User talk:Msnicki

From Wikipedia, the free encyclopedia
Jump to: navigation, search
Editors Barnstar.png The Editor's Barnstar
for being so clear eyed, consistent, and even in tone while working on the whole Deluna page/umpteen socks investigation. Good work!Tao2911 (talk) 21:05, 24 August 2011 (UTC)

My goodness. Thank you. Msnicki (talk) 21:19, 24 August 2011 (UTC)

A barnstar for you![edit]

Peace Barnstar Hires.png The Barnstar of Diplomacy
For keeping peace and letting everybody use their opportunities. APS (Full Auto) (talk) 22:37, 14 December 2011 (UTC)

Thank you! Msnicki (talk) 22:45, 14 December 2011 (UTC)

Barnstar of Integrity Hires.png The Barnstar of Integrity
Your one of the best debaters I have seen in some time. I have great respect for you my new friend. Moxy (talk) 20:22, 10 January 2014 (UTC)

Note on search results[edit]

per Talk:Cannabis_(drug) and newspaper articles, it seems that you have some familiarity with search engines; however I should point out that the estimates from google (and other engines) are only that - estimates - and should not be used to determine frequency, especially if the ratio differentials are so tiny. A much better technique is to do focused searches of particular sources and then click all the way to the last page of result to get the actual number of hits. Sometimes the actual number of hits vs estimates are off by one, two, or even three orders of magnitude. If there can be such a swing in actual search results vs estimates, a factor of 2x or 3x is irrelevant - the accuracy of your measurement is determined by the granularity of your ruler. see Wikipedia:Search_engine_test#Search_engine_tests for more details. An example of how I used this technique is here: Talk:Deadmaus/Archive_2#Break_2.--Obi-Wan Kenobi (talk) 16:59, 15 April 2014 (UTC)

