26
$\begingroup$

One great use of this website I discovered is digging through the highest voted answers authored by users with high reputation. I've discovered some really interesting discusions this way, and I recently developed a data explorer query to weed out the "most interesting" users on the site. I thought others might be interested and wanted to share my new tool.

Query

The cv_user field is a direct link to the "answers" tab of the appropriate userpage, sorted by vote score. Also, I filtered out results from the "best statistical jokes" and "best statistical quotes" questions because I'm only interested in technical discussions.

My intention here is not to stir up a competitive environment or anything like that, simply to highlight users whose answer histories are rich with interesting discussions.

Enjoy!

Update: Per whuber's suggestion, here're some slightly modified versions of the above query

  • Answers that received more than a threshhold of votes (default = eight) within a limited time period (default two weeks) of being written -- Query
  • Users who have submitted questions meeting the above criteria -- Query
  • The unique questions associated with answers meeting the above conditions, sorted by the number of upvotes the "best" answer was able to achieve within two weeks Query

If anyone else has other suggestions, I'm taking requests ;)

$\endgroup$
  • 6
    $\begingroup$ (+1) A quick glance at the output suggests you might also want to filter out cartoons and famous statisticians. Please consider, too, that to some extent votes are a measure of popularity and visibility more than "interest." Depending on what you like, you might prefer to highlight answers (even with moderate upvoting) that have a lot of $\TeX$ and figures within them--but finding them with a SQL statement might not be easy! Consider perhaps looking at question upvotes too. $\endgroup$ – whuber Aug 13 '13 at 21:05
  • $\begingroup$ filters added, thanks. I'll think about what I can do with the tex/figures idea. $\endgroup$ – David Marx Aug 13 '13 at 21:08
  • 6
    $\begingroup$ I should also add that age of answer is an important covariate: older answers (especially those posted during the first two weeks or so) tend to have accrued many votes purely from more exposure. Although we could quickly get into murky details of how to model and measure "interest," I believe that controlling for exposure (as measured by time elapsed and maybe page views) may be relatively simple and useful for identifying well-crafted newer posts that haven't yet risen near the top and help promote the efforts of newer members of our community. $\endgroup$ – whuber Aug 13 '13 at 21:12
  • 3
    $\begingroup$ @whuber added some modified queries based on your suggestions regarding the "age" component. $\endgroup$ – David Marx Aug 14 '13 at 4:12
  • 1
    $\begingroup$ Filtering out 'community wiki' questions altogether might give a very different profile, and a much higher proportion of answers that are statistically helpful, since a lot of cw questions are opinion based (about preferences rather than statistical concepts) $\endgroup$ – Glen_b -Reinstate Monica Aug 14 '13 at 21:36
  • 1
    $\begingroup$ If you're looking for the most consistently insightful user on CV, you don't need a SQL query, it's cardinal. See whuber's comment here: How should a new user navigate cross validated to learn more about statistics? The second most consistently insightful user would be whuber. If you want to develop a measurement, you might be interested in my answer here: How should performance answering questions best be evaluated? $\endgroup$ – gung - Reinstate Monica Aug 15 '13 at 2:18
  • $\begingroup$ I don't know if SE's data site will let you access the views associated w/ a question at different points, but if you could, I think the best SQL query would return the # of answers above a threshold (8 net upvotes seems fine) w/i a certain # of views (say 200). That is, is the number of views when +8 is reached minus the # of views when the answer was posted < 200? This count would be divided by the # of answers that have been posted for >200 views to yield a percentage. That % may be the best that can be done, but note @chl's comments to my answer about how to evaluate performance. $\endgroup$ – gung - Reinstate Monica Aug 15 '13 at 2:21
  • $\begingroup$ @gung, assuming it's not possible to look at vote count over time (Maybe it is. Idk.), it would be almost as good to use (# of +8 answers w/i 2 weeks)/(# of answers to questions with >200 views). As you know, most of the "action", in terms of views/answers/votes usually happens near the beginning. This would mainly miss anyone who gamed the 'Necromancer" badge and a relatively small handful of other answers. Of course, there's no simple way to correct the time-varying voting rates whuber refers to. BTW, thanks for the official gung Top 2. Why not create and release the full gung Rankings? ;-) $\endgroup$ – Macro Aug 15 '13 at 14:55
  • 1
    $\begingroup$ @Macro, you're honorary #3. I just didn't want to inflate your ego. $\endgroup$ – gung - Reinstate Monica Aug 15 '13 at 14:58
  • 2
    $\begingroup$ Thanks @gung! You're #3 in my official rankings too. Of course, I'm #1 and #2 (there's my ego :-/) but #3 still isn't bad! #4 is that kid who used to post unintelligible questions (kittens at play) and then get very indignant about it when asked for clarification. The rest of the top 10 is rounded out by people with user####### names. Cardinal and whuber are tied for 11th. My rankings are strange. $\endgroup$ – Macro Aug 15 '13 at 19:34
  • 3
    $\begingroup$ I'm glad I was able to disseminate my tool without encouraging a competitive environment ;) $\endgroup$ – David Marx Aug 15 '13 at 20:51
  • $\begingroup$ @DavidMarx is it possible to query something along the lines of specifying a value of reputation and then seeing who was fastest to reach that value in terms of days? $\endgroup$ – user25658 Sep 20 '13 at 0:12
  • 1
    $\begingroup$ @BabakP, I can't answer your question about querying but I can say you're way up there. I can also say with certainty that, without some restriction on the number of questions answered, it would be nearly impossible to outpace Michael Chernick, who quit visiting the site quite a while ago. Just look at his first four weeks. He kept up this astonishing pace for months. $\endgroup$ – Macro Sep 20 '13 at 18:49

You must log in to answer this question.

Browse other questions tagged .