Talk:Usage share of web browsers

Apples, oranges, ravens and office desks

There has been talk of such things here. I understand that we are only measuring one thing, viz how many web visitors use this, that, or the other browser. The article goes into some detail as to how this is actually very difficult to measure. Then we take all the known sources of major information and provide detailed figures going back months and years from each of them. In between, we try to summarise. We choose five major measuring sources, that each provide data based on billions of web visits in the most current month, and it is surprising by how much they differ. So we provide a summary. The summary includes medians. End of. If we had a jar of sweets and asked some Americans to estimate how many there were, then some Europeans, then some people who look up things on the Wikimedia sites, we'd get different estimates. What could we do? Summarise and show a median. If we had estimates from people interested in X, Y or Z, we could include those. It wouldn't affect the fact that all the measurements were of the same thing, just different subsets from the true, global, (unmeasurable) world population. The more we got, the better. I suppose if some of the estimates were from a society for the blind, or from the association of sweetshop assistants, we may have to think about those. I don't see any apples and oranges problem here. --Nigelj (talk) 22:37, 12 December 2011 (UTC)

I certainly do, some of the numbers are adjusted with CIA census data to estimate users from which there are not any hits. Others such as wikimedia use raw hits, and to not try to account for regions which don't have access to their servers, or where there servers are not very well used. As with the article I linked in my !vote these kinds of difference mean effectively that your measuring different quantities. I do believe there is an apples and oranges situation here. Thenub314 (talk) 23:30, 12 December 2011 (UTC)

Nigel, I know stats can be confusing, but you do not appear to understand the arguments made against the use of this "median". inter alia, these different measures are looking at different jars, not the same one several times - your statement that "we are only measuring one thing" is flat out wrong. The targets can be described in vague English in the same way, but that's just a distraction. Furthermore, some of these jars do not appear to be relevant in the real world. For example, how many sites independent of the Wikimedia foundation cite Wikimedia stats as representative of the Internet? It's probably close to zero. Per WP:NPOV (seriously, read it) we should not give these stats any weight at all. Clicky and w3counter seem pretty obscure too.VsevolodKrolikov (talk) 23:51, 12 December 2011 (UTC)

The stats sources are estimating the actual usage share of browsers by measuring the usage share of a biased sample. The jar in this case is the actual usage share of browsers. The estimate is the measurement of their biased sample. In the case of a crowd of people estimating the number of jellybeans in a jar, each person is using their own mental model to determine their estimate. But just because their models are different does not mean that they are estimating different things -- they are all estimating the number of jellybeans in the jar, each in their own different and biased way. When you take the median of their estimates, it tends to approximate the actual number of jellybeans well. -- Schapel (talk) 00:08, 13 December 2011 (UTC)

No, that's wrong. They measure different populations and have different definitions of "browser usage". Look at the dramatic difference between Net Applications, Statcounter and Wikimedia stats. The differences are not reflections of margins of error in measuring. (You're confusing an imaginary situation constructed by some of the editors in this discussion here with the real one.)VsevolodKrolikov (talk) 01:41, 13 December 2011 (UTC)

I don't see any dramatic differences between browser stats. I see numbers that agree fairly closely. I understand that the differences are not related to errors in measuring. The differences are caused by different biases in the samples that they take. Similarly, if you ask many different people how many jelly beans are in a jar, they will report wildly different numbers, each biased by the method they used to estimate the number. If you take the median of those guesses, you get an accurate number. -- Schapel (talk) 02:02, 13 December 2011 (UTC)

So 35% and 50% (IE) is close agreement? 6% and 11% (Safari) is close agreement? 15% and 22% (Chrome) is close agreement? You're kidding, right? These are large differences in market share. If a browser managed to increase its share by 50-90% it would be big news. Please don't pretend to engage in discussion if you've no intention of considering the evidence, as that would be acting in bad faith.VsevolodKrolikov (talk) 02:11, 13 December 2011 (UTC)

Which one of them is correct? I feel that picking one is a bigger WP:OR than simply calculating a median for illiustrative purposes. Can we just consider the median an illustration in a text form (and use it as such)? No original research involved. 1exec1 (talk) 12:30, 13 December 2011 (UTC)

Yes, those are pretty close. When you ask people the number of jelly beans in a jar, they give results that vary by an order or two of magnitude. So, yes, the numbers you cite are in close agreement relative to other situations where the wisdom of the crowds works well. So it seems to me that we can use it here. I am considering evidence, but I don't see you providing any. Do you have any evidence that the median of the measured usage share of browsers is not close to the actual value? All I see is people claiming that we can't do so, without evidence. -- Schapel (talk) 13:16, 13 December 2011 (UTC)

@1exec1. Choosing one of them is not original research at all, as we would not be creating new information. The policy we have to be careful of is WP:NPOV - that we are biasing the article. NPOV requires us to present viewpoints in proportion to their preponderance in reliable sources. If we are going to choose one measure for the lede, we choose the most popularly used measure (or possibly - in this case - the most popular two measures). As it stands, we're putting StatCounter and Net Applications - which are cited rather a lot judging by a rough google news archive count - on the same footing as Clicky and W3counter, which appear to be really rather more obscure, and Wikimedia foundation stats which I gravely doubt are cited by any independent source as a measure of general browser usage. In other words, we're contravening NPOV by giving equal credence to these measures in our calculation of this "median", particularly with the Wikimedia one, which no RS employs. (A graph of the Wikimedia results does no harm in the article, but it should absolutely not play any role at all in our headline descriptions of the market shares; that would break both NPOV and OR.) Look at it like this - if an editor were interested in using Wikipedia to undermine one browser or promote another, s/he could easily just keep adding or removing various measures to push the result towards the one s/he wanted. NPOV policy is a core policy for a reason, and it applies just as much in computer articles as in political ones.

The issue is thus not "which one is correct", but "which one is most commonly used". This is how we keep our neutrality in relation to the real world. It's clearly either Net Applications or Statcounter. I wouldn't object to either; it has struck me that having both side by side in the lede might be educational as to the different results one gets from different ways of counting, sampling and adjusting.VsevolodKrolikov (talk) 14:20, 13 December 2011 (UTC)

The problem is that deciding which source is most commonly used is itself OR. There are so many RS talking about browser share, that we need to use (statistical ?) methods such as GoogleNews, Google search, etc. Which one of them represents reliable sources? They give vastly different results (see my comment in the poll), how we can address that, WP:DUE and WP:NPOV then? Use median to combine the results of different methods that represent RS in some way? As you can see, we quickly return to the same problem we are solving, just with much more complexity and OR. Thus I do not think that picking some browser statistics engine is going to solve anything with less OR involved. 1exec1 (talk) 14:45, 13 December 2011 (UTC)

I've answered above - use better search terms. I get Net Applications and StatCounter coming out consistently way ahead of others. For the life of me I do not understand how any of you can think Wikimedia stats have any place in a general calculation. You've all been on here a while, you should all be familiar with NPOV. VsevolodKrolikov (talk) 15:29, 13 December 2011 (UTC)

@Schapel. I don't know how to put this delicately, but you don't seem to have grasped at all what the differences between these measures are. You keep talking as if there is only one population to count and one aspect of that population these measures are counting. This is simply not the case. The reason for the differences is not statistical or human error. It's because in practice they're not all trying to count the same thing. You may have what you think is a straightforward idea of "usage share" in your head, but establishing what that means in practice has resulted in varied approaches and definitions. Real world statistics are not like statistics one does in the maths classroom. (As for considering differences of 50% and more in results acceptable in statistics - that really strains credulity).VsevolodKrolikov (talk) 14:20, 13 December 2011 (UTC)

How are they counting not the same thing? They all measure global browser statistics, just with different biased samples of the global population. You suggest that picking one biased sample (one browser stat engine) and presenting that in the article is better than presenting several biased samples. Can you provide any statistically sound method how we can select the biased sample to present? 1exec1 (talk) 14:56, 13 December 2011 (UTC)

Woah - you haven't understood that either? OK: First of all, the populations are different. Wikimedia foundation stats are reflective only of people who use wikimedia websites. This is going to be biased towards people who edit the sites (disproportionately white, male, college/technically educated, anglophone) as well as those looking up information (likely to have a higher level of education than average). Net Applications works with 40,000 partners, but at least they have a range of commercial, corporate, content, public and other sites. StatCounter gets information from over three million sites, but only those that volunteer to carry its tracking code. Its website says "StatCounter Global Stats are based on over 15 billion hits per month, by a random sample of people worldwide, to over 3 million global websites, covering multiple interest areas and geographic locations".

As for bias: While both statcounter and Net Applications have general bias problems in the sample (Judging by statcounter's figures, towards the US and away from places like China and Japan), only Net Applications deals with this by re-weighting their stats geographically in accordance with CIA stats on internet usage. That is, Statcounter does nothing to remove the bias. In other words, they look at different populations to begin with, and then Net Applications does something to their data which changes it a lot. So all three measures have very different samples, and only one of them does anything to get rid of the bias in their sampling. They're not all measuring "global" usage, or even representative usage in a single country.

There is also a different way of defining "usage". Is it a unique visit (NA, W3counter), or is it every page hit? This matters because it views different browsing habits differently.

