Talk:Usage share of operating systems/Archive 4

Archive 1

Archive 2

Archive 3

Archive 4

Archive 5

Archive 6

Archive 7

Summary of issues with median row of the usage share table

RFC: Median calculation across multiple sources

The article (and a related article Usage share of web browsers) routinely calculates the median over operating system usage statistics from multiple sources. Article also uses the median numbers as basis a usage share graph. An editor has raised concerns that

• median may not be a routine calculation (and thus original research)
• the specific application of a median creates a synthesis over multiple sources
• the specific application of a median is not statistically safe as the sources have known (and declared in article) geographical and demographical biases
• the sources over which the median is calculated are selected by Wikipedia editors, the selection not being supported by any source
• some of the sources have been adjusted by Wikipedia editors to allow for the cross-source calculation because the sources do not break out the numbers in the same way (mobile usage share).
• medians calculated for operating systems individually yields a set of usage share percentages the total of which may exceed 100%

Some of these concerns have been discussed before, and editors have previously through a vote decided that median is applicable and not WP:OR. --Useerup (talk) 09:30, 6 November 2011 (UTC)

Because the debate above quickly focused on single, specific sub-issues (like that the median doesn't represent a conclusion but is just a median), I felt it necessary to create this summary section to make sure that what I consider the main issues are being addressed through the debate. --Useerup (talk) 09:30, 6 November 2011 (UTC)

For those who came here through RfC: please also read the previous discussion (#Median constitutes improper synthesis and original research), since there are already many arguments raised, some of which might not be repeated here. 1exec1 (talk) 14:58, 6 November 2011 (UTC)

I proposed a change to WP:CALC here. I think there is an issue with the policy itself. Please discuss. Thanks! 1exec1 (talk) 14:31, 7 November 2011 (UTC)

Median is not a routine calculation

Unlike simple arithmetic functions and conversions, the median is a statistics function and as such assumes a number of properties about the set over which it is used. So while it is well-defined it is by no means simple to apply appropriately (routinely), and consequently is not covered by wp:Or#Routine_calculations. There is a very big difference between calculating the age of a person from his birthday to creating a web usage share by calculating a medians. The former is uncontroversial, the latter is not.Useerup (talk) 10:12, 31 October 2011 (UTC)

Is this opinion based on anything? Dmitrij D. Czarkoff (talk) 18:58, 7 November 2011 (UTC)

yes. Read above Useerup (talk) 20:37, 8 November 2011 (UTC)

I've read above that You find it difficult to calculate median. I wanted You to state, what in the process of calculation was so difficult for You? I asked it in more detail in another thread of this discussion, but You don't reply there, so I asked in this thread. — Dmitrij D. Czarkoff (talk) 20:51, 8 November 2011 (UTC)

I don't understand how computing the median of a few numbers could be considered anything but one of the most routine calculations. I've frequently seen it done quickly and correctly by people whose grasp of mathematics is like that of a 10-year-old. A suggestion it is anything but routine without giving some reason why the particular case is somehow less routine than other instances of computing medians should be dismissed as absurd. Michael Hardy (talk) 15:37, 10 November 2011 (UTC)

I don't think the policy means in the calculation itself is simple, more the reasoning why the calculation is done is simple and a person would straightforwardly think of doing it. When a person calculates miles per gallon that is in some ways a more complicated calculation as it involves division but it is still much simpler in the sense of the policy as it is an obvious thing both to calculate and to choose to do and is done routinelyin the appropriate circumstances. Dmcq (talk) 17:19, 10 November 2011 (UTC)

The numbers are not homogeneous

In this case the set consists of numbers expressing usage share of a selection German language sites, usage share of commercial sites in the U.S, usage share of a web-designer oriented sites etc. These are not homogeneous and treating them as such is inappropriate. The median of Linux usage share risks ending up being a mean between the usage share in Germany and usage share with web designers. Useerup (talk) 10:12, 31 October 2011 (UTC)

The more diverse the user groups are involved, the more accurate the information is. Is being more accurate violates any Wikipedia policy? Dmitrij D. Czarkoff (talk) 19:00, 7 November 2011 (UTC)

Accurate? accurate??? I could create a whole list of native European language oriented stat counters, and that would create a heavy bias towards usage share in Europe (many diverse languages compared to, say, North America or even South America). How does usage share in China factor in? Accurate towards what? World share? How are the stat counters weighted then, considering that some very populous regions may be left out? The idea om "summarizing" these numbers is as far fetched as apples and oranges. More observations only improve a sampling when the samples are selected from a homogeneous set. That is not the case here. These stat counters were selected from what was "available", but they express wildly different shares (geographically, demographically and sampling methodology: unique users or page impressions). Useerup (talk) 20:47, 8 November 2011 (UTC)

The more data You add the more accurate result is. Any statistical study is somehow biased, and that's why this article needs median, because this is a statistical instrument to decrease bias as much as possible with no WP:OR.