Yes, I was one of the original developers of the Bing engine, so I'm familiar with search engines, how the estimates are made and that they can be off. But can they all be off and so consistently in the ratios, even sliced and diced so many different ways? (If the counts of "marijuana" were inflated, what makes you think the counts of "cannabis" were inflated the same way?) If anyone (or if you would like to do it now) had done the research to go to that last page and show that my ratios were faulty, that would have been substantive. But to claim search engine figures could be wrong, therefore they can be ignored is just lame. I did my best and put a lot of effort into what I believe was a competent analysis and in my responses to the actual questions raised. Msnicki (talk) 17:18, 15 April 2014 (UTC)
I have, on many occasions, clicked to the end to find 22,000 results = 956, and 38,000 results = 350, etc. The point is, those estimates can vary by many ORDERS of magnitude, but your ratios were for the most part < 10. Thus, those results from a scientific perspective are irrelevant, even if they are consistent (there may be some other magic going on which causes those things to behave in that way). I don't know what the real data is, but google search won't return any results past 1,000 anyway, so any numbers beyond that are pure guesswork, so you need to constrain the space to get below 1000. It's much better to do constrained searches, as the link at Deadmau5 shows, and then you can get really robust results. In any case, I think a ratio of 1.5x or 2x is not sufficient justification for COMMONNAME, you need to look at other WP:CRITERIA in those cases in any case, since commonname esp with such broad search results can be polluted with all sorts of crap. On the Chelsea Manning case we actually mapped individual articles and studied them, that is another way to go. I'm not saying all of your searches were bad, I didn't read through the discussion nor study them in detail, but the comparison of 80,000 vs 70,000 results is meaningless IMHO.--Obi-Wan Kenobi (talk) 17:27, 15 April 2014 (UTC)
Fine. And if you had raised that argument in the debate, we could have had a lively discussion and I would have been delighted to dive deeper into the results to show that your objection did not have merit. But you did not raise that argument then and to complain that I didn't respond to an argument that was never made is just dumb. Msnicki (talk) 17:32, 15 April 2014 (UTC)
I only came across the discussion today, and I'm not complaining that you didn't respond there. On this: "But to claim search engine figures could be wrong, therefore they can be ignored is just lame." - the point is, I can't use your results and "click to the end", since they aren't constrained enough - so Google at least will never change the estimates. You need to build searches that will ultimately have < 1000 hits, in order to get relative frequency, the best way to do this is to constrain by time, or by publication. The reason I'm engaging with you here is that you seem to know about search engines, and I'm pointing out that your methodology there was flawed for the reasons I've pointed out above.--Obi-Wan Kenobi (talk) 17:39, 15 April 2014 (UTC)
Here's an example, fwiw, since you constrained by pub later on: you show 10,300 results for USA today, vs 3850 for cannabis. However, if you click to the end, you get 531 for marijuana and 484 for Cannabis, thus a ratio of 1.09 - hardly enough to call it for Marijuana. NY Times similarly, you have a ratio of 26 to 1 (93,300 vs 3,470), but clicking through to the end gives 518 for cannabis and 542 for Marijuana, a ratio of 1.04. Those are just two I chose randomly, but they demonstrate my point. I don't yet even have an opinion on the debate, fwiw.--Obi-Wan Kenobi (talk) 17:49, 15 April 2014 (UTC)
How many ways can I say this? This would have been an interesting argument if only you'd made it at the time. You didn't. If you had, I don't know I how I would have responded. For all I know, I'd have done more research and decided you were right. I do follow the data and I did the best I could to do just that. The evidence I found, using my best efforts, said what it did and I reported it. I think the close was a joke but realistically, I've said all I care to say about whether I think the data shows marijuana to be the more common term.
But while we're on this topic of how search engines work, let me explain something you obviously don't know about how they work. The last page of results you get from any search engine is NOT necessarily the last of all the possible results in the index. In fact, it almost certainly isn't, especially on any search for which the results are estimated in the millions.
Search engines not only estimate how many pages might match, they also estimate whether they've already found enough of the best to produce a reasonable result, c.f., the Secretary problem. In addition, search engines deliberate truncate results because they simply cannot allow queries to run to completion. If they did, that would open them to DDOS attacks with queries that produce millions of results or otherwise force the engine into scanning the entire index. (Ask me how I would know!) The deeper into the index, the more costly it becomes, e.g., because each page needs to evaluated (ranked) and because you switch from small index in memory to large index on disk, etc. (You have to go to the large index for "tail queries" that don't produce usable results in the small index. But you don't want to go to large index if you don't need to so you try to satisfy most queries out of the small index.)
When you get to the last page of results, there's no way of knowing if that's the last page to be found or if they just refuse to give you any more but if you're already at page 25, it's more likely the latter. It should obvious to anyone, not just search engine developers, that when someone starts clicking through to page 25 of the results, it's starting to get unlikely that there could be anything you could report on page 26 that would make them happier. So they stop those queries in their tracks to avoid bogging down the whole engine. Msnicki (talk) 18:18, 15 April 2014 (UTC)