I'll stress again - real world statistics is not like stats at school. When you ask a question like "which browser is used the most", there is not one clear way of defining that question quantitatively. Operationalising concepts for measurement is not a straighforward hand-wavy procedure.VsevolodKrolikov (talk) 15:29, 13 December 2011 (UTC)

Different stats companies estimate usage share using different methodologies in the same way that different people estimate the number of jelly beans in a jar using different methodologies. I fail to see how the two cases are different, such that taking the median of dumb guesses that vary by orders of magnitude is valid yet taking the median of measurements that vary by 50% is not valid, and the only argument that anyone can seem to produce is along the lines of "They just are different, can't you just see that?" No, I can't see that. Please cite some sort of source that explains the difference so I don't have to simply take someone's word that it's different. Before you do so, I suggest that you read Richard Gill's thoughts on the matter again. -- Schapel (talk) 15:56, 13 December 2011 (UTC)

Adding to Schapel. Our own methods to assess whether the stats engine is appropriate is itself OR.1exec1 (talk) 16:02, 13 December 2011 (UTC)

We don't assess. We choose the one or two most used in RS. It's really simple. It's how Wikipedia works. VsevolodKrolikov (talk) 16:08, 13 December 2011 (UTC)

But do we have an precise method to evaluate which is most used in RS? For browser stats, there are a lot of RS that cite the stat engines. We need to use some method to assess them all at once. I argue that selection of the method (be it Google Scholar/News/Search, Bing, etc.) is OR, especially since the search query is supplied by us, and the results for different queried differ dramatically. This means that our selection of sources is also OR. Now, WP:OR says that we should not include OR, that means we must not select the sources. The more sources we provide, the less impact the OR of selection has. With which part of this answer do you disagree? 1exec1 (talk) 16:17, 13 December 2011 (UTC)

Strawman argument; not what VsevolodKrolikov said. We can pick any stat counter for use as lede, as long as it is being reported by a reputable source with meaningful editorial oversight and as long as we disclose that that is what we did. Selecting any such source is well within WP policies as long as the criteria and validation is disclosed (or obvious). We are under no obligation to pick "the best" or "most representative" source. Just a verifiable source. The attempt at calculating global usage share or some other variant of a single number signifying the usage share of a browser assumes that such a number exists and is meaningful. Only a single source attempts to do that in the first place (by adjusting for its own selection bias through CIA population estimates by country). Dumping different numbers (usage shares of different populations) into a calculation is WP:OR because it advances a viewpoint not supported by any source. The viewpoint is the resulting number, produced by WP editors, and claimed in here as well as by context in the article to represent the usage share of each browser.--Useerup (talk) 06:28, 14 December 2011 (UTC)

I'll try to address each point you raise separately.

1. Where do you see a strawman argument? VsevolodKrolikov says: "We choose the one or two most used in RS"; I say we can't do that without inducing OR. Can you elaborate where I'm missing something?
2. "We are under no obligation to pick "the best" or "most representative" source." WP:NPOV (specifically WP:DUE) begs to differ. "[NPOV] means representing fairly, proportionately (#1), and as far as possible without bias (#2), all (#3) significant views that have been published by reliable sources." Emphasis mine of course. All current stat engines are cited widely. Thus excluding most of them would violate #1 and #3. Using an arbitrary method to remove all but one or two sources would violate #2, specifically as far as possible part, since excluding less sources would mean less bias.
3. "<...> is WP:OR advances a viewpoint not supported by any source." Can you elaborate what viewpoint is advanced by dumping lots of numbers into the table? I can only to repeat again: median is for illustrative purposes and will not be referred to anywhere in the main text.
4. claimed in here as well as by context in the article to represent the usage share of each browser. Sorry, that's a strawman argument. Can you specify where in the main text median is cited as the usage share? Various notes already warn that the data of the sources probably does not represent the real usage share. If the implication that the median also doesn't represent the usage share is not straightforward for you, that is not a problem of the median. We can add another warning to solve this issue if you feel it is so important.

1exec1 (talk) 13:24, 14 December 2011 (UTC)

We are under no obligation to pick "the best" or "most representative" source for the lede. The article itself already presents fairly, proportionately, and as far as possible without bias, all significant views. That's the table. In fact, because the table presents those sources, the median is uncalled for. VsevolodKrolikov was talking about the lede. The sources all have bias - some significantly so - and that bias is not erased by any median calculation. Medians may help erase sampling errors but you cannot compute the median of fruits by counting bananas and slivers of orange peel. I can only repeat again: Median is misrepresenting (mistaken for global usage share) and it is not comparable between browsers. It has no value and makes no sense. The median is claimed as "the" usage share by condensing multiple numbers into a single number, thus implying that the number is somehow a valid summary. It is not as the sources are sampled from different populations, have been corrected by the sources in different ways and have been collected in different ways. --Useerup (talk) 15:24, 14 December 2011 (UTC)

1. "We are under no obligation to pick the best or most representative source for the lede": It seems we are discussing different things. I strongly disagree with all sources or the median being in the lede. Though I disagree with picking one for the lede too. I think the best option would be to be very vague in the lede to limit OR (e.g. different sources report IE usage share between about one third and a half).
2. "Median is <...> mistaken for global usage share", "The median is claimed as "the" usage share". No it's not. Repeating that it is doesn't make it true. You can add a warning saying that median can't represent the usage share correctly, if you feel it's needed.
3. "It has no value and makes no sense". That's your opinion. Do you have any real evidence to back it up? An expert in statistics, Richard Gill, thinks otherwise, as do about ten editors here.

So again, what's your argument against median in the table? 1exec1 (talk) 14:09, 15 December 2011 (UTC)

I think the argument against medians is becoming more reasonable. I agree that finding the median is original research, because it is a position not supported by any single source. However, some original research is allowed in Wikipedia. Computations involving simple arithmetic are allowed, provided there is consensus about applying the arithmetic. However, finding a median of five sources involves absolutely no arithmetic, so it is not clear that a consensus is needed to add it to an article. It seems like most people who have expressed an opinion want it, and professional statistician Richard Gill agrees it does readers a service. So, I have to ask, why the crusade against medians? Why not live and let live, and leave the median be? Or, if some editors really are so dead set against them, do what you need to do to clarify the Routine Calculations portion of WP:NOR to indicate that even calculations that involve no arithmetic must have consensus, or to indicate that finding the median of data form different sources is disallowed. -- Schapel (talk) 13:43, 14 December 2011 (UTC)

Didn't you mean something else with the first sentence. I feel it's somehow out of sync with the rest of the comment. Or did I misunderstand something else? 1exec1 (talk) 14:22, 14 December 2011 (UTC)

If 1exec1 and Schapel are being genuine in discussing this issue , I'll try to contribute further.

@Schapel: The median calculation involves 5 sources. There are no selection criteria for these sources. There are lots more than five sources quoted in the media. Why have these five sources been chosen over and above all the rest? Why is Wikimedia - which no independent RS uses for a general usage share - being used at all? Why include "Clicky" when it's a rarely used measure? If we publish a median based on our own judgement what measures are relevant without any RS to suppport that judgement, we are providing original research to the reader.

@1exec1: "Original research" refers to what we publish in article space. If we only use one published measure, it's not original research because we didn't make the measure. The information the reader sees has not been created by us. We are not Statcounter or Net Applications. We didn't make the numbers. It's not from us. Someone else calculated it. It's someone else's research, not ours. We didn't do it.

The danger with selecting a measure is not that it's OR. It's that it may violate our policy on neutrality. But "neutrality" does not mean treating all measures equally (which is what the median calculation does). It means treating views of browser usage in proportion to their prevalence in reliable sources.

Determining which measure is used most is not OR, because it does not, in itself, create original content. 1exec1's idiosyncratic belief that searching google news, books and scholar is OR directly contradicts, for example, commmon practice in determining WP:article titles, as well as the practical enforcement of NPOV in general across the encyclopedia. Search term choice is not "OR" in the wikipedian sense - it's a constant part of the discussion of sources and NPOV terms. VsevolodKrolikov (talk) 16:27, 14 December 2011 (UTC)

I feel that we could've been arguing about different things. Did you mean that we should select one or two sources for the entire article or just for the lede?

If it's the latter, most of my current arguments do not apply, that's for sure, but I still think that picking anything for the lede might involve too much OR or fail NPOV. For the lede I'd prefer a a very conservative and vague approach (e.g. "different sources report IE usage share between about one third and half").

If you actually argue for removing most of the sources from the tables, etc., I've already said why I disagree with that: WP:NPOV (specifically WP:DUE) says "[NPOV] means representing fairly, proportionately (#1), and as far as possible without bias (#2), all (#3) significant views that have been published by reliable sources." Selecting only one or two sources would certainly violate #3, #2 and possibly #1. #3 because we wouldn't include all significant sources, only the few most significant ones. #2 because we would use an arbitrary method to select sources (I'll cover that below) and thus add some bias. #1 because the less significant sources would be not represented at all. I agree that we already violate #1, but I think it's impossible to represent the sources proportionally, since can't pick grey color, we have only black and white: we either include a source or not. I think that including more sources would violate #1 less than just picking one or two, but that's only my opinion.

As for OR, "Determining which measure is used most is not OR": I don't disagree with the principle, I just say that in this case we must treat the results of the searches very carefully. We can't even build a search query without OR. We can only vaguely know how popular is each of the stat engines. However, we can't give more weight weight for one source, less for another, the only option is include or exclude the source. Given that there's so much on the table, I think the apparent popularity of the stat engines should be considered very conservatively, thus we should be excluding less sources. :::Regarding the median I argue that it is added only for illustrative purposes without any implications whatsoever (i.e. it's just a number). Thus there's no issues with median implying equal weight of different sources or median being referred as the usage share, since it should not do that. 1exec1 (talk) 15:51, 15 December 2011 (UTC)

"Nigel, I know stats can be confusing, but you do not appear to understand the arguments", "an imaginary situation constructed by some of the editors in this discussion", "You've all been on here a while, you should all be familiar with NPOV", "I don't know how to put this delicately, but you don't seem to have grasped at all...", "Woah - you haven't understood that either?". You do not win an argument among intelligent and educated people by proclaiming that their baffling inability to agree with everything you say is directly due to their personal stupidity. I don't intend to discuss this any further in these terms. I assume you'll get bored here eventually and go off to 'argue' somewhere else. Good luck with that. --Nigelj (talk) 18:18, 14 December 2011 (UTC)

Nigelj, it's not about agreement, it's about engagement. For example, if you look at the professor's opinion, he acknowledges the problem of the table not adding up to 100%, but in a detailed way he explains why he feels it doesn't matter. That's engagement without having to agree. It's acknowledging the issues that other people raise. You may be right to say it's unhelpful to point out that not engaging with objections in the specific manner people have done here genuinely comes across as not understanding them. However, if editors are going to dismiss objections so high-handedly, something needed to be said. !Voting rarely produces a happy outcome when the sides are not listening to each other. I'm certainly not calling anyone "stupid", and I apologise if you think I meant that of you.VsevolodKrolikov (talk) 01:34, 15 December 2011 (UTC)

@Schapel: I hope we agree that the median involves some level of arithmetic (placing numbers in order, finding the middle, and possibly averaging depending are all skills taught under the heading of arithmetic). That being said my own personal objection was never that this should be ruled out on WP:CALC reasons. But because a calculation is simple doesn't mean it isn't OR, and all of the questions that VsevolodKrolikov asks are a good indication of why. We choose the sources, we choose how to measure the central tendency of these sources, we report the results. I personally am happy with leaving all the sources as they are with no median. If others feel the table is too complex then reducing the number of sources appearing in the table, but keeping the information on which sources we are aware of, seems like a reasonable compromise to me.Thenub314 (talk) 19:43, 14 December 2011 (UTC)