So You have a simple choice: either ask for AfD or stop complaining.

Dmitrij D. Czarkoff (talk) 20:56, 8 November 2011 (UTC)

The selection is opaque and controlled by Wikipedia editors

The median picks the middle value (or the mean of the two middle values). Adding or removing a statistics counter will shift the numbers. The statistics counters have been picked by editors and it is not exhaustive (where's China, for instance, or accounting-oriented sites?). The selection criteria is not clear and certainly not supported by any source. Even if a criteria existed, qualifying the sources would in itself constitute original research.Useerup (talk) 10:12, 31 October 2011 (UTC)

Doesn't this raise questions regarding the table itself, and not the median? 89.180.27.148 (talk) 02:47, 4 November 2011 (UTC)

No, not in itself. Each source has been named, referenced and the potential bias has been explained. The reader can thus make his own judgement. The article does not infer anything. But when calculating the median, the editors implicitly claim a number of things:

• That the sources are comparable in the first place (same demographics, cultures) since the median should only ever be used for comparable numbers. Think about weights, for instance. The table has a source which is heavily biased for German language sites. How many people frequent those sites, compared to commercial sites in the US? What then does the median calculated over those two express? Market share in the world? Market share in US? Market share in Antarctica? Now also factor in that some of the sources use unique visitors and other use page hits.
• That the list is representative under an objective criteria. This is important because any addition will shift the median numbers. This list very well may be exhaustive, but that is not verifiable. There is no source that we can refer to which states that the statistic counters of this list are representative.

Useerup (talk) 06:25, 4 November 2011 (UTC)

The proper resolution here would be to add the sources You consider important here. And remove the WP:OR notice, since there is nothing original in it. — Dmitrij D. Czarkoff (talk) 11:10, 7 November 2011 (UTC)

The issue is not which sources I consider important. This concern is that the number being presented as "the median usage share" (even used in a graph) of a given operating system is not directly supported by any of the sources. Rather, that number is a synthesis from a list which will at all times be composed by the editors. I could question why a German-language oriented stat counter is in there. Depending on the outcome of that discussion the stat counter will be included or not. That in itself is not WP:OR. What IMO is synthesis is when those editor choices are reflected in a number being presented in the graph as well as "the usage share" of an operating system. That claim is not - and cannot be - supported by any of the sources. Useerup (talk) 15:12, 7 November 2011 (UTC)

It seems You fail to see the difference between cited value and directly supported value. The median number is synthesis at the same degree as each and every word in Wikipedia — they all are not literally taken frrom somewhere; instead they are the result of synthesis of all the sources. Median values are not special in this regard. — Dmitrij D. Czarkoff (talk) 15:24, 7 November 2011 (UTC)

No, the median is synthesis at the same degree as combining multiple sources in a way not foreseen and not supported by those sources, which is expressly forbidden. When the observations over which the median is calculated is under the control of editors it is not supported by any source. The simple calculation may be trivially verifiable, but the selection (or the criteria) of the list over which the median or average is calculated is not verifiable. That alone makes it OR. Useerup (talk) 16:42, 7 November 2011 (UTC)

What are You talking about? The median is the most trivial statistical instrument with absolutely no discretion on the editor's side. It is trivially verifiable. The verification method:

1. exclude the maximum and minimum value pairs until no more then two remains;
2. if there are two remaining values, sum them up and divide by two;
3. the result should be checked against the value stated in the article.

Now, please tell me, which part of the instruction above You can't accomplish and why? Dmitrij D. Czarkoff (talk) 18:32, 7 November 2011 (UTC)

The problem is in the sampling. There is really no need to be derogatory and imply that I don't know how to compute a median. I can compute the median of total fruits observed on orange- and apple trees. I really can. But, does the median then say anything about fruits per tree, oranges per tree or apples per tree? These observations are heterogeneous. Page hits and unique users. German language sites and US commercial sites. Small oranges, big oranges, red apples and green apples. Useerup (talk) 21:30, 8 November 2011 (UTC)

It is the way the statistical values are calculated: You take as much results as You can regarding the base an count them. The median represents the best one can get from the referenced material.

