food for thought I suppose.

Argghh. This is the study on the French language Wikis and (trust me, or check it yourself) Francophone IP editors in Belgium and France don't behave like their counterparts on en.wiki


I never spun nearly universal semi-protection (obviously you need some IP proving ground to let people get the edits to get registered) as the "best answer." Oy, the best is ever the enemy of the good. It's simply part of the answer. It's not meant to be a replacement or alternative for flagged revisions. It's simply something easy we can do NOW. Unlike flagged revisions (that perennial red herring to distract us from the fact that nothing substantial is being done about this problem).

To me, the objection that IP vandals, when blocked, will simply be more tricky and register, is a non-starter. If you register, you blow an autoconfirmed account with every simple vandalism, and who has the patience to keep doing that? You have to wait 4 days between vandalisms, or else you have to keep records and let accounts ripen in parallel and you have to have an infinite pool of IP addresses and paid email accounts to do all this with. Most vandals are lazy, stupid, and (by their very nature) incapable of anything but nearly immediate gratification. If you think you're going to turn most of them into CIA opperatives by semi-protecting articles, forget it.

The metaphor for all this is the idea that if you let police use radar guns to catch speeders, all it will do is make speeders slow down in places where a cop car will be hiding, or else make them buy radar detectors, forcing the cops to go to lidar laser units in a never-ending escalation of detection.

To which the answer is: so what? Some people will slow down in areas where radar using cop can be hidden. This is good. Some nerds will buy lidar detectors, but the idiot kids and drunks who you want most to detect when they drive too fast, probably will not.

And I can push the parallel further. Suppose we can't detect heavily stealthed corporate shills, being paid to add well-sourced informational stuff into articles on their businesses and products? How is the world going to suffer from this, exactly? Is it even approximately the problem that BLP is (which I'm assuming we're going to get rid of sompletely)? And what makes you think it won't (and doesn't) happen ANYWAY?

No, it wasn't; it was a study on the English Wikipedia conducted by Seth Anthony in 2005 and presented at Wikimania 2006. A study on the French Wikipedia, a low-traffic wiki with a totally different culture, would be almost completely useless to the purpose of comprehending and predicting editor behavior on the English Wikipedia.

Semiprotection effectively stops random, opportunistic vandalism, which is the easiest type of vandalism to detect and the least harmful type of vandalism (much of it is just blankings or replacement of articles by obvious nonsense; occasionally it is "harmful" in the sense that it puts profane words or images in places where people are not expecting them, but fundamentally that's a minor harm at best). A substantial fraction of that sort of vandalism is already being autoreverted by bots anyway, so semiprotection doesn't really help much there except insofar as it reduces database size and server activity. Opportunistic vandals aren't going to take the time to register an account in most cases, and they certainly aren't going to wait around for their account to autoconfirm.

Semiprotection does almost nothing to deter purposeful vandalism; these people have mechanism to create accounts by the bucketload (Mediawiki's captcha methodology is easily defeated by any of several approaches) and can easily maintain a stockpile of autoconfirmed accounts for whatever they're trying to accomplish. Furthemore, the vandalisms these people introduce are often subtle enough to not be noticed by vandalism patrollers anyway. Flagged revisions is likely to help at least somewhat here, although whether it does or not will depend on how approval rights are handed out and supervised. Given Wikipedia's track record for user right management, it will not be done effectively.

I'm reasonably certain that there are vandalism patrollers who are also vandals (that is, they vandalize with one hand and remove their own vandalism with the other), with the goal being to obtain adminship. I'd put the odds that there's at least one admin who got there via this technique at even money. There's also any number of Internet marketing entities (of various levels of scrupulousness) constantly trying new ways to insert their preferred content into Wikipedia. Some of them are quite blatant, others much less so. My experience as a checkuser is that some of these entities maintain dozens of apparently distinct identities on Wikipedia (either good actors, or else multiple people at the same firm, probably more the latter than the former). Given that advertisers are paid to manipulate public opinion, it should be of no great surprise that at least some of them good at manipulating the Wikipedia community as well. I've also seen some evidence of ad firms directing their employees to edit from home, via 3G links, or via public access points (e.g. Starbucks wifi) in order to conceal their association.