According to Richard Gill, finding the median is not arithmetic. I agree that it is original research, but without arithmetic we don't seem to need consensus to add a median to an article. Most people seem to want it, and Richard Gill thinks it serves a useful purpose. Again, if you really think medians should be forbidden or should require consensus before they are added, please do whatever needs to be done to clarify Wikipedia's policies to make that clear. -- Schapel (talk) 13:37, 15 December 2011 (UTC)

WP:CALC talks about calculations. Are you trying to re-label median as "not a calculation"?. I am sorry, but median most definitively is a calculation and requires consensus. You cannot sneak around that by trying to make WP:CALC out to be about narrow arithmetic. Either we reach consensus or the median is gone. That is why I don't get why we have to protract this. --Useerup (talk) 16:01, 15 December 2011 (UTC)

Yes, it talks about calculations. It says requires "consensus among editors that the arithmetic and its application correctly reflect the sources". But finding the median requires no arithmetic, as does finding the largest number or smallest number. Therefore, I don't think we require consensus for these statements. If we require consensus for stating a median value, would we not also then need consensus to say that one number is larger than another? After all, that's a calculation also. Any summarizing at all requires computation of some kind or another. Even stating what the range of values is requires computation. Do we need consensus to do that also? I don't think so. It seems simple and straightforward enough to list five values and state which is the largest, which is the smallest, what the range is, and what is the median, e.g. among the numbers 3, 5, 6, 8, and 9, three is the smallest, nine is the largest, all the numbers are in the range 3 to 9, and the median is 6. If you're so dead set against it, please have the WP:CALC updated so it's clear that a consensus is required for these statements. -- Schapel (talk) 19:32, 15 December 2011 (UTC)

You are cherry picking the part of WP:CALC you like, ignoring the part which does not serve your purpose. You are even ignoring that 1) median may very well include calculating the mean if you have an even number of observations and 2) calculating median requires following an algorithm. You are also ignoring that one of the core objections of the opponents of the median is precisely it's applicability. Sorry, but calculating the median is squarely covered under WP:CALC, and these kind of discussions is the very reason WP:CALC exists: To make sure that if there is the slightest doubt that a calculation introduces new facts or represents a new viewpoint, the calculation is removed and not allowed. Also, see WP:BURDEN. These are core content policies which means that Wikipedia would rather err on the side of caution than risk jeopardizing it's purpose. I state again: Unless consensus is reached here (seems unlikely), the median is gone. It really is that simple.--Useerup (talk) 01:58, 16 December 2011 (UTC)

I'm not cherry picking. All of the examples in WP:CALC show arithmetic. The sentence involving consensus specifically mentions arithmetic. If non-arithmetic computation is disallowed as original research, the I don't see how Wikipedia editors can do anything but simply regurgitate information shown in sources. Summary by definition involves some sort of computation. There has to be a line somewhere, and that line is the line at which arithmetic is performed. If you wish to include median, finding the greatest number, finding the lowest number, and other computations that do not involve arithmetic, that needs to be made explicit so we have a clear demarcation of where the line is where consensus is needed. -- Schapel (talk) 02:34, 16 December 2011 (UTC)

Median is a calculation. It is covered under WP:CALC. Take it to the OR noticeboard if you disagree. I'm pretty certain what the outcome will be. So I suggest you just deal with it. --Useerup (talk) 19:14, 16 December 2011 (UTC)

Yes, finding the median is a calculation. So is finding the maximum, minimum, ordering numbers, and so forth. If we start including those calculations in WP:CALC, it's a slippery slope that will lead to people removing information from Wikipedia they don't like using the excuse that it involves minimal calculation. If a calculation involves no arithmetic and should be covered by WP:CALC, it should be listed explicitly there. All I'm asking for is a clear demarcation between what does and does not need consensus according to WP:CALC, because if is going to be expanded to include calculations without arithmetic, there's no clear boundary. Any summary involving numbers could be considered a calculation. Right now, that demarcation involves arithmetic. -- Schapel (talk) 13:41, 17 December 2011 (UTC)

The demarcation is: If it is a calculation it requires consensus. No need to try to parse WP:CALC to mean anything else by cherry picking the 2nd part and interpreting it out of context of the 1st part. Calculations are covered by WP:CALC because any calculation has the potential to introduce original research, arithmetic or not. Even maximum and minimum are prone to that. You could take that table an claim that the maximum usage share of Firefox is XX. And that would be original research and would patently disregard that other statistics could show a higher share. Same deal with median. --Useerup (talk) 15:26, 17 December 2011 (UTC)

Regarding this statement: slippery slope that will lead to people removing information from Wikipedia they don't like using the excuse that it involves minimal calculation. If the calculation is straightforward and useful, surely you will either have consensus that it is applicable or you will be able to cite a reliable source who has performed the calculation. That will remove the problem of "people removing information from Wikipedia they don't like". The problem here is that no source whatsoever has calculated the number being presented - it is a number produced by Wikipedia editors based on a selection of sources by editors. No other published reliable source has been able to challenge or validate the applicability of calculation nor the selection process. That is what makes the median blatantly WP:OR.--Useerup (talk) 15:50, 17 December 2011 (UTC)

I don't think you're thinking this through. When people summarize information in Wikipedia, they apply non-arithmetic calculation. They find the largest number, smallest number, sort dates, count items (gasp!), and so on. If some small group of people does not like the result of this calculation, they could block consensus on whether it should be allowed. For example, we provide the number of known vulnerabilities for each browser and list the oldest vulnerability. If an Internet Explorer fanboi does not like these results, he could find likeminded people and claim these is no consensus that we can perform those calculations. What you call "blatently OR" is routinely done in Wikipedia. If we need consensus for all of it, we'll be dropping material right and left. It seems simple enough to just include calculation of a median in addition to arithmetic calculations to WP:CALC. Then the problem I mention wouldn't happen, because the median would be the only non-arithmetic calculation which would require consensus. -- Schapel (talk) 13:03, 18 December 2011 (UTC)

@Thenub314. I agree that reporting median as the results (conclusion/summary) is certainly OR and should not be used in Wikipedia. However, I think that adding median just for illustrative purposes does no harm. By saying for illustrative purposes I mean that there must be no implications in the text that the median serves (or could serve) as a valid proxy for market share. We could have an explicit warning that the median does not represent the global market share. As for the calculation of the central tendency, I think that we can't do that without too much OR whatever method we choose. Just reporting what the sources say in a very conservative and vague way (e.g. "different sources report IE usage share between about one third and half") could be a sensible solution for that issue. 1exec1 (talk) 14:30, 15 December 2011 (UTC)

1exec1, in principle I really see your point. The difficulty is there is no good way to make it stand out that this is for illustrative purposes. Foot notes/asterisks are fairly easy to miss. To be a bit anecdotal for a moment about a month before getting into this conversation I looked up the corresponding information about usage share of operating systems, and glanced at the article quickly, found the number in bold on the table and mentioned it to a colleague. Only later when I was looking back in did I notice the debate and have a chance to reflect on the question at hand.

Now I understand this is not a really good argument, it is really a situation of shame on me. But, I am a fairly analytical guy, I have a Ph.D in mathematics, and I was mislead. I think it is fair to assume that the average reader would be as well. My own personal experience just reminds my why NOR is important, and why we should be relatively strict about it. Also, I coould imagine my position might be different if the table had 100 rows instead of 5, but in all honesty I see no need for an central tendency in this particular setting, the list of numbers is short enough to glean what you care to know from it directly. Thenub314 (talk) 03:36, 16 December 2011 (UTC)

I think it's possible to solve this problem. We can add a warning at the end of each table. Since the median is also at the end, the notice wouldn't be left unnoticed. For example:

Usage share of browsers for September 2011

Source	Internet Explorer	Firefox	Chrome	Safari	Opera	Mobile
Net Applications	50.9%	21.1%	15.2%	8.0%	2.7%	6.0%
Statcounter	38.9%	25.0%	22.0%	6.6%	3.1%	6.7%
W3Counter	35.1%	26.1%	20.9%	6.0%	2.4%
Wikimedia	35.5%	23.8%	19.3%	11.0%	4.8%	9.2%
Clicky	41.3%	26.3%	21.8%	9.3%	1.3%
Median	38.9%	25.0%	20.9%	8.0%	2.7%	6.7%
Note: median is here only for illustrative purposes. It does not represent the global browser usage share.

What's your opinion about this? 1exec1 (talk) 13:22, 16 December 2011 (UTC)

My opinion is that this is a horrible idea. It is not at all clear what "illustrative purposes" is. The median is just flat out wrong and this illustrates it more than anything. Just leave it out. The table is small enough to make the median pointless anyway. --Useerup (talk) 19:18, 16 December 2011 (UTC)

"The median is just flat out wrong": I've already asked you to provide real evidence for that. You haven't done that yet. There are a lot of people who think median is useful (including a renowned professor of statistics Richard Gill) for whatever reasons, so your argument is moot. Also note, that illustrative purposes does not need to mean anything precisely defined, the important part is the second sentence. 1exec1 (talk) 11:19, 17 December 2011 (UTC)

I have provided ample evidence that median is WP:OR, violates WP:DUE and is being misquoted by other articles throughout this debate. Go read it; I am not going to rehash it here. You may think that the median is useful - and it may be to some readers. But the point is that it potentially useful original research, a synthesis of multiple sources and a number controlled by wikipedia editors through the selection of sources. Useful is not the criteria for including anything in Wikipedia. Verifiability is a core criteria and the median fails that because the applicability of the median is being challenged under these circumstances. On top of that, the median has great potential for being misunderstood by readers (and editors) who do not realize that they A) should NOT compare medians across browsers and B) the "usefulness" of the median is hampered by the fact that it is calculated from sources which sample different populations and measure different patterns and cannot be usefully thought of as a median of the population (according to Richard Gill).--Useerup (talk) 15:39, 17 December 2011 (UTC)

The environment has changed a lot since your arguments were laid out and lots of them are being challenged. You should know better which your argument still apply. Regarding the above:

1. "is being misquoted by other articles throughout this debate": this is not a problem of median, but of the articles. Also, the warning in the table specifically addresses the absence of the implications of the median..
2. "But the point is that it potentially useful original research"": it falls within WP:CALC as you've admitted numerous times. Per that policy, there's no OR involved in simple calculations.
3. "<...> a number controlled by wikipedia editors through the selection of sources": all content of the article is controlled by wikipedia editors, it's always a problem. Why do you think this concern applies to median, but not, e.g. the table it derives from? Can you back your opinion with evidence, especially since the significance of the median is downplayed by an explicit warning?
4. "and the median fails [verifiability] because the applicability of the median is being challenged". Can you provide evidence that verifiability is affected by applicability by quoting a Wikipedia guideline or other relevant source? Without it, the argument is moot, since the median can be verified mathematically and therefore is not OR per WP:CALC policy.
5. "the median has great potential for being misunderstood by readers": an explicit warning specifically addresses that. If an editor uses median wrongly even after being warned, it's his problem. Can you specify how exactly the warning might be misunderstood by a reader?
6. "<..>should NOT compare medians across browsers <...>": do you have a source for that? Why shouldn't one compare medians of usage share as reported by various sources (i.e. not referring to the global usage share, but to the usage share data reported by various sources)? Comparing medians for some datasets does provide information about the relation of the datasets and is an accepted practice in statistics.
7. "the "usefulness" of the median is hampered by the fact <...> and cannot be usefully thought of as a median of the population (according to Richard Gill)". Could you specify where exactly does Richard say that? He says exactly the opposite: median is useful even in spite of the fact that it does not correspond to the median of the population. Quote:

   The numbers in question can't be usefully thought of as a sample from some population, so their median can't be thought of as an estimate of the median of the population, but so what? The median is very simply calculated and one can imagine that many readers would like to see it, so adding it to the table does those readers a service.

   So unless you provide evidence with more reliability than that of Richard Gill, your argument is moot.

Could you reply to the above arguments point by point? If I haven't addressed some of your points, please say so. 1exec1 (talk) 00:03, 18 December 2011 (UTC)

No. --Useerup (talk) 00:32, 18 December 2011 (UTC)

I've provided valid criticism to your claims. I've asked evidence for your argument and provided my evidence why I think your argument is wrong. Note, that repeating an allegation does not make it true. If you don't respond, we may just consider that the consensus to leave median has been reached; see WP:CONSENSUS : "Ideally, it [consensus] arrives with an absence of objections". Also, "Here editors try to persuade others, using reasons based in policy, sources, and common sense". You provided either flawed reasoning or no sources. 1exec1 (talk) 00:42, 18 December 2011 (UTC)

I have clearly stated my opinion. You have not provided arguments which can persuade me to change my opinion. I have stated and repeated arguments many times over and I'm not going to repeat it here just because you demand so. Go read the discussion. All of your arguments have been presented and rebutted before. Do not threaten to assume consensus just because I refuse to meet your tiring demands. I am not going to repeat arguments ad nauseam. --Useerup (talk) 01:09, 18 December 2011 (UTC)

Repeating an opinion does not make it true. " All of your arguments have been presented and rebutted before." No they weren't. If you are not eager to build a consensus, so be it. 1exec1 (talk) 01:46, 18 December 2011 (UTC)

To answer 1exec1, I am not that crazy about the proposed table, it still bolds the median and the warning message is small and not very clear. To return to a subject someone brought up earlier, why are we quoting wikimedia data? Is it quoted by any third parties? Thenub314 (talk) 04:24, 18 December 2011 (UTC)

It has some advantages over the other stat collecting sites, like you don't have to pay to get counted. Daniel.Cardenas (talk) 06:25, 18 December 2011 (UTC)

Yes, Wikimedia data is quoted by RS, e.g. [1], [2]. As for the table, it's not the final variant, the message could still be improved. 1exec1 (talk) 12:36, 18 December 2011 (UTC)

Thanks for the replies and links. I am content about including wikimedia in that case. Thenub314 (talk) 03:23, 19 December 2011 (UTC)

I don't think it says much about general browser usage though, that should be left to professionals who actually earn a living producing such stuff. How do we know whether there are special features about the people browsing Wikipedia? Besides of course them having good taste and using fave browser/ having bad taste and of course that explains them using fave hate browser. More to the point I think sticking in Wikipedia figures is undue weight when matched with those others. Dmcq (talk) 16:25, 20 December 2011 (UTC)

And BTW when is that median figure going to be removed? It just makes me annoyed seeing it in there making Wikipedia look like some fanboy blog making up figures. Dmcq (talk) 16:30, 20 December 2011 (UTC)

The RS (linked by 1exec1)([http://techrights.org/2010/09/10/us-oriented-operating-system-surveys/ techrights.org) says "Wikipedia numbers are not accurate as a global indicator and there are other shortcomings to these numbers. Christian75 (talk) 21:57, 21 December 2011 (UTC)

Why on earth would you include the median when these sites are measuring different things? Even if they were measuring the same thing, the median would depend arbitrarily on how many sources the authors decide to include. It’s a nonsensical idea.

In terms of what’s being measured, Net Applications measures users and Statcounter measures hits. These are absolutely not the same thing, any more than the number of cars is the same as the number of kilometres driven. The normal practice when discussing market share of countable goods is to discuss units (e.g. unique users or cars), and not intensity of use (hits or kilometres driven). Would you measure Volkswagen’s market share using the number of kilometres driven by VW drivers? Of course not. The correct measure is the number of cars, e.g. the number sold during a given interval or the number registered at a given point.

For ‘usage share’, the focus should be unique users, and hence the Net Applications data. Intensity of use may be interesting as a supplementary point, but it is simply not what normal usage of ‘market share’ refers to. Moreover, Statcounter’s sample sizes vary enormously and non-randomly across countries, which means its aggregate figures (but not country level figures) are simply meaningless. This is not a matter of opinion, it’s basic statistics. Faagel (talk) 12:44, 19 January 2012 (UTC)

Medians in Usage share of web browsers

Since there is really no consensus above and everyone involved can agree on nothing, I ask for outside comment on whether the medians should be included.Jasper Deng (talk) 00:12, 5 January 2012 (UTC)

• Suggestion: I believe that this is more about original research (synthesis of multiple sources) than it is about math/arithmetic. So perhaps Wikipedia:No original research/Noticeboard would be more appropriate? Certainly, allowing statistical analysis on sources selected by WP editors would set far-reaching precedent for WP:CALC. --Useerup (talk) 01:21, 5 January 2012 (UTC)
  • Note:: I have created an entry on the OR noticeboard since I believe this is firmly about WP:OR. --Useerup (talk) 17:04, 6 January 2012 (UTC)
    • It's more than that. The amount of math used to get the data is what we're disputing over regarding original research.Jasper Deng (talk) 01:25, 5 January 2012 (UTC)
      • The actual amount of maths is not a major matter provided it is obvious. The real problem is the obviousness and meaningfulness of it all given what it is operating on which I regard as sticking a finger in the air and then trying to dress it in mathematical clothes to gain some sort of credibility. That according to WP:OR is something we shouldn't do. Never mind it is just all unnecessary and could be done better without all this trouble. Dmcq (talk) 08:51, 5 January 2012 (UTC)
      • The amount of math for calculating of median is much less then needed to satisfy WP:CALC. The disputable points are the media itself and the calculations in the table before application of median. — Dmitrij D. Czarkoff (talk) 22:31, 6 January 2012 (UTC)
  • May I ask that we remove the median (in accordance with WP:CALC) until a consensus is reached. When there is a dispute the general rule is to exclude disputed content until resolved. This is all the more important when it comes to WP:OR which is a core content policy.--Useerup (talk) 01:04, 5 January 2012 (UTC)
    • There has been consensus in the past. I remember reading some policy that says not to delete content because of lack of consensus. Does someone know which page says that? Daniel.Cardenas (talk) 21:48, 6 January 2012 (UTC)
      • The poll above shows that there is consensus to keep median, so until this RFC is closed, nobody is entitled to remove it. — Dmitrij D. Czarkoff (talk) 22:20, 6 January 2012 (UTC)
        • Wrong. The poll above showed lack of consensus. Consensus is not a majority vote. Get over it. --Useerup (talk) 02:17, 7 January 2012 (UTC)
          • Useerup, neither it is the unanimous decision. The keep position received support by voters, while the remove position didn't. That's it. — Dmitrij D. Czarkoff (talk) 07:09, 7 January 2012 (UTC)
            • Dmitrij D. Czarkoff you really don't get this, do you? When there is still a significant number of editors with objections which have not been addressed, then there is no consensus. And under WP:CALC - which is a core content policy - consensus is required for any calculation no matter how "routine" you consider it. I will let this debate run it's course. By then, if there is still no clear consensus the median will be removed. --Useerup (talk) 20:50, 7 January 2012 (UTC)

• Comment - See WP:PRESERVE. If the information is removed from the article, yet is still being considered, then at the very least is should be included on this talk page. Otherwise, there's no information for people to view and discuss. Northamerica1000^(talk) 16:14, 10 January 2012 (UTC)

• This is an oppose or support situation. I oppose the median.Jasper Deng (talk) 01:13, 5 January 2012 (UTC)
• Oppose using a median value in the table. This is my first time viewing this particular article and commenting on the talk page, so forgive me if I missing something that was already gone over above. I'm not seeing why a median figure is necessary when there are only (currently) five figures for each browser in that table. The problems with using a median value seem to outweigh any benefits, in my opinion. - SudoGhost 08:41, 5 January 2012 (UTC)

  What problem do you see with using a median value? Daniel.Cardenas (talk) 21:41, 6 January 2012 (UTC)

  I would rather ask what problems would be caused by not having a median value. There's no point in having it in the article, and the lack of the median value would not be a detriment to the article in the slightest. It is not a reliably sourced aspect, and while I have no opinion on whether or not it is WP:OR, I don't see any good arguments for inclusion of an unsourced median value that shows the median value of only five figures. - SudoGhost 00:07, 7 January 2012 (UTC)

• Oppose for the reasons I gave above - the figures being operated on are grossly incompatible. Also the results don't give a percentage of he result. Plus I think any graph done of the figures can be done better otherwise without the synthesis. If somebody outside wikipedia wants to do this we can report on their results no matter that they are silly, for us to do it ourselves is just wrong and why are we making up things that nobody outside of Wikipedia can be bothered making up and writing about? Dmcq (talk) 18:23, 6 January 2012 (UTC)
  • That said, there are basically two ways of making statistics accessible by readers:
    1. choose among sources of statistic (rather tricky, as involves evaluating biases, and is evident WP:OR) or
    2. collect whatever is available (and passes WP:V bar) and summarize (using median or any other tool agreed upon).