In effect, the referenced material as such exists without Wikipedia and each of its member isn't notable on its own, so either we somehow sum it up here (calculate median) or we just AfD the page as per WP:N. Just that simple. As the topic itself is notable, the editors did their best to give the most valuable data it is possible to derive from sources without WP:OR. In form of example: the median of oranges per tree and apples per tree gives more information about the amount of fruits per tree then the raw data itself. — Dmitrij D. Czarkoff (talk) 21:43, 8 November 2011 (UTC)

The median calculation combines multiple sources

The median calculation combines numbers from multiple sources. This is in direct contradiction to WP:Or#Synthesis_of_published_material_that_advances_a_position. The position being advanced is the idea of quantifiable operating system market share. This position is not supported by any of the sources.Useerup (talk) 10:12, 31 October 2011 (UTC)

The market share numbers have been "corrected" by wikipedia editors

The numbers used in the calculation have themselves been "corrected" because they did not factor in the same way. This is improper WP:Synthesis in itself, but it makes the median even more original research/synthesis.Useerup (talk) 10:12, 31 October 2011 (UTC)

Doesn't this raise questions regarding the table itself, and not the median? 89.180.27.148 (talk) 02:46, 4 November 2011 (UTC)

Yes, the numbers in the table should not have been "corrected". That is clearly improper synthesis where an editor guesses at how to compensate for lack of statistics for mobile units. My suspicion is that this was considered necessary in order to be able compute the median. If you don't try to arrive at a single number, we could just make a note for the row in the table (and the mobile cells) that this is the case. Useerup (talk) 14:22, 4 November 2011 (UTC)

What "correction" exactly You believe to be improper synthesis? — Dmitrij D. Czarkoff (talk) 16:42, 8 November 2011 (UTC)

Relevant numbers are marked with improper synthesis?. Useerup (talk) 18:19, 8 November 2011 (UTC)

If You disagree with presumed results of Desktop/Mobile split, You have my full support here. But Your RfC was about Median; how do these issues relate? — Dmitrij D. Czarkoff (talk) 18:31, 8 November 2011 (UTC)

The median uses the "corrected" numbers as input. This is just one of the many issues with the median. Useerup (talk) 20:49, 8 November 2011 (UTC)

This is not an issue with median. This is an issue with table. Don't You notice the difference? — Dmitrij D. Czarkoff (talk) 21:07, 8 November 2011 (UTC)

Suppose that we now remove this correction (mobile split) and the usage share for the other (non-mobile) OSes are reset back to what the sources state (desktop shares increases). Now you have the median calculated across multiple sources where two sources have higher observed relative shares for desktop OSes compared to the others. I suppose the median magically erases that error source when it is calculated across the multiple sources, some with mobile split and some without? Useerup (talk) 21:38, 8 November 2011 (UTC)

Median is inappropriate even if the numbers were homogeneous

The median yields a line where the total of the usage share does not even come to 100%. Because the median of each column is calculated in isolation, the numbers of the median line end up not being comparable to each other. The problem is compounded (and further illustrated) by the fact that calculating the median on the "Median / Windows All versions" column will yield different results depending on whether you calculate median for each version and sum the medians or calculate the median of the "Windows all versions" in isolation. So not only are the numbers volatile with respect to the selection, they will also yield different results based on how editors factors versions, and the numbers still end up not being comparable to each other. Useerup (talk) 10:10, 31 October 2011 (UTC)

Is this the reason that a bar chart is used where a pie chart would be more appropriate? Because a pie chart is not possible without "correcting" over the others column, and because the sum can actually exceed 100%? Useerup (talk) 22:55, 1 November 2011 (UTC)

The pie chart can be used with any data. One can construct a pie chart on raw numbers. The most probable reason for a bar chart is that it is more illustrative. Dmitrij D. Czarkoff (talk) 19:04, 7 November 2011 (UTC)

Regarding the switch to bar chart from pie chart, this brief discussion may have been relevant. It was a discussion about the charts on another article, but there are enough editors who work on both that the logic may have been carried across. Certainly the way a set of medians work out was part of the thought process. --Nigelj (talk) 19:21, 7 November 2011 (UTC)

Thanks for that insight. I believe that the graph should display a stacked bar chart normalized to 100% instead of a simple bar chart. Each bar could illustrate each source. That way there wouldn't be a need to summarize the numbers, the sources can be clearly illustrated and the reader can seek out the explanation for deviating usage shares between sources in the text/table. The problem of breaking out Windows versions against OSes with no version break-out could then be easily solved by indicating the family relationship between Windows versions through coloring or patterns Useerup (talk) 20:08, 7 November 2011 (UTC)