I'm aware that they don't give you all of the results for obvious reasons - thanks - but since we can't get to the "actual" number of actual results, it's more objective to find a <1000 query (at least on Google) and compare that way. As I mentioned, I wasn't in that discussion, and I'm bringing this to you here so that next time you do a move request you might consider using a slightly better methodology, that is all... At least for google, they seem to be rather consistent, and they will sometimes return 1000 results for big queries, but for more constrained ones they will return less (claiming other results are dupes) - importantly, I've found the results don't vary that much over time so I think they're much better than the estimates which can vary from machine to machine, geography to geography or day to day.--Obi-Wan Kenobi (talk) 19:03, 15 April 2014 (UTC)
This is your opinion. Mine is different, I've explained why and you really haven't answered my objection that the last page is not a reliable indicator of the total number of matching pages. Further, you make it sound like I only did a few searches and that none that were constrained to small numbers of results. That's simply not true by inspection. I did LOTS of queries (and invited suggestions for others) and many of them returned only hundreds or low thousands of results. In fact, those were the ones that tended to show the greatest preference for marijuana. Look, I understand you would like to relitigate my search results. But I'm really not interested. You're too late. Msnicki (talk) 19:25, 15 April 2014 (UTC)
Not trying to relitigate. It's just a pet peeve of mine. Just plz avoid the big searches in the future, I think it's really misleading.--Obi-Wan Kenobi (talk) 19:28, 15 April 2014 (UTC)
Once again, this is your opinion, supported by ... your opinion. Msnicki (talk) 19:44, 15 April 2014 (UTC)
erm, no. It's part of an official wikipedia how-to. Wikipedia:Search_engine_test#Google_distinct_page_count_issues, which is directly referred to in WP:AT policy at Wikipedia:COMMONNAME#Use_commonly_recognizable_names. More importantly, since users use search engines to find things, the only results they see are what the engine returns - thus clicking-to-the-end gives a much better sense of the total space of "names" the user will be exposed to, vs the off-by-10x-1000x estimates Google throws out. You're far from the only one to use raw google results in this fashion, it happens ALL the time - but again since I read you had SE expertise I wanted to reach out - that is all.--Obi-Wan Kenobi (talk) 19:49, 15 April 2014 (UTC)
And I have explained to you that going to the last page is not reliable, either, and I've explained why. For someone who claims to have reached out because you thought I had some relevant expertise, you seem completely unwilling to consider that I might be right. The search engine test article is correct that estimates are often wrong, especially when searching for quoted strings because it's difficult to estimate conjoint probabilities, especially of exact phrases given that the engines typically only know the number of documents containing each word and the total number of occurrences in all documents, not how many also contain other words without searching. But none of my searches were for multiword queries, much less for quoted phrases. Moreover, it's just plain wrong to imply (with the phrase, "actually exist") that if you get to the last page of results, that's all there are. Whoever wrote that appears to be unaware that search engines deliberately truncate results not because there are no more but because they refuse to return more. Msnicki (talk) 20:18, 15 April 2014 (UTC)
I understand your points, but you're missing mine. Users will never be exposed to the 99,000 purported results that allegedly exist on the LA times website - google will never let them see them. It's really relative frequency within a bounded space that is more important in measuring user experience and exposure to given terminology, and looking at the possible set of results users might actually be exposed to (vs. theoretical estimated results) is a much closer estimation of relatively frequency. In other words, if I search for marijuana at the NY times in the past year and I get 100 hits, while a search for Cannabis gives me 10 hits, that's a very good indication of frequency of use and what's more, I can have a fair bit of confidence in those numbers. OTOH, stating that, over the history of the NY Times, they have... allegedly... written the word "Marijuana" 99,000 times and the word "Cannabis" 54,000 times, but by the way those estimates could be off by one or more orders of magnitude, is literally meaningless - but you presented those results as if they were the truth, and even averaged the ratios together somehow... it's all very misleading, and reasoning from bad data. My suggestions don't give you perfect results, either, but I still maintain they are much better and reflective of user experience than the raw results, and our own guidance on the issue suggests the same.--Obi-Wan Kenobi (talk) 20:27, 15 April 2014 (UTC)
Your analysis is wrong. For unaugmented single word queries, there's no need to estimate the number of matches. That's actually known. The engine does know exactly how many total occurrences of each word exist in the index and it also knows exactly how many documents there are in the index that contain that word. It'll be a 64-bit number in each case. The engine only needs to estimate the number of matches if you narrow the query. Further, truncation is done on queries that produce lots of matches but exactly where the engine chooses to truncates is not proportional to the number of possible matches. Rather, it's function of how long the query has run and the estimated quality and number of pages already found. A lot of that is just some developer's heuristic.
For example, consider what would happen if the engine had a (not unrealistic) heuristic that said, once you get to page 25 of the results stop as soon as you hit a result isn't better than the worst one we already have. If you compared two queries, one for which there were an actual 25M results with one for which there were only 1M actual results, they might very likely both stop on the very same page. At that point, you would falsely conclude that the estimates were wrong and that the actual numbers were about equal.