    So the question basically is: what is better? Regardless of this and other discussions the maintenance of this article will lead to one of these options. — Dmitrij D. Czarkoff (talk) 00:56, 7 January 2012 (UTC)

• Support for the reasons I gave anove and in all the preceding discussions on this topic. — Dmitrij D. Czarkoff (talk) 18:44, 6 January 2012 (UTC)
• Support, what? again a pool? Come'on close either the pool as non consent or use the small majority as yes or no. The word will also destroyed (did say the Mayas), so who cares? mabdul 18:50, 6 January 2012 (UTC)
• Support the medians. Are we just going to keep repeating the poll until people get bored and someone wins by default? I think this is getting disruptive, as the summary tables have not been updated since September last year. This, if I recall correctly, was when significant regular contributors were driven off the article by these interminable arguments. Months and months of a few people holding the summarisation of the article to ransom. Puh. --Nigelj (talk) 19:56, 6 January 2012 (UTC)
• Comment My initial thoughts are that not only is this original research, it's faulty original research. But I'll withold judgement for now. I've asked a few questions at WP:ORN and await responses.[3] A Quest For Knowledge (talk) 20:02, 6 January 2012 (UTC)
• Support Its helpful. Yes helpful is not an excuse for putting something in, but wikipedia exists because it is helpful. Correct it is not necessary, but is wikipedia necessary? Daniel.Cardenas (talk) 21:41, 6 January 2012 (UTC)
  • ...but misleading. The median is basically out-of-date, not current, data.Jasper Deng (talk) 23:20, 6 January 2012 (UTC)
    • ...as all usage shares only show data which is at least one month old - no current data! So where's the point in having this article at all? mabdul 23:38, 6 January 2012 (UTC)
      • The median would be even worse than that.Jasper Deng (talk) 23:39, 6 January 2012 (UTC)
        • I'd be quite happy with it if that was all that was wrong! Dmcq (talk) 23:41, 6 January 2012 (UTC)
          • No, because he market share won't change dramatically within a day, a week or even in a month. There are only a few exceptions like the first few days of Christmas - because they use more Firefox since they have to use IE at work. (common example in Europe) mabdul 20:54, 7 January 2012 (UTC)
            • OK, so my concerns about it being outdated are not valid, but, as below, it is not an accurate representation of the actual market share.Jasper Deng (talk) 21:02, 7 January 2012 (UTC)
• Oppose because
  1. The median may be a simple calculation but it's applicability (as required under WP:CALC) is anything but simple in this case. The sources use different observations, different methodology. The sources sample different demographics/populations. Some sources try to account for their recognized bias by "correcting" using CIA numbers about Internet use in each country. The end result is a mess of incomparable sources being treated with equal weight even though some of them sample only a small fraction compared to others.
  2. The median is calculated across multiple sources which are selected by WP editors. Thus, the median is controlled by WP editors and not supported by any one source. No source is cited which directly supports such a calculation or the chosen selection. This is improper synthesis
  3. The numbers in the table over which the median is calculated have been "corrected" by WP editors. Because not all sources break out the observations in the same way (some don't report "mobile"), editors have found it necessary to "correct" those sources using the total/mean of all the stat counters. Thus, those "corrected" numbers are not supported by any source! This alone is violation of WP:SYN, but is necessary because editors want to calculate the median.
  4. The median numbers are useless for comparisons. Because the median of each column is calculated in isolation, the medians do not come to 100%. Indeed, the current numbers add up to 102.2%. So the medians tell us that more browsers than 100% of are being used?

  --Useerup (talk) 00:29, 7 January 2012 (UTC)

• Oppose Medians should not be used, because the sources are not the same size and their usage is not comparable. For example, 80% of Canadians support the Queen as head of state, therefore the median level of support for constitutional monarchy in North America is 40% (assuming it is zero in the U.S.). It is unusual anyway to apply a median to percentages. The most appropriate comparison would be to provide a total for all the sources then provide an average of users for each browser. But then we might want to provide weightings for each of the sources. That however is something that we would want to find in a source, not conduct ourselves, per WP:OR. TFD (talk) 00:57, 7 January 2012 (UTC)
• Medians are a fine calculation but oppose WP:OR of the sources of data to apply the medians too; they should be removed from this article. Nobody Ent 04:09, 7 January 2012 (UTC)
• Note Since the local community cannot seem to come to a consensus, I have posted this RfC at Centralized discussion. This should attract many more editors to help determine a solution. Hasteur (talk) 05:37, 7 January 2012 (UTC)
• Oppose If you add up the median values in the Sept 11 table you get 102.2%. Its mathematical nonsense. --Salix (talk): 20:10, 7 January 2012 (UTC)
  • Sorry, Salix, but You just have shown why the RFC for this topic was a very bad and damaging idea. The median values are not supposed to add up to sums of the source values. This is not an issue here, and this is no way connected to concerns this RFC is associated with. If You still want to participate in this discussion, You might want to give a glance to the discussions above this section. — Dmitrij D. Czarkoff (talk) 20:45, 7 January 2012 (UTC)
    • He is entitled to his own opinion, let's respect that. The whole point here is to solicit outside comments, and it doesn't have to be about the OR of this.Jasper Deng (talk) 20:46, 7 January 2012 (UTC)
    • Sorry, Dmitrij D. Czarkoff but Salix has a perfectly valid point: Each browser median will be used to compare it against the other browsers' medians. And when reported as a percentage readers will expect the numbers reported to be "fractions of 100". The fact that the sum of the medians can exceed or fall well short of 100% (not due to rounding errors) illustrates how useless they are, apart from being WP:OR. The editors have even avoided illustrating the "median shares" in a pie chart because they ran into this very problem. So rather than realizing that the medians are wrong, they swept the problem under the rug by using a bar chart instead. --Useerup (talk) 21:01, 7 January 2012 (UTC)
      • @Jasper Deng: Sry, I know that we are all no experts and that might be good - non experts writing an encyclopedia for non experts - but this isn't even statistics - that is math of the 9th grad (or so) and having a !vote based on a wrong memmory (in the case that he/she has learned it somewhen)?
      • @Useerup: Please check our last (or that before) archive why we are using a bare chart. (the short answer is: because pie charts are evil) mabdul 21:07, 7 January 2012 (UTC)
        • That's besides the point here (@ your reply to me).Jasper Deng (talk) 21:11, 7 January 2012 (UTC)
      • That just doesn't make sense to me. If it was true, these talk pages would be flooded and the tables were in a middle of a constant were. Effectively, the fact that the situation is the opposite only shows that median does its job: most readers just seek for a summary, and most of the rest understand the use of median. The questions appear when RFCs or other discussions draw public attention to the line. — Dmitrij D. Czarkoff (talk) 21:57, 7 January 2012 (UTC)
        • This is assuming that people that do not understand medians will know how to use a discussion page. Lack of something is not proof of the opposite. - SudoGhost 22:12, 7 January 2012 (UTC)
          • Yes - and it also assumes that they are not all just "going away" happy with the "answer" (or laughing at us, or confused). If they are, and the "answer" is sub par, then we have failed. Not everyone complains or comments. Begoon ^talk 07:55, 8 January 2012 (UTC)
            • Which answer would not be "sub par"? Our problem is not with identification of issues, but with addressing them. So if You know the better way to help the readers understand the data, could You please share Your thoughts on it? — Dmitrij D. Czarkoff (talk) 09:28, 8 January 2012 (UTC)
              • No, I don't know a better way to present it, sorry, but more importantly I really don't think it's the sort of aggregation, interpretation and analysis of sources we should be doing. Sorry. Begoon ^talk 13:09, 8 January 2012 (UTC)
                • So what is your vision of the right way to cover the subject? The only goal of the whole discussion is to find a viable solution that can be accepted as a consensus. Eg. Useerup stated that a table of data needs no summary, VsevolodKrolikov suggested to represent summary with a chart of a most cited source. I say that only a numerical summary can help. Do you share any of these opinions? Any other idea? Or what is Your input to the consensus building? — Dmitrij D. Czarkoff (talk) 14:02, 8 January 2012 (UTC)
                  • Ok, framed thus as a question - no summary. I'm not comfortable with us deciding how to combine and summarise data from disparate, dissimilar sources and constructing any analysis of that, even an average, because that is research we should not be doing. That's the best I can do to frame my opinion in your terms. Begoon ^talk 14:49, 8 January 2012 (UTC)
                    • Thanks, that was what I wanted to see. — Dmitrij D. Czarkoff (talk) 14:54, 8 January 2012 (UTC)
    • The problems with the medians are a classic case of why we have WP:NOR. Statistics done right is the selection of meaningful data and a lot of work in stats departments goes into deducing what is meaningful. Here we are attempting to do a Meta-analysis of multiple studies but not using an established technique, if we are using a median of percentages, not something I've seen before. A weighted mean would have little more statistical pedigree. The technique clearly has flaws, not adding up to 100% is just one. What the data is really telling us is that sampling effects are strong when measuring web-browser usage, for example wikimedia is clearly not a representative sample of web users. As some of the data ranges are from 35.1% to 50.9% its questionable if we should be reporting that many decimal places, indicating false confidence in the data. If we want to report this data faithfully we should really show error bars letting the user know how much trust to put into the data.--Salix (talk): 08:55, 8 January 2012 (UTC)
      • You address it as if we were doing a statistical study. Our data is actually known to have not only different samples, but also a different population; our sources are known to have biases, but not disclose them. I think the weighted mean wouldn't be any more accurate. And for wikimedia specifically: why do You think other sources to be more credible? — Dmitrij D. Czarkoff (talk) 09:15, 8 January 2012 (UTC)
• Oppose So we self select 5 sources of data, in many ways different samples (Wikimedia sample leaps out in this sense as limited), then we present the median (3rd largest) as somehow significant or useful? That's how it appears from looking at the article, and I'm trying to do just that to give an opinion unbiased from the reams of discussion above. If that's correct about what we are doing here - I oppose. We shouldn't be doing this research. The median (3rd ranked) of 5 figures garnered in this fashion? I just can't see where or how that is useful. Couple that with the problem that they look like percentages to the casual reader, who would therefore expect them to sum to 100. If my take on this isn't correct, please say so and I'll reconsider. Begoon ^talk 07:50, 8 January 2012 (UTC)
  • AFAIK, the amount of sources isn't the matter of selection. Sources that are found to pass Wikipedia policy for sources are included. Eg., the sister article about OSs has twice as much sources. I strongly agree that we should not select sources, but we need a way to represent the stats to the reader. You oppose median; which form of summary do You propose to replace the median? — Dmitrij D. Czarkoff (talk) 09:15, 8 January 2012 (UTC)
    • I honestly don't know. I think the problem I really have is that it seems to me that once one decides to aggregate and interpret data like this, from disparate sources, what one has, in effect, done, is produce one's own "Survey of Surveys" or "Poll of Polls". If that is done somewhere else, with published methods and rationales as to choices of source, summary methods etc..., we might be able to use it as a source - but if it's actually our research creating this analysis, well, I guess you see where I'm going with that. Begoon ^talk 09:23, 8 January 2012 (UTC)
      • I see no conclusion. You think we should just report the data? Or to avoid data completely? — Dmitrij D. Czarkoff (talk) 14:02, 8 January 2012 (UTC)
        • Report the data by all means. Just don't provide a number calculated as an average of 5 dissimilar sources as though they were perfectly comparable. And in the event you still do, despite it being wrong to do so at all in my opinion, don't use a median. People understand means, and that's what they expect to see, generally - anything else is likely to confuse. If all that means you can't summarise at all, then don't summarise at all. That's really about as far as I can go to match my opinion to your question. Begoon ^talk 15:08, 8 January 2012 (UTC)
          • The problem with the mean is that it would make it even more obvious that there is a very serious problem with weight in this table. Should the sources have the same weight (obviously, no) or should we compensate/guess using some other source (more WP:OR)? The basic problem is that the sources - despite all reporting browser usage shares - are not compatible at all and we should not be doing any type of calculation which assumes that.--Useerup (talk) 15:57, 8 January 2012 (UTC)
• Comment. Each median value is a percentage, and it is comparable with 100 in the same way as any other percentage. It says, "Of the N most reliable figures Wikipedia can find for this browser for this month, the median usage figure for browser X is A%". And so on for browser Y, Z etc. If a browser's median usage figure creeps over or under 50% for example, that is significant, whatever the row of medians adds up to. That's the only thing that makes no sense - adding up the row of medians. Just don't do this, as it gets you nowhere. That's one of the reasons why we dropped the pie chart - in effect it adds up the row of medians, which is a mad thing to do. The fact that you can't add them up does not make each of them invalid in its own right. --Nigelj (talk) 12:19, 8 January 2012 (UTC)