You are wrong, pie chart is not appropriate for "any kind of data". Sure you can put anything into a pie chart, but pie charts inherently expresses shares of a whole. The pie chart would expose the problem with the median numbers: They are not comparable. The bar chart is inappropriate because it misrepresents the data as it breaks out the Windows versions in separate columns but keep the other OSes versions in summarized columns. Bars of a bar chart does not indicate share of anything like a pie chart does - the bars convey the idea of absolute numbers where one bar can increase without the others need to decrease. There are several graph types which are appropriate for shares: Pie charts, area charts or stacked bar charts; bar chart is not. However, creating a pie chart with usage shares where the others is labeled as 3% (the median of others, but takes up an 8% slice of the chart would be openly dishonest. It does go to show how the median is meaningless.Useerup (talk) 19:20, 7 November 2011 (UTC)

So what is this? ;-) — Dmitrij D. Czarkoff (talk) 20:44, 7 November 2011 (UTC)

I see a pie chart with no labels. Your point? Useerup (talk) 00:10, 8 November 2011 (UTC)

My point is that I've produced it from current median data in the article. Though data doesn't sum up to 100%, it does sum up to something, and the pie chart shows the shares in this total. — Dmitrij D. Czarkoff (talk) 00:50, 8 November 2011 (UTC)

A few questions:

1. What title would you use for the graph?
2. What does the entire disk represent?
3. How would you label each slice if you were to show both the observation and the slice percentage size of the disk (as is customary for pie chars)?

Useerup (talk) 16:04, 8 November 2011 (UTC)

Here we go:

1. "Operating system usage share (Median)".
2. see above.
3. Name and percentage as stated in table.

Dmitrij D. Czarkoff (talk) 16:40, 8 November 2011 (UTC)

The point I made ages ago in the discussion I linked above is that, if the data in a pie chart are percentages but they don't add up to 100%, then we can get anomalies like 48% looking like more than half, or 55% looking like less than half. No percentage is accurately represented by the angle used to represent it. This is misleading. One answer would be, if the total adds up to less than 100%, simply to have an unaccounted for gap in the pie (you cannot label it 'other', as there is likely already a median 'other' segment present). I am not sure what you might do if the medians add up to more than 100%. This is why we dropped pie charts of medians, as far as I remember. --Nigelj (talk) 18:41, 8 November 2011 (UTC)

Wrong. You are dodging the questions but are nevertheless proving my point. Calling the graph Operating system usage share would be dishonest and inaccurate:

1. The proper title would be Operating system medians shares* share of sum of medians for all operating systems. with the caveat that *) For windows the usage share medians for individual versions are used and that the sum of those medians does not add up to the median of Windows overall. Yes, it is that bad. The graph does not depict operating system usage share.
2. The entire disk represents the sum of all medians, not usage of operating systems. Same caveat as above.
3. These would be the proper labels:

Windows 7: 33.0% (34.9%)

Windows Vista: 10.8% (11.4%)

Windows XP: 34.4% (36.5%)

Mac OS X: 8.2% (8.6%)

iOS: 3.7% (3.9%)

GNU/Linux 1.2% (1.2%)

Android 1.4% (1.4%)

Symbian 0.2% (0.2%)

Blackberry 0.4% (0.4%)

Other 1.2% (1.3%)

Need to explain that first percentage numbers are medians while percentages in parenthesis are each median share of total medians. The total of the medians comes to 94.4%. That is why each median's share of the total is higher than the median itself. If you had chosen to show one slice for Windows (like for the other OSes) instead of one for each version the median of Windows would have been (according to table) 79.8%. Summing the medians for the Windows versions only yields 78.2%. That difference would mean that the slice sizes would change yet again. Get that? The total and the size of the other slices change depending on whether you break Windows up or not! But then again, we can also just pretend that there is no problem by not including the labels.

This is so amateurish that I cannot believe I have to explain these things. The usage of median as well as the chart is so utterly wrong. And I have to argue against someone who thinks that if one can create an SVG that is proof positive that median is appropriate. I am going to call in an expert in statistics. Useerup (talk) 18:51, 8 November 2011 (UTC)

The reason there is an argument about it is because it was set up by an editor on Wikipedia rather than being something 'out there'. It is amateurish and wrong but it is the sort of thing some people in reliable sources sometimes do and it gives a general feel. That does not mean we should do it. And we should not do it. It is what Wikipedia calls original research. Dmcq (talk) 19:13, 8 November 2011 (UTC)