Admittedly, most of my queries were not unaugmented one-word queries insofar as I constrained them in various ways. The queries were one word (either marijuana or cannabis) but I added constraints on web addresses I would consider, etc. So, indeed, some of the estimates of numbers of matching pages were indeed estimates of conjoint probability. But you offer no reason (and I'm not aware of any) to think that there would be a bias in the calculation, that the estimate would favor marijuana over cannabis when that's the only difference in the query.
I think my numbers were reliable, more reliable than your "last page" comparison. And btw, I did your experiment, comparing NYT news articles in the last year. You said, "if I search for marijuana at the NY times in the past year and I get 100 hits, while a search for Cannabis gives me 10 hits, that's a very good indication of frequency of use and what's more, I can have a fair bit of confidence in those numbers." Great, because I did the search and got 3 results for cannabis and 34 results for marijuana. I'll say again, I find it funny that you come to me because you had learned I had some expertise, but then refuse to consider that I might know what I'm talking about and that you could learn something from me. Instead, the whole agenda here seems to be to convince yourself that you know more. Msnicki (talk) 21:07, 15 April 2014 (UTC)
The reason I offered is that those estimates are, as stated, estimates, that may under-estimate or over-estimate the actual number of results by one or more orders of magnitude. For example, this paper which studied Windows live found variation of 18-180% in # of results reported vs "actual" hits (they used a technique of splitting the query space to get more than 1000 results). Also see this and this, specifically "The Estimated Result Count is just that - it's an estimate of the number of results available. Due to the nature of calculation, this estimate is not stable, and can change from request to request, mimicking the similar count on google.com." If your measuring stick isn't sufficiently granular to differentiate between 60,000 and 70,000 (which, I'm arguing, such estimates are NOT) - then any such result comparing the two is scientifically MEANINGLESS - especially if the resultant ratios are, as you showed, only on the order of 1-3x. The question is not whether there is a consistent bias, it's that if the results could vary that significantly (especially given those search results may vary by many thousands on a given day from the same query!), does it even make sense to talk about a ratio. On the other hand, a bounded search - even if truncated as you note - will give a better sense of the sets of pages real users will be exposed to, given that people find things online usually via search engines, and in my experience those results are more consistent and invariant over time; if you focus on bounding to reliable sources and filtering by time, it gives you a much better sense and can sometimes reveal a much more dramatic differential. For example, deadmau5 vs deadmaus gives a 2-to-1 ratio, but if you bound the search to Rolling stone you will get over a 10x multiplier. I see your point that the truncation may have little to do with the "actual" number of results amongst the billions of web pages out there, but the estimates also seem to have little to do with the actual number of results and at least the truncated returned results are aligned with what real users will see. Why don't we agree that both methods are crap, and given your knowledge of how SE work, why not help us develop a better method? Are there backdoor APIs we can access or maybe we can give wikipedia a developers-link to the major SEs that would give us more raw data about number of results without worrying about returning actual URLs? re: the NY times search above, Google news is now broken, it only goes back like a month, so maybe using the internal NY times engine is better in those cases.--Obi-Wan Kenobi (talk) 21:46, 15 April 2014 (UTC)
Okay, I did the search on the NYT site, just as you suggested, just like I have consistently done any search anyone ever suggested was needed. I have never made any attempt to cherry pick the results. And guess what, the NYT estimates 502 results for cannabis in the last year and 2950 results for marijuana. The best way I know to understand search engine results is to do lots of queries and that's what I did and you have not. Msnicki (talk) 22:05, 15 April 2014 (UTC)
Msnicki, I think you may mistakenly believe I have an opinion about the proper name. I do not. So I'm not talking about marijuana or cannabis, I frankly don't care that much. But for the NY times, again if you click to the end you get different results: [1] and [2] - in one case it returns the full 1000 results (but won't give any more), while for Cannabis it only returns 99 and won't return any more. Now THAT is a robust result, with a 10x differential - and we know it's a real 10X because the website actually gave us 1000 URLs vs 99 URLs. Especially with a low hit count like 99, I now actually have a lot of confidence that NY times simply doesn't have more than 99 results to give me. I never suggested that you cherry picked the results and I never believed that you have acted in anything but good faith. My discussion here was about methodology, not the results of that particular discussion.--Obi-Wan Kenobi (talk) 22:16, 15 April 2014 (UTC)
Further, on NY times, see [3] vs [4] - when limiting to last 30 days you get MORE results. Again, I'm highly suspicious of any numbers > 1000 returned by search engines. The example above is perfect - since we don't know, actually, how many hits there are for marijuana, but we CAN say with confidence it is > 1,000, because we've seen with our own eyes at least 1000 URLs. OTOH, we can also say, with a fair amount of confidence I believe, that there are only 99 for cannabis. The fact that this is a 10x differential makes this quite strong. If the differential was much smaller than that, I wouldn't see a case. But 10x, now that is a very strong case, at least for that paper. Show the same for 10 other papers and you have it locked (at least for media sources, then you get into medical literature, blah blah). In any case, I have much more faith in URLs actually delivered to the user that I can see with my eyes. If one query promises 1,000,000 URLs but delivers 500, and another query promises 3,000,000 URLs but delivers 100, for the average user the "500" result is more meaningful - it means more pages they are likely to be exposed to, in the long run.--Obi-Wan Kenobi (talk) 22:28, 15 April 2014 (UTC)