Yes, but, notwithstanding my basic objection that this is analysis/research we shouldn't be doing, isn't this true?

• One of the main arguments for using a median is to reduce the influence of "big outliers" in a large sample.
• It is, here, being applied to a sample of 5.
• The sample data for the median is percentages.
• By definition, percentages are confined to a range of 1-100, somewhat reducing the likelihood of "big outliers".

And, if we are honest, isn't there, anyway, a tiny hint here that we are using median as something that might avoid WP:CALC, because we really, deep down know that we're crossing, or over, the line of doing our own research here? (yes, I read the rest of the page, now).

Apologies if my maths/statistics knowledge isn't fully up to speed, I'm largely basing my supposition on medians and their usefulness from a discussion I had with a real estate agent, explaining that it helped to exclude massively overpriced palaces from local property price averages. Begoon ^talk 12:43, 8 January 2012 (UTC)

@NigelJ: And yet the medians are plotted in a graph directly encouraging comparison of the medians; omitting the fact that readers should actually re-scale the medians if they want to compare them. Of course, comparing the medians would be wrong since they are created from sources which doesn't even claim to state the same kind of numbers. Some sources tries to extrapolate to global usage shares, other sources report their raw usage shares. Doing any type of summary on such numbers is just flat out wrong. It's apples compared to slivers of orange peel.--Useerup (talk) 15:40, 8 January 2012 (UTC)

"the fact that readers should actually re-scale the medians if they want to compare them" is actually wrong. Each median is a percentage and so is comparable with 100%, and therefore is comparable with other percentages, and medians of percentages. All you cannot do is add them up and expect to see 100%. It is perfectly valid to say, "Based on the most reliable figures Wikipedia has been able to identify, the median usage of A just went above 50%", "Based on the most reliable figures Wikipedia has been able to identify, the median usage of A is now two percentage points greater than the median usage of B", and "The usage shares reported by statistics provider P are usually within 5% of the medians based on all the most reliable figures Wikipedia has been able to identify". --Nigelj (talk) 16:12, 8 January 2012 (UTC)

• Oppose . I agree with already said arguments against median. In our graphs we can choose a single source, i propose StatCounter, already used in some. The only valid "pro" of median is the synthesis, but due to the few sources, in my opinion it is useless. Subver (talk) 13:54, 8 January 2012 (UTC)
  • As all sources have biases, using one of them as a source for a plot will constitute a plain violation of WP:WEIGHT with no benefits. Having a fueled debate to avoid something the minority of editors regard as violation of policy to replace it with something that is plain violation of policy is... strange (wording optimized per WP:NPA). — Dmitrij D. Czarkoff (talk) 17:53, 10 January 2012 (UTC)
• Oppose. A median is a meaningless number when the inputs are not comparable. kop (talk) 06:16, 12 January 2012 (UTC)
  • But the input is perfectly comparable. It only differs in biases — that exact thing median is supposed to fix. — Dmitrij D. Czarkoff (talk) 11:12, 12 January 2012 (UTC)
    • The sources sample different populations and they may very well sample different behavioral patterns (unique users versus page impressions). The populations they sample are of very different sizes. One of the sources tries to extrapolate to global usage shares; others don't. They are not comparable. Yet, in a median (or mean) calculation they are given equal weight, the result (global usage share???) is not clearly defined and if you compare percentage points you err because they are not scaled to 100%. if the sum of the medians hit 110 (which is possible although right now it "only sums up to 102%), comparing percentage points and concluding that browser A has 2 percentage points more usage than browser B you would err by about 10% --Useerup (talk) 14:53, 12 January 2012 (UTC)
    • How can you say they're comparable when they're not reproducible, not verifiable, and, pointedly, are computed based on populations that are not randomly selected and which therefore represent nothing but themselves? The meaning of each metric is therefore questionable; and entirely unknown with respect to global browser share, which is what the median is suppose to pertain to. Further, as you note, arguments which pertain to the median also pertain to the mean. Yet nobody is arguing that the mean is meaningful -- it's obvious that the mean is not meaningful because it can't be weighted when sample size is unknown. It should be equally clear that when you take a median you must know what you're taking the median of, and nobody knows how to compare the different survey's sample populations. kop (talk) 08:15, 15 January 2012 (UTC)

---

Note: this RFC was supposed to help building consensus. Therefor it's not enough to say whether you support or oppose the median. Please also state your view on how the user agent statistics should be presented. Eg., a table with raw data, a table and a plot (which plot?), a table with a weighted mean line, a table with a median line, just a text that such studies are performed, or any other way. Please, make sure you not only criticize, but also suggest something. Otherwise your effort will actually turn out to further fuel the dispute. — Dmitrij D. Czarkoff (talk) 14:09, 8 January 2012 (UTC)

---

Note: Unlike what Dmitrij D. Czarkoff claims above, you are not obliged to present any alternative way to present the data. However, if you support or oppose please state your main reasons for doing so. If the median and "correcting" calculations are found to be original research it is simply deleted. If there is no clear consensus either for or against, it is simply deleted (WP:CALC requires consensus for a calculation to add it or keep it in).--Useerup (talk) 15:04, 8 January 2012 (UTC)

How many new polls have there been on these medians, here, at Talk:Usage share of operating systems and elsewhere in recent months? --Nigelj (talk) 15:28, 8 January 2012 (UTC)

---

• RFC reply It seems to me to be a big non-issue; the median it returns is effectively the same as the statcounter result anyway. IRWolfie- (talk) 23:51, 9 January 2012 (UTC)

Ending the RFC

The conclusion is that there is no consensus on whether the median is an appropriate calculation. According to WP:CALC there must exist consensus for keeping the median; otherwise it must be removed. The median is already removed through other changes and there seems (absense of edits) to be consensus that the changes are appropriate (good work!). I have removed the RFC tag. --Useerup (talk) 19:36, 27 January 2012 (UTC)

Shouldn't the beginning of Usage share of operating systems' page updated to remove the "A discussion is being conducted..." text, then? 195.23.92.74 (talk) 19:34, 19 March 2012 (UTC)

Wikimedia percentages

In this edit I have made what I hope are some improvements to the new, simpler summary tables. It was good to see Wikimedia data back represented, but it appeared to be utterly at odds with the other figures, whereas in the past, Wikimedia usually provided the majority of the median figures - i.e. it was often right in the middle of the spread. I looked into it, and the reason was the separation of mobile and non-mobile data. When the Wikimedia stats page said 29.2% for MSIE, it meant 29.2% out of the total of 87.5% of non-mobile visits. No wonder it wasn't comparable! The simple arithmetic required is perfectly allowed by WP:CALC. I copy-and-pasted the Wikimedia table into a spreadsheet and added a column based on =B2/B$26*100 to produce true percentages of the non-mobile visitor figure (which happened to be in cell B26). This was so easy that I did the same for the mobile table below it, and added these figures too. I found 'Other' figures in both cases by adding up the figures used (after rounding to 1 D.P.) and subtracting the totals in each case from 100. This is all simple, accurate and useful, and hopefully will not present any problem to maintain. As for the Wikimedia section table in the main body of the article, I have already complained about the complexity of this here, and now do not really know what to do with it. --Nigelj (talk) 20:19, 17 February 2012 (UTC)

Psdie (talk) 15:52, 9 March 2012 (UTC): I think the whole decision to use the Wikimedia stats for the "headline" usage chart is suspect - they serve to heavily under-represent Internet Explorer usage. I smell an anti-IE agenda (popular amongst tech-savvy users, but does no favours when trying to objectively monitor real-world IE market share). Reasons for under-representation:

1. By counting based on page views instead of unique users, the Wikimedia stats over-represent page-refresh-intensive users of the Wikimedia sites, i.e., Wikipedia editors. Thus the browsers used by Wikipedia editors will be over-estimated in the Wikimedia stats. I suggest that editors are likely to be more technically savvy than "typical" visitors, so are more likely to have an alternative browser installed - i.e., non-IE (standard browser with the most popular desktop OS, MS Windows).
2. The Wikimedia stats combine desktop and mobile stats. IE has no mobile presence, so its share will be significantly diluted in stats that merge mobile usage (currently ~13%). It's not necessarily unreasonable to present combined mobile/desktop usage as the headline figure, particularly given the rising importance of mobile, but this should be made clearer in the labelling.

Personally I believe an aggregate stat (median wasn't too bad, traffic weighted mean would surely be better) as the headline chart would present a more realistic picture. If that's prevented by WP:SYN (and not exempted by WP:CALC) then perhaps omitting a headline figure altogether is the fairest approach - otherwise Wikimedia's stats are being presented as more authoritative and accurate than other sources, which I'd dispute based on #1 above.

--Psdie (talk) 15:52, 9 March 2012 (UTC)

Clickz

There are some stats from ClickZ at http://web.archive.org/web/20090711201800/http://www.clickz.com/stats/stats_toolbox .Smallman12q (talk) 00:28, 1 March 2012 (UTC)

Wikimedia server logs

Generally not accetable

I just want to remind everybody that graphics of the Wikimedia server logs, like the one here are not acceptable, for a variety of reasons:

• They are original research
• Syntehsis
• We are giving undue weight to a particular website, and something not discussed in reliable sources
• They are self-referential
• The limit the re-usability of the charts, (in wikipedia mirrors, for example) because they refer to the wikipedia servers, and therefore are not "universal" or appropriate when discussing general browser market share

If anywhere, they could be used in the Wikipedia or Wikimedia articles, they certainly would be somewhat relevant, but the issues of WP:OR and WP:SYNTH would still remain if the information is not discussed in reliable sources. --SF007 (talk) 23:02, 13 March 2012 (UTC)

The only concern that can be considered at least marginally valid is that of WP:UNDUE, though each of the stats providers have known biases. There is nothing even close to WP:SYNTH, WP:OR and sself-reference. — Dmitrij D. Czarkoff (talk) 23:16, 13 March 2012 (UTC)

I dare to say it is much more than "marginally", since this is not discussed in any reliable source whatsoever. And while this might technically not violate WP:SYNTH or WP:OR, from my own POV, it certainly violaties the "spirit" or "principle" of those policies. It is arguably a self reference, while it does not mentions "Wikipedia", it mentions the "parent", wikimedia. Why should we present the stats from wikimedia? Are they representative in any way of market share? Why not just choose the sats from any other random website? Simply because Wikimedia websites are popular? Because Wikimedia runs Wikipedia? The answer to those questions should have already came from reliable sources... sadly, it is hard to justify the inclusion of such information. --SF007 (talk) 00:08, 14 March 2012 (UTC)

Even if the stats were based on accessing this image it wouldn't be self-referencing for a pretty evident reason: it doesn't reference content at all. It is not WP:SYNTH and WP:OR at all neither in spirit nor in fact: the data is referenced. And we all probably are well aware that squid data is itself pretty reliable source. At least more reliable then known unreliable sources like all those you left intact in the article. That's it: Wikipedia is the 3rd most visited site itself, so Wikimedia projects altogether are at least that much used (not to mention the fact that Wikimedia Commons' content is used throughout the web. If we are talking about the spirit of core content policies, then Wikimedia stats were the only reliable data in the article, as Wikimedia projects are known to have widest possible auditory in contrast to the rest of the sources, and thus the trustworthiness of these stats is out of question. The data in question is collected in the most neutral way possible and is verified in the most objective way – automatically; its sources are easily traceable and can be re-examinated; the chance that this statistics gets purposely misinterpreted in favour of one's commercial interest is neglictable... It is the ideal source for the purpose of all the policies you name. — Dmitrij D. Czarkoff (talk) 00:33, 14 March 2012 (UTC)

I don't think you really address the issue raised by SF007 at all. The problem is not whether you or any other editors considers squid data reliable. When we use raw data to produce a graph we implicitly validates and assign credence to the data. The fundamental problem here is that no reliable source has discussed these numbers, and thus it *is* WP:OR. No reliable source has taken a critical view on the data and opened up for quoting. Thus this is in violation with the goal of WikiPedia. Put another way, if you consider these data reliable and relevant, what source can you quote that these are reliable numbers? What source can you quote that these are relevant? What source can you quote that these numbers are representative for some population? --Useerup (talk) 15:40, 14 March 2012 (UTC)

The reliable source that produces these numbers is the reference given. These are the stats for over 150 billion web requests in a single month, across over a dozen of the busiest websites on the internet. The figures are worldwide and have been made by web users with every conceivable interest. have you got any source that says this is not a reliable source? WP:OR - reproducing results published by a reliable source is not OR. WP:SYN - we do not combine these figures with any others, no sysnthesis of multiple sources takes place. WP:UNDUE - this is a very large sample, and so is significant. WP:SELF - we do not assume that the reader is reading Wikipedia and we don't refer to this or any article on Wikipedia in any special way. Therefore these figures and their refs make perfect sense on any mirror server. Wikimedia is an important part of the web. I see that SF007 (talk · contribs) has gone ahead and unilaterally deleted all that material from the article regardless of this discussion. I shall reinstate it per WP:BRD and it should now stay in the article until this discussion has reached a consensus. --Nigelj (talk) 00:21, 15 March 2012 (UTC)

The burden of evidence lies with the editor who adds or restores material. WP:BRD is not a policy and cannot be invoked as a reason for undoing an edit you disagree with. As for the points:

• The Wikimedia server logs are WP:PRIMARY. That does not rule out using them, but they should be used with care. They have not been used with care here.
• You state that "These are the stats for over 150 billion web requests in a single month". This number is meaningless unless put into context. You need a RS which say something about how representative or for which demographic this source is representative. You can have 150 trillion web requests, if they are all sampling the same demographic it is not more useful than this number. Sheer volume is meaningless unless put into perspective. By a reliable source, please.
• You state that "The figures are worldwide and have been made by web users with every conceivable interest.". Got any RS for that? If so then please put it in the article. If not, your point is moot. Editors don't get to make such assertions.
• You ask "have you got any source that says this is not a reliable source?". You are seriously misguided as to what Wikipedia is. I or anyone else do not need to provide any source for removing unsourced or improperly sourced material (this being a case of the latter). It is you who need to provide a WP:RS which verifies why this stat is significant, prominent and relevant. Read WP:BURDEN.
• Regarding WP:SYN, agree, there is not WP:SYN as far as I can see. That is not the main problem.
• You state that "WP:UNDUE - this is a very large sample, and so is significant.". No. It is WP:UNDUE because it is given a more prominent position in this article than what has been discussed by reliable sources. Keep in mind that, in determining proper weight, we consider a viewpoint's prevalence in reliable sources, not its prevalence among Wikipedia editors or the general public. Read WP:UNDUE again.

--Useerup (talk) 08:21, 15 March 2012 (UTC)

In case of each source the reliable source itself is the source of stats. Neither of figures are discussed, for none of them the population or relevance to any population is discussed and all of them are reliable sources on their own. WP:OR requires that we use reliable sources for content, not that we support reliable sources with other reliable sources. WP:RS and WP:V also don't request that the sources we use should be discussed in other sources. Please just don't start another lame war with no proper grounds – this article is already damaged severely enough. – Dmitrij D. Czarkoff (talk) 09:01, 15 March 2012 (UTC)

In support of the reservations about highlighting Wikimedia stats over others (given bias created by its counting by page views, which are skewed by high admin activity), see my comment under Wikimedia_percentages above. If Wikimedia stats were based on uniques, I'd be more open to highlighting them as typical (which they aren't at present). --Psdie (talk) 15:21, 15 March 2012 (UTC)

@Useerup, I am very familiar with WP:BURDEN, thankyou. It says, "You may remove any material lacking an inline citation", which does not apply here. I won't repeat what Czarkoff just said; it seems obvious to me. Perhaps you should look at WP:EDITWAR, which says, "A potentially controversial change may be made to find out whether it is opposed. Another editor may revert it. This is known as the bold, revert, discuss (BRD) cycle. An edit war only arises if the situation develops into a series of back-and-forth reverts", which is what you just did. That is from WP:V, which is core policy. --Nigelj (talk) 21:41, 15 March 2012 (UTC)

@Psdie, your original point was about the use of a piechart of Wikimedia stats for the "headline" usage chart, was it not? That is something I'd gladly throw into the negotiation pot if everyone was willing to discuss and negotiate rather than delete and edit war. It's interesting that you see these stats as part of a pro/anti Microsoft stance. Did you know that there have been allegations in the past of people being paid specifically by Microsoft to edit Wikipedia?[4] We never find out who may have been paid to come here and add/remove content, but it's always something to be mindful of, within the context of WP:AGF. --Nigelj (talk) 21:41, 15 March 2012 (UTC)

The thrust of this objection (please correct me if I'm wrong) is that the Wikimedia stats are not discussed in other references, and so we are only dependent on a primary source for all of them. Is that correct? In that case, we are also going to have to delete the Statcounter figures, as they are only referenced to statcounter.com, and we don't have any references to other WP:RSs discussing them, their sample size, their methodology, or their reliability. Oh, the same is true for Clicky - totally sourced to getclicky.com. Same for W3Counter. Net Applications seems to call itself Net Market Share these days, and the same is true there. StatOwl.com is the same. It looks like there won't be much left. Which one of you would like to do the deletions? There'll have to be a new explanation written to take their place, as there won't be much left of the article. If these deletions don't go ahead, I'll assume that there was a mistake somewhere in the logic and replace the long-standing Wikimedia stats for our readers' benefit soon. --Nigelj (talk) 23:09, 16 March 2012 (UTC)

Wikipedia probably is not representative of the population due to all us open-source fans. I would vote "no". — Preceding unsigned comment added by 2.80.217.197 (talk • contribs) 04:53, 17 March 2012 (UTC)

Guys, stop edit warring. I've requested that this page be protected for that.Jasper Deng (talk) 04:57, 17 March 2012 (UTC)

So you think that Wikimedia stats are less reliable due to the higher load by users of open source OSs/browsers? Why do you think it is the case at all? Why do you think that StatOwl counting visitors of several Windows-related forums doesn't suffer from the similar issues? Do you know what issues do other figures suffer from? — Dmitrij D. Czarkoff (talk) 07:33, 17 March 2012 (UTC)

Use a source which has been reported by reputable mainstream media then. That's a reliable source. What is your problem with that? The Wikimedia server logs may be accurate, but they are raw data and certainly a primary source. As a primary source it is unacceptable that it is given WP:UNDUE weight over proper secondary sources. As I also cannot find any mainstream or acceptable tech medias which report on statowl, that source should also not receive undue weight considering that we have netmarketshare which is widely reported on in the media. We have to observe WP:DUE and not give undue weight to certain sources because WP editors believe that they are accurate. I have no problem with including the table with proper disclaimer about demographics (other than a bit unease about WP:NOTSTATSBOOK), but giving it prominence in the form of lead graphics is seriously WP:UNDUE considering that it is a primary source. It means nothing what you or any other WP editor thinks or believes about the sources and possible "issues". What matters is what reliable sources thinks about the primary source. --Useerup (talk) 10:30, 17 March 2012 (UTC)

@Useerup, you seem to have missed my point above: we have nothing in the article about what any secondary sources think about any of the primary source statistics. They should all go, by your logic. --Nigelj (talk) 20:11, 17 March 2012 (UTC)

Don't try to put words in my mouth, please. Netmarketshare seems to be quoted a lot in the media. Just follow WP:DUE and use that as the lede. Do not give a primary source with multiple potential issues a more prominent position than the sources which are usually quoted by reputable secondary sources. Simple. --Useerup (talk) 21:43, 17 March 2012 (UTC)

Just to be clear, are you arguing against the appearance of a Wikimedia pie chart in the lede, or are you arguing in favour of deleting all Wikimedia tables and removing all Wikimedia statistics from the article? It's important to be clear. --Nigelj (talk) 22:35, 18 March 2012 (UTC)

I am against using Wikimedia as a representative graphics in the lede. I believe that with proper caution (based on raw data with possibly skewed demographics) the stats from Wikimedia does have a place. I just don't think they should be given more weight than, say, Netmarketshare. --Useerup (talk) 00:23, 19 March 2012 (UTC)

RFC

Do Wikimedia's server logs constitute original research? If yes, should it be kept? Is the current use of them due or undue weight?Jasper Deng (talk) 01:34, 18 March 2012 (UTC)

• Keep the material under debate until a clear, material, cogent criterion for its exclusion is established. The working definition for OR is by now so muddy that searching literature or news for a particular quote, or finding a new means of representing data a la Edward Tufte or combining points of fact from different publicly available sources, or paraphrasing or summarising publicly available views or data, get pilloried as OR whenever it suits partisan editors. OR is the most convenient mud to fling. Accordingly, like patriotism, it is the first resort of the scoundrel who finds truths and logic inconvenient. The fact that the OR-rules (and patriotism -- and morality and...) are rooted in good intentions does not detract from our responsibility to examine them with great care and due cynicism whenever they are presented as justification for prohibitions. The Book of Words is for good sense and guidance, not for pettifogging, not even Wikipettifogging, and we should be on our guard against such.

Consider for example the fact that a given graphic representation of particular data includes data concerning that very graphic representation; is that self-reference? Certainly. The fact that a given argument about argument in general by definition deals with itself and is self-referential, is beyond question; it has been a cliche for a long time. But that does not mean that either of these examples is in itself unacceptable or even undesirable. They may be in any given case, but it is necessary to consult good sense, good conscience, good consequences, and a lot of other goods before we invoke hysterical subjunctives and Cretan liars for every text we disapprove of or disagree with. An alarmingly large number of such arguments in WP are settled by exhaustion or appeal to authority. This is unhealthy. (Now, there is a bit of OR, and make the most of it!) Similar principles apply to all the other holy Wikipillars.

Now, then. Truth and reason above all. I hold no brief for either side in the article under discussion, but I vote for the fair, good-faith, good-sense and constructive use of any representation, even though I have some very snotty views on snappy pie charts. (Edward Tufte had some really good points!) If anyone has a better presentation, bless him and go for it, say I. But if the best he can come up with is lawyering about data that might refer to WP among other subjects, or that unearthing publicly available data or data that can be displayed publicly in an illustration, but does not already appear in other textbooks counts as OR, then go away and explain yourself elsewhere. I have seen nothing in the arguments so far that moves me to forbid the material. JonRichfield (talk) 07:17, 18 March 2012 (UTC)

• Keep: the original research is the the contribution, that is primarily based on contributors' own experience and/or knowledge. The rendered Wikimedia usage stats is published independently of all the Wikipedia contributors and constitute a valid secondary sources (with the primary source being Wikimedia's logs). For the purpose of WP:OR they fall under the same category as all the other sources of statistics, though they are less affected by known biases due to well-defined methodology and population. The removal rationale is specifically flawed, as it is based on the assumption that these stats as source should also be a subject of coverage in reliable sources; this in fact means that the reliability of a source is assumed to depend on publisher's notability, which is not the requirement on Wikipedia. — Dmitrij D. Czarkoff (talk) 20:16, 18 March 2012 (UTC)
  • I missed the "due/undue" thing. Each stats item on this page references exactly one source. That is: StatOwl references StatOwl, Wikimedia references Wikimedia, Net Applications references Net Applications, etc. Consequently all the stats have equal weight in sense of WP:DUE policy. — Dmitrij D. Czarkoff (talk) 22:30, 18 March 2012 (UTC)
  • Equal weight does not mean that all sources should be given the *same* weight; rather it means that viewpoints (stats) should reflect the weight given to them by secondary sources. I have not seen Wikimedia visitor stats used by any secondary RS. On the other hand I often see Netmarketshare used. This means that Netmarketshare should be given more weight than Wikimedia and certainly not the opposite. --Useerup (talk) 00:30, 19 March 2012 (UTC)
    • Could you please name the policy, guideline or at least essay that states that we can only use sources that were discussed in other sources? — Dmitrij D. Czarkoff (talk) 00:52, 19 March 2012 (UTC)
• Keep - The discussion seems silly. Did someone invent the numbers? No, they are cold hard facts. Daniel.Cardenas (talk) 22:06, 18 March 2012 (UTC)
  • That's always an old argument used to keep original research. Could you please elaborate on that?Jasper Deng (talk) 22:08, 18 March 2012 (UTC)
    • Can you elaborate on "that's always an old argument ...". What I've seen of original research are people inventing things and then trying to put them on wikipedia. Daniel.Cardenas (talk) 00:42, 19 March 2012 (UTC)
      • Saying something is true is often used as an argument by (new) users, in my experience, to justify original research.Jasper Deng (talk) 00:46, 19 March 2012 (UTC)
        • You have doubts that the provided references and the data they point to are accurate? Daniel.Cardenas (talk) 00:53, 19 March 2012 (UTC)
          • That's one of the questions here, that I will not give my personal opinion on. However, my question here is whether, regardless of actual accuracy, this data is in compliance with OR.Jasper Deng (talk) 00:56, 19 March 2012 (UTC)
• Keep - I think the debate above is mostly founded on a confusion over Wikipedia, Wikimedia, and individual Wikipedia editors. If an individual editor, or some group of them, set out to trawl through some Wikipedia pages and thereby produce some statistics about the web in general in order to add some point to an article, then that would fail WP:OR. If when they added the point they said in the article, "We found this out by searching other Wikipedia pages", then that would fail WP:SELF too. This case is quite different: the figures were being published by Wikimedia long before they were added to this article; Wikimedia is an established and very significant web publisher worldwide; and references to Wikimedia as one among many independent sources of significant web visitor statistics are nothing like a problematic self ref. There is no requirement imposed on any of the other sources of stats that they have been discussed or validated by any tertiary source, so the only reason such a requirement is being suggested for these seems to be due to a misunderstanding regarding these preceding points. The limitations of any individual set of web usage statistics are well discussed in the article. In the days when we used to add median figures to the summary, the Wikimedia figures often supplied a significant number of the median figures (or were part of the pair that did). This shows that they are not outlying or surprising in any way - they are another solid source of valid figures, close to the middle of the spread seen from the various other sources each month. --Nigelj (talk) 22:29, 18 March 2012 (UTC)
  • Wikipedia is about verifiability, not truth. It doesn't matter at all whether you believe the numbers are in line with other statistics. What matters is whether you can find a reliable source which has dealt with that issue and has a viewpoint on it. If you believe that Wikimedia statistics is widely held as representative for the web population in general, then you should have no problem finding a source which supports that assertion. Useerup (talk) 00:37, 19 March 2012 (UTC)
    • On the other hand, which reliable source said the numbers are flawed?Jasper Deng (talk) 00:40, 19 March 2012 (UTC)
      • None. But reliable sources routinely use netmarketshare. My problem is with the WP:UNDUE weight given to these numbers. Someone likes to play statistician and make nice graphs out of the numbers. Given that they are numbers from primary sources and there are legitimate concerns about how representative they are, Wikimedia server logs or any "illustrations" based on them should not be presented as more prominent than numbers for which there actually *are* sources which use them. Remember WP:BURDEN? Useerup (talk) 00:51, 19 March 2012 (UTC)
        • No, WMF server logs aren't primary sources because we didn't make any browsers. It's clearly verifiable because we aren't claiming that the logs are an absolute count, only that they are our count. BURDEN does not apply to DUE.Jasper Deng (talk) 00:53, 19 March 2012 (UTC)
        • I don't think WP:BURDEN applies here, it does appear to be reliably sourced. A source need not be independent to be reliable, and I believe WP:SELFSOURCE applies here as well, the numbers don't claim to be representative of the internet as a whole, but specifically of readers of Wikipedia. These numbers therefore seem pretty relevant to the interests of, well, readers of Wikipedia. - SudoGhost 00:58, 19 March 2012 (UTC)
• Comment: I need comments from uninvolved editors for this RFC to be useful.Jasper Deng (talk) 00:40, 19 March 2012 (UTC)
• Keep - Wikimedia statistics are no different to statcounter, netmarketshare and other statistics. Wikiolap (talk) 04:54, 20 March 2012 (UTC)

Protected

The article has been fully protected two weeks due to the edit war. A WP:Request for comment is one way to get consensus on what belongs in the article. Since this is now the third time the article has gone under full protection, it may be reasonable to use blocks to deal with any warring that continues after expiry. Protection can be lifted if consensus is reached on talk. EdJohnston (talk) 16:47, 17 March 2012 (UTC)