Wikipedians have a tendency to assume that they're smarter than the vandals and all the others who want to manipulate Wikipedia content to nefarious ends. They'd be wrong.

True enough. Is this published? Because it's not the published study that everybody quotes whenever this debate comes up.

I have no idea if Seth ever published. He's still around (I see him on Facebook occasionally) but I do not believe he actively does anything with respect to Wikipedia anymore.

Well, then, what can I say? Except {{fact}}.

I did some analysis posted here on WR, of English and French chem element articles and English US president articles which I cannot find now with the search engine (probably tarpitted like all my best writting) and it was very obvious that sprotection stopped most vandalism (I tracked not just IP vandalisms, but ALL edits, and classified them by "hand"). I put out a challenge for anybody to find an article of any decent size which had less than 20% (IIRC) of its edits IP-vandal edits, if it was unprotected.

Ah, HERE we are:

http://wikipediareview.com/index.php?s=&sh...ndpost&p=160672

You can't enter more than one word at a time into the search box unless you want to to use that as a "phrase." I'd like to know how to use this site's search as a Boolean AND. Put the words separately in quotes, as you do with Google?

That's pretty much what I would expect. Most vandalism is opportunistic vandalism and the very low barrier created by semiprotection increases the opportunity cost enough that it's not worth it. If your goal is to put "JOSH IS A FAG" on the Internet, it doesn't matter what article you do it to; at the same time, you aren't going to put a lot of effort into this enterprise, either.

What semiprotection does not do is significantly impede purposeful vandalism, which should be of greater concern anyway since the whole "biography of living persons" problem is primarily due to the relative ease of libel offered by Wikipedia, which clearly (to me, at least) falls within the scope of "purposeful vandalism".

I did find the link:

http://wikipediareview.com/index.php?s=&sh...ndpost&p=160672

Seth's presentation is abstracted here. There may be a videograph, but I'm not arsed to dig through all the crap on the WM06 site to try to find it.

I went through the powerpoint slides and found nothing about semi-protection. Seth wasn't even clear if any of the articles he was looking at were sprotected. I assume SOME were not, since there were IP edits, but if some were and some weren't, that would skew "high content" edits in favor of nameusers (who made 70% of the total, but is that people they make 100% of the good edits, due to making 100% of ALL edits, in some sprotected articles in the mix?).

There's 5% vandalism but I have no way to know what that means. Was all of it done by IPs? What fraction? I was amused that none of the high content editing in the small sample came from admins.

Basically, I can't find anything here which is relevent to the "sprotect most of WP" argument. And with this analyses of 50 edits at a time, I actually have more data in my own WR look at element wikis that are not protected. And it clearly shows 30-40% of the edits are vandalisms. In English. Less than 10% in French for the same articles.

Semi-protection does jack squat, as can be seen by the modus operandi of the Virgin Killer vandal, current incarnation is "Fruue".

To reach autoconfirmed status, a user account myst be 4 days old and have 10 edits. That's it. So...

User account created 15:03, 19 July 2009

1 hour past the 4 day mark (how's that for OCD zeal?),

http://en.wikipedia.org/wiki/Special:Contributions/Fruue

he picks a random user page, makes 9 random text addition/removal edits, then reverts it all back on the 10th. Next edit to to a random reverter or two of his past vandalism (I've gotten it a few times). Finally, it is off to the articles for Virgin Killer and the Internet Watch Foundation and Wikipedia.

Semiprotection didn't exist in 2005 (it was added in December 2005 IIRC), so of course he didn't mention it.

I guess alot of this depends on numbers - i.e. proportion of impulsive IP vandalism to more sneaky stuff. At least removing some of the former allows more time to review the latter. Plus easier to block accounts than IPs, no worries about inadvertent blocking of potential constructive editors.

This is another reason why I like FA/GA as mega/consensus review/flagged revision-type things to act as staging points, and compare down the track.

Some interesting points though..
Cas

Okay, but we still have the problem if what study MBisanz was talking about, above. I assumed the French one, you assumed the Seth X one. Obviously we're both wrong, because neither of these showed what Bisanz claimed:




The Seth llys/Anthony (user:Sethant) study shows no such thing-- name users who aren't admins do the heavy content contribution lifting. On the other hand, he finds amazingly little IP vandalism. Nothing like what I see in the kinds of articles I edit. But I should try it using SPECIAL:RANDOM (try it). "Proper sampling techniques" Seth's edit groups analyses are too small (they're even smaller than mine) for that. Perhaps he describes how he picked his articles, but it's not in his powerpoint summary.

The French study interestingly found that IP users contribute more of the lasting content than name users, although with time this drops for IPs and increases with nameusers (perhaps as bad IP-users are blocked and good ones register).

I find that IP vandal edits are typically 20-40% of large articles with common "encyclopedic" topics (science, history, geography, politics). It tapers a bit for shorter articles and stubs, and very technical stuff.

But test it yourself. Enter special:random to the searchbox until you get your first unprotected articles more than 2 pages long. Compare the vandal edits on the last page to the name-user vandalisms. Then look at who has been contributing content.

I had to do this 10 times to get an article decently long:

Goulburn,_New_South_Wales

The last 50 edits include those by about 7 different IP vandals. Roughly half the edits are no good at all. Many name users are there simply to revert the vandals. What a frigging waste of time. That's Wikipedia.

Okay, again (takes 7 more tries to get more than a paragraph or dab)

Laurie_Anderson

BLP of a singer:

Only one IP vandalism on the last page!
http://en.wikipedia.org/w/index.php?title=...oldid=276479332

I have hypothesized that pop cultural stuff is less heavily vandalized than "encyclopedic" educational stuff.

In any case, I do not see why more than a couple of IP vandal edits per page should not be enough to trigger off a robot sprotect.

If I've parsed the negators properly, I find that I do not disagree with your opinion.

Litotes!

Hahaha, can't knock a town with the world's largest cement sheep. Shoulda been in New Zealand not Oz though....

Why? Oz has way more sheep than NZ.

Well, yeah, but...y'know...

Seriously, I have no idea but might have something to do with the density (sheep/sqm)...

You are attributing the quotes wrong. That was a comment of Kelly Martin's that I quoted in my post, unless you are talking about my own non-scientific anecdotal evidence?

Yup, looked at quote box wrong and it was Kelly. Well, then, Kelly, all I can say (again) is Seth's study didn't show what the French one did, and his doesn't agree with what I see on en.wiki with just as much data, and since he didn't publish, then ONE MORE TIME: {{fact}}.

It's pitiful. There are 120 odd chem element articles, and they all have (or had) the same IP vandal problem. But they've had to protect them ONE BY ONE... More this month than last month. But not as many as next month....

Thus does time pass on Wikipedia. It's sort of like watching Grand Canyon get deeper. The whitewash slowly dries on Tom Sawyer's fence....

The only entertainment for me anymore is watching them argue on the TALK page for WP:CONSENSUS. Seems they can't agree at all about what "consensus" actually means. This has been going on a loooooong time.

Seth's study was in 2005. It's now 2009. Nobody reasonably believes that Wikipedia has not changed in four years.

My point is that I think the "evilosity" of IP-identified editors is overstated; at the same time, IP-identified editors are certainly a significant source of low-grade vandalism. If Wikipedia is serious about being a user-contributed encyclopedia, it needs to find some way for casual readers to contribute to the project while at the same time decreasing its "surface area" (as the security people put it). So far, the glorious crowd-sourced wisdom that rests within the Wikipedia community has yet to come up with a solution for this one.

As I've said on Wikipedia Review and elsewhere, the main problem is that the current implementation of semi-protection sucks. Or rather the current implementation of page protection sucks. Currently, you can block all users who are unregistered and who don't fit an arbitrary set of criteria (4 days since registration and 10 edits for the English Wikipedia) or you can block all non-admins from editing a particular page.

It's all about barrier to entry (as we know it's always possible to get through page protection with enough determination). For some articles, blocking just anonymous users would be completely sufficient. Those not lazy enough to register an account could then be easily blocked. For other articles, you might want to set the configuration to 30 days since registration and 40 edits. Or set it to allow only people in certain user groups. Or set it to allow only people with a valid e-mail address set. How e-mail addresses correspond to vandalism or mischief, I don't know, but you hopefully see my point: the issue is a lack of sufficient flexibility when protecting a page.

The AbuseFilter is sort of a step in the right direction. However, unfortunately, it was never really intended as a substitute for page protection, so to my knowledge, using it widely will cause problems.