────────────────────────────────────────────────────────────────────────────────────────────────────

further research here [5], one of their findings verifies what I have found, which is that if you click all the way to the end the number of "hits" displayed seems to be more accurate and consistent. Also note their findings re: clustering of results across different days - for Google for example 11% of queries returned ~5x the number of results for a few days in October 2009, and another cluster of queries went upwards of 8x. While you may be right that the search engine knows how many matches there are, at least for Google, it's not at all clear that the number they are returning is the same as that number of matches in their index, and if the index is being updated (or your query hits a part of the index that is being updated) the results can be off by a lot. In any case, I'd love it if we could update our google search instructions with your insights, and more importantly if you can teach us better secrets to get better results. I've thought for a long time we should have a standard policy for determining commonname through google/bing/etc searches, it would be great to develop one like that, esp if it could use a backdoor api or something.--Obi-Wan Kenobi (talk) 22:16, 15 April 2014 (UTC)

Tokai - sorry about that![edit]

For some reason the AFD page is VERY slow at showing updates to AFD discussions, new comments, etc - when I looked at it, I didn't see Tokai discussion as closed, but as still being open. As soon as I realised it HAD been deleted, I reverted my reversion - apologies! Mabalu (talk) 21:46, 27 April 2014 (UTC)

No problem at all. I think I've had this happen to me, too. Sorry to be slow responding. (Just busy.) Msnicki (talk) 21:30, 30 April 2014 (UTC)

Move review notification[edit]

Because you participated in the most recent discussion regarding the proposed move of Hillary Rodham Clinton, you are hereby notified per Wikipedia:Canvassing#Appropriate notification that the administrative determination of consensus from that discussion is being challenged at Wikipedia:Move review/Log/2014 May. Please feel free to comment there. Cheers! bd2412 T 19:22, 20 May 2014 (UTC)

Wikipedia:Articles for deletion/At (Windows)[edit]

You are invited to join the discussion at Wikipedia:Articles for deletion/At (Windows). Thanks. — Dmitrij D. Czarkoff (talktrack) 21:09, 15 June 2014 (UTC)

Thank you[edit]

I wanted to express appreciation for you using good editorial judgment regarding Rich Salz, having heard all the arguments. That's the sort of patient consideration Wikipedia needs.Pigdog234 (talk) 16:26, 23 June 2014 (UTC)

And thank you. I tend to argue strongly, but I also listen to strong arguments and they often cause me to change my mind. Thanks for reaching out. Msnicki (talk) 16:32, 23 June 2014 (UTC)

ANI[edit]

Information icon There is currently a discussion at Wikipedia:Administrators' noticeboard/Incidents regarding an issue with which you may have been involved. Thank you.

Doing this for the editor that neglected to do this. Dennis Brown |  | WER 20:11, 16 July 2014 (UTC)

Okay, thanks for the heads up, though it looks like you all have it under control. I probably was a little rude and he hadn't also been rude, he'd have gotten an apology. But you guessed correctly, I'm just worn out by his behavior, including the careless accusation of vandalism and all the whining. Msnicki (talk)
It ended up getting unclosed, and I just put it on the table, the fact that I knew they were a paid editor socking. Ponyo was nice enough to run a quick CU and linked two, so both are indef blocked. Ironic. But on the other point, I understand getting pissy sometimes, I do myself, but we both just need to be careful with it. As I said there, unless it is a pattern, I'm not inclined to do anything else, although it isn't up to me alone. Dennis Brown |  | WER 22:37, 16 July 2014 (UTC)
Ah, yes, WP:BOOMERANG. Yet another reason why I try very hard to avoid ever going to ANI even to defend myself. Better to let things simply take their course. And, yes, I totally get that even if I am frustrated, I need to watch my tongue. What can I say? I am not a perfect person and I'm aware of at least some of my flaws, like this one. Sadly, despite my awareness, the flaws persist and still pop out at the worst possible times. Thank you for understanding. Msnicki (talk) 07:25, 17 July 2014 (UTC)

A kitten for you![edit]

Kitten-stare.jpg

Thanks for finding additional notability evidence for ConnMan

Yacz (talk) 20:24, 18 July 2014 (UTC)