Maegara

To keep conversations together, I prefer to reply on this page to messages left here.

Yo

Welcome!

Hello, Maegara, and welcome to Wikipedia! Thank you for your contributions. I hope you like the place and decide to stay. Here are some pages that you might find helpful:

• The five pillars of Wikipedia
• Tutorial
• How to edit a page
• How to write a great article
• Manual of Style

I hope you enjoy editing here and being a Wikipedian! Please sign your messages on discussion pages using four tildes (~~~~); this will automatically insert your username and the date. If you need help, check out Wikipedia:Questions, ask me on my talk page, or ask your question on this page and then place {{helpme}} before the question. Again, welcome!

I get the sense that you know much of the above already, but welcome all the same.--Father Goose (talk) 07:27, 15 March 2008 (UTC)

Well, help would be welcome

I don't think statistical tests are the way to go, honestly. I think that the numbers actually hurt this case. Instead, I hope to develop tools that find patterns suggestive of quacking. One of the main complaints is that these methods are untested, so I would first and foremost like to test them against real accounts on RFCU—before the ip check results are posted. If I can establish a good track record there, then it should engender some confidence that these methods are reliable.

One way to do this is by making a battery of tests. When most of the tests do or do not suggest socking, I would post predictions. Over time, it should be possible to determine which tests are most strongly correlated with sock puppets among different classes of socks—and there are at least two classes: regular and proxy socks. Proxy socks are the difficult problem, but I suspect that there are editing patterns associated with them as well. Cool Hand Luke 16:58, 15 March 2008 (UTC)

I understand where you are coming from. I've had the suspicion that even if I could have come up with a statistically significant result in the Mantanmmoreland case, no one on ArbCom would have understood the methods. Let me know if I can help in any way. --Maegara (talk) 17:50, 15 March 2008 (UTC)

Thanks. I think what I'll do is come up with a list of characteristics that we could test for. I'll let you know for feedback. Then the best we can do is apply tests to possible socks on CU.

Each time we get a match, or a proxied ip or a non-match, we could keep track of the test results for each. That is, if a result is positive, or if it places an account into the Xth percentile of editors, or whatever, we'll list that datum along with the results to determine which characteristics are well-correlated. It's a messy problem because there are several things socks may or may not do based on the type. For example, proxy socks might have less interleaving than non-proxy socks, and we can never be too sure whether a proxy result is a true sock or not. Proxy socking will be a tough nut to crack.

If a test is poorly correlated to socking, we'd reject it. We could also make new tests by tweaking possible leads. For example, if editing correlation coefficient turns out to be a good predictor, we could test whether weekday/weekend correlation is even better—but we can't really be sure any method is better until we've tested it against new CUs. Data snooping problem. Anyhow, I think experimentation with confirmable accounts the only way.

A priori theorizing about the behavior of socks is just not good enough.

I'll be coming up with a list of possible tests over the next few days. Cool Hand Luke 02:48, 17 March 2008 (UTC)