(e/c) I think I finally see what you're trying to say (Useerup), and it is almost exactly the same as I was saying except I was referring the matter to the pie chart. I can't parse the title "Operating system medians shares* share of sum of medians for all operating systems" at all. Have you left out an apostrophe, or accidentally used the wrong word somewhere in it? The point that makes our medians valid is that it is true to say, "The median of our selected eight estimates of the usage share of Android is 1.39%". We clearly show the eight we chose, and give reasons why we chose them. The calculation is trivial per CALC. Where it becomes tricky is to take a row of medians and try to do something else with them as if they were another row of data. They are not. They are a set of discrete facts, not members of a row that adds up to 100%, or meets any other criteria as a set. That is why you cannot display them in a pie chart. You certainly can't start calculating the percentages in parentheses above, 'each median's share of total medians'. That is effectively what the pie chart does, and both are wrong. It makes little emotional difference with the figures we have for most OSs at the moment, but at the time when the usage share of the MSIE browser was dropping through 50% it might have seemed very important to some people to keep 48% looking greater than half... Your parenthesised figures might have allowed that, and that would have been wrong. There is nothing wrong with each median taken individually, though, and quite a lot that is useful and right. --Nigelj (talk) 19:25, 8 November 2011 (UTC)

So, if the sum of "median usage shares" comes to more than 100% (as it does at the browser share article and which could very well happen here) don't you think that there's something wrong with the share? Yet, the medians are included side-by-side, encouraging comparison of numbers which clearly are not comparable and even plotted in a graph, further encouraging comparison of incomparable numbers. I would say that the fact that you can arrive at a total share above 100% is a pretty damaging impeachment against any such calculation. This is what happens when statistics is being applied without proper knowledge of the field.Useerup (talk) 20:53, 9 November 2011 (UTC)

May be You would actually state, in which way is it amateurish? To date You just make statements. What about actually proving them? — Dmitrij D. Czarkoff (talk) 20:12, 8 November 2011 (UTC)

It seems You don't have even the most general knowladge about statistics. First, let's pass to questions:

1. To prove Your position You truncated my answer. Impolite and shows You can't find a proper grounding for Your position. The Median is a valid and widely used statistic method, so the heading "Operating system usage share (Median)" is absolutely OK, as wel as is OK the heading "Usage share of web client operating systems. (Source: Median values from Usage share of operating systems for August 2011.)". The heading "Operating system usage share" is not OK, and that is the reason You can't find it on the article page.
2. The entire disk represent the total of the median values, which is exactly what is supposed to represent the disc of "Operating system usage share (Median)" pie chart.
3. The labels You propose would not be the proper labels as the pie chart represents the median values, which are not supposed to sum up as 100% per se.

Do You really understand what is the difference between mean and median values? Do You understand, why median is prefered for representation the total result of similar statistical research on different user bases?

P.S.: Please, stop screwing indentation! This is not the first time You reply shifts right in the middle.

Dmitrij D. Czarkoff (talk) 20:12, 8 November 2011 (UTC)

Oh, now, after a week of rioting here You actually state the lack of expertise. May be just shut the whole thing up? You are clearly in the minority, and there is an evidence of the consensus on the question before. — Dmitrij D. Czarkoff (talk) 20:45, 8 November 2011 (UTC)

I added the tag because I hope that an expert can explain this blatantly obvious misuse of statistics better than I can. Useerup (talk) 20:53, 8 November 2011 (UTC)

Well one of my degrees is a masters in statistics but I do not believe the main problem here would be solved by correct use of statistics. It is a basic failure to follow Wikipedia policy. We should not be calculating figures like this, it is against WP:CALC. If it requires special expertise to know whats right and an expert to explain it s just not a routine calculation. You could easily put all the figures into one chart using different colour lines and that would require zero thought to understand and explain. Just get rid of the unnecessary calculation. Dmcq (talk) 00:21, 9 November 2011 (UTC)

Why do You think that median calculation requires any skills or any expert? What can You cite in favour of Your opinion? Please stop just complaining and pass to actually proving Your complains. The only argument so far was that values don't sum up as 100%. The next obviously needed step for this discussion to ever turn into real discussion is proving that there is at least something wrong with it. And to actually make it possible to somehow resolve the issue You have to provide a vision on how to summarise the referenced data. — Dmitrij D. Czarkoff (talk) 01:24, 9 November 2011 (UTC)

About expertise as the plumber said knowing where to hit the pipe is what you're paying for. Yet another of Wikipedia's policies besides WP:CALC is WP:BURDEN. We don't need sources to remove stuff. We need sources to justify keeping stuff in. We should not be proving anything. And we do not need to prove anything. Even if there was a burning need to make a simple graphic I've pointed out how you can do that without causing problems. Just get rid of the median line in the table and that form of the graphic. Dmcq (talk) 08:45, 9 November 2011 (UTC)

WP:BURDEN has no relation to this discussion: the median is referenced with its data sources, which are probably the best referenced material in Wikipedia. And these references are out of scope of this discussion. If You have problem with the referenes, please start a new thread.

Here we discuss the current representation of data, so please, actually be exact and answer the questions in my previous reply. Or just say You can't. — Dmitrij D. Czarkoff (talk) 11:17, 9 November 2011 (UTC)

Your stuff simply is not in line with policy. But okay I'll answer. There are multiple representations. For instance many people use geometric mean to stick a whole lot of disparate data of this type together. Then there's other ways which take more account of that they are all from slices of a whole. Then there's all the other problems like compatibility and weighting by reliability. The figure produced is in essence an judgement, in other words dependent on a point of view. What entitles anybody here to make such judgements? This is the sort of stuff the original research policy was designed to combat, people using their own judgements to stick their own ideas into articles. WP:CALC is deliberately minimal and this just drives a coach and horses through it. If a reliable source does this sort of thing we can report it an I would not normally feel too worried about the problems but we should not do it ourselves. I really do not need to explain this to you, it is straightforwardly against policy. Dmcq (talk) 12:11, 9 November 2011 (UTC)

So Your point is that if there are many methods out there, wikipedian can't use any of them. That effectively means that wikipedians can't add any data, as there are always alternative approaches. Your reading of the policy is too restrictive and is itself a mere POV. — Dmitrij D. Czarkoff (talk) 15:25, 9 November 2011 (UTC)

Even if there was only a single way it would still not be just a routine calculation like changing kilometres to miles. The only way I could see for justifying the graphic is under WP:OI as an illustration rather than an accurate sourced fact, but the median line would have to be removed from the table or put somewhere where it clearly was just for illustrative purposes and not something justified by the sources. I'd just remove the median line in the table as after all it is easily calculated. Put in the table it is an editors own analysis, and one's own analysis is original research on Wikipedia. Just because a number is easily calculated does not mean it is a routine calculation. Dmcq (talk) 16:04, 9 November 2011 (UTC)

Even without a degree in statistics it should be obvious that how to apply statistical analysis (such as a median calculation) requires deliberation, not only on whether the data is safe for such an analysis, but also on whether the result is at all meaningful. This is one of the situations where not only is the median meaningless, it is also original research. As Dmcq says: It is not a routine calculation. This very article demonstrates why. Sure, a median is calculated, but then it is immediately compared to other medians in the table, and plotted in a graph. Comparing medians is meaningless. Readers will believe the graph shows usage shares compared, while in reality it shows mangled data. Useerup (talk) 16:32, 9 November 2011 (UTC)

OK, as You don't say anything new again, I assume You have nothing more to say. So, I quit this discussion for now as I don't see neither valid points on your side nor significant amount of Your supporters to be able to call Your position WP:CONSENSUS. — Dmitrij D. Czarkoff (talk) 17:03, 9 November 2011 (UTC)

I see noi consensus here for keeping the current median line in the table. I think I can agree with keeping the graphic under WP:IMAGES#Pertinence and encyclopedic nature "Consequently, images should look like what they are meant to illustrate, even if they are not provably authentic images". I think it scapes in under WP:OI "Original images created by a Wikipedian are not considered original research, so long as they do not illustrate or introduce unpublished ideas or arguments". The median line in the table however does not satisfy WP:CALC "This policy allows routine mathematical calculations, such as adding numbers, converting units, or calculating a person's age, provided there is consensus among editors that the arithmetic and its application correctly reflect the sources". That is clearly for things like converting miles to kilometres. It wouldn't be routine even if a consensus here thought it was. Dmcq (talk) 18:10, 9 November 2011 (UTC)

Agree with Dmcq. Median is not supported by a source and can be removed on this basis alone. Other issues remain as well (improper synthesis, disparate sources, sources selected by WP editors). Useerup (talk) 18:27, 9 November 2011 (UTC)

Please refrain from wikilawyering: changes need WP:CONSENSUS, so it's Your task to convince everyone. BTW, someone here has mentioned the previous consensus which has resulted in the current median line. — Dmitrij D. Czarkoff (talk) 20:07, 9 November 2011 (UTC)

A change does not need consensus when the change is to remove original research. Don't game the system. WP:NOR is one of 3 core content policies and generally trumps guidelines. To outline this pretty clear policy:

• The term "original research" (OR) is used on Wikipedia to refer to material—such as facts, allegations, and ideas—for which no reliable, published source exists. This includes any analysis or synthesis of published material that serves to advance a position not advanced by the sources . (my emphasis of the parts relevant to this debate):

Median is synthesis of published material. You can try to call it "summarize" or use other words for it. But it remains synthesis: A new "fact" is derived from the sources but not attributable any of them.

• To demonstrate that you are not adding OR, you must be able to cite reliable, published sources that are directly related to the topic of the article, and directly support the material as presented . (my emphasis of the parts relevant to this debate):

None of the sources directly (not even indirectly) supports usage share as presented through a median. In fact, they couldn't, could they? It is a synthesis of multiple sources.

There may not be much point in debating this further. The median is WP:OR and is going to be deleted per the core WP policy. If you cannot accept this you can try to raise it on a noticeboard. Useerup (talk) 21:19, 9 November 2011 (UTC)

General responses

None of the eight sources is based on web devlopers' usage share. That source (w3schools) is not used in this table.

I think I've spent enough time on this dispute now. This table has been around for a long time and is the product of consensus here. The sudden arrival of one new editor is not sufficient to disrupt that, no matter how aggressive, disgruntled and sure of his opinions he is. So until Useerup's view gains more support, I assume he's on his own and will spend my time doing something more interesting than repeating arguments which have now had a good airing on both sides.--Harumphy (talk) 11:34, 31 October 2011 (UTC)

I actually support his view. The median is unimportant, as it often represents outdated data, you know what I'm saying here?Jasper Deng (talk) 21:48, 31 October 2011 (UTC)

What now?

Afer this summary... what are the next steps? It was proposed previously in the discussion that we should vote (again) about having or not the median line there. What are we waiting for, exactly? -- 89.180.146.171 (talk) 20:05, 1 November 2011 (UTC)

In my initial attempts at discussing these topics only a few of my points were addressed. But the discussion was quickly going in the direction of whether the median line constituted a "conclusion" or not. My concern was that we were splitting words while the core issues were being ignored. That is why I broke them out. I'm waiting for someone to discuss these issues. My concern is that the median line is wrong (median is the wrong function to apply here for multiple reasons) and it is contradiction to one of the WP pillars which says no original research. I am willing to have this discussion. For the record, I do not consider a vote appropriate on these issues, per WP:NOTDEMOCRACY. Useerup (talk) 22:45, 1 November 2011 (UTC)

Just open a request for comments, since there's no consensus over this issue. WP:NOTDEMOCRACY does not apply here since it concerns with using voting as primary means to reach consensus. Here we use arguments and that's accepted. 1exec1 (talk) 21:58, 5 November 2011 (UTC)

Thanks for the suggestion. I opened an RFC. Useerup (talk) 12:21, 6 November 2011 (UTC)

Rfc discussion

• I see no reasons I can agree with for changing WP:CALC in the discussion. If a source does not provide something like what is wanted we should not be doing so either. It definitely should not be provided standing on its own as something that satisfies WP:Verifiability or WP:Original research. Even as an illustration of data in the text which is something that can be set up by an editor and does not need verifiability I'm a bit chary of it, but that is the least I would expect to justify its inclusion. Dmcq (talk) 15:23, 7 November 2011 (UTC)
  • This RFC isn't about changing CALC. WhatamIdoing (talk) 15:42, 7 November 2011 (UTC)

I've had another look and I think it can be counted as an illustration rather than content which lessens the requirements. It should however be in or just beside the section with the source data that it is illustrating rather than at the top and it would be far better if it also illustrated the minimum and maximum lines for each. Dmcq (talk) 18:15, 7 November 2011 (UTC)

The illustration was actually a secondary issue. The primary concern was about the median line in the table, not the illustration. Hence such a debate. 1exec1 (talk) 18:51, 7 November 2011 (UTC)

I'd missed that. Sticking median into the main text without a special marker saying it is just something we thought of doing is wrong. We really do need to distinguish very carefully between illustrations and examples and the main text. I think the median line should be removed from the table. Dmcq (talk) 19:17, 7 November 2011 (UTC)

What kind of marker are You talking about? The row is specifically called Median, which is a bold marker. If You don't like medians in general, please explain why. Until You do so, You don't participate in discussion — You just pollute the discussion with baseless statements. — Dmitrij D. Czarkoff (talk) 20:35, 8 November 2011 (UTC)

• Simple descriptions of mathematical information are always permissible. There are some issues above about whether the underlying data is appropriate, but saying that "this is the median for our dataset" is no more prohibited than saying "The median height of US Presidents is a bit less than six feet tall". Encyclopedias are supposed to summarize information (including mathematical information), not just dump raw data on the readers and tell them that if they want to know more, they should buy their own stats software. WhatamIdoing (talk) 15:42, 7 November 2011 (UTC)

agree. Reasoned in discussion threads. Dmitrij D. Czarkoff (talk) 19:23, 7 November 2011 (UTC)

Graphic not requiring medians, in fact requiring minimal thought to produce

We should not be reporting median heights for presidents either unless a source reports it. We can represent the data without this messing around by having something like different coloured lines for each source in a single table. It would take no more room than the current table, it would require no explanation, it wouldn't cause a dispute, and it would show the variation in the sources which this doesn't. As to presidents we should not even have a table of their heights in the first place unless a source starts listing them out. Just because we can do something doesn't mean we should. Dmcq (talk) 12:20, 9 November 2011 (UTC)

If the subject of the article or section is about the physical characteristics of US Presidents, then we most certainly should—and we should do it by saying that the median height is a bit less than six feet, or by saying that they range from 163 cm to 193 cm, not by typing the eye-glazing details, e.g., "US Presidents have had the heights of 193 cm, 193 cm, 189 cm, 188 cm, 188 cm, 188 cm, 187 cm, 185 cm, 185 cm, 185 cm, 183 cm, 183 cm, 183 cm, 183 cm, 183 cm, 183 cm, 183 cm, 183 cm, 182 cm, 182 cm, 182 cm, 182 cm, 180 cm, 180 cm, 179 cm, 178 cm, 178 cm, 178 cm, 178 cm, 177 cm, 175 cm, 175 cm, 174 cm, 173 cm, 173 cm, 173 cm, 173 cm, 171 cm, 170 cm, 170 cm, 168 cm, 168 cm, and 163 cm", even if the particular source we are using happens to provide the full data. WhatamIdoing (talk) 23:42, 9 November 2011 (UTC)

If a source lists the heights but does not give the average then the average is not something that has shown itself to have WP:DUE] interest. Producing figures like that just because we think they are interesting or should be done is original research. Dmcq (talk) 23:57, 9 November 2011 (UTC)

Here for example is a graphic which required no original thought to produce from the data. I just copied it over to Excel and selected the very first graphic type with no options. As you can see it requires no decisions. I'm sure something a bit prettier could be done but as you can see this gets the data over okay. Dmcq (talk) 13:41, 9 November 2011 (UTC)

You graphic represents why there must be a median. On its own your graphic may be only of some use in the article about the relation on statistical research and reality. This article is no place for such has no place here. — Dmitrij D. Czarkoff (talk) 15:19, 9 November 2011 (UTC)

It could have gone in straight per policy but you're saying you just don't like it. Life isn't always simple. I've said above how WP:OI might be justified for the original image but the median line should be removed from the table whatever way it is done. Dmcq (talk) 16:10, 9 November 2011 (UTC)

Is there a consensus to include the median line?

Support

1. Support use of median in the table, as some form of summary is absolutely needed in situation of considerable amount of RAW statistic data. As median obviously does not violate WP:CALC but instead is absolutely required as per WP:NOTSTATSBOOK, WP:LINKFARM, WP:NOTDIR and WP:NOT PAPERS, it must be present. The title of the page is Usage share of operating systems, so the user comes here to get information about the usage share, not for the means of calculation of the usage share and the organisations that work in this area (though this information should also be present). Leaving the user with numerous reprints of the statistic data from the bunch of researchers is not enough. — Dmitrij D. Czarkoff (talk) 23:01, 9 November 2011 (UTC)
2. Support, but with an explanatory footnote so that readers (and future editors) know where these numbers came from. WhatamIdoing (talk) 23:50, 9 November 2011 (UTC)
3. Support. I'm still under the belief that the median doesn't violate wikipedia's original research guideline. The reason for this is that the median is clearly stated as a "median" and we are not claiming to be an original source. Likewise, the graph clearly states where the data comes from and that it is not meant to be an original source. Could there be a use of the median which does OR, yes but I don't think that's happening here. Instead, it adds a concise summary to the table, similar to an intro paragraph, but for data. Furthermore, it allows for the creation of the pie chart that people are still interested in, without which, a single "biased" source would have to be used to represent this page. Showing all sources next to each other in a bar graph is hard do read because of the clutter. Also, because the sources are fairly close to each other, visually you will estimate the median any ways to look past the clutter. Clearly, the use of the median has more benefits than any negative or inconsistencies that may exist in keeping it. That's why I continue to support the use of the median for this page. Jdm64 (talk) 00:09, 10 November 2011 (UTC)
4. Support. The use of median provides a useful summary of the data, the data is fully sourced, and the calculations are explained in the footnotes to ensure verifiability. Without the median, the table just becomes a sea of raw numbers in search of a purpose. An encyclopedia exists to provide a summarised introduction to a subject for lay readers. It does not exist to appease the purist obsessions of academics.--Harumphy (talk) 08:19, 10 November 2011 (UTC)
5. Support. That table is not very helpful if there's no reasonable summary (arguments per Jdm64). I agree with the argument that the desktop/mobile split is WP:OR and should be removed, but this is a different issue. If this is done, median or mean to summarise a table is definitely not WP:OR. 1exec1 (talk) 09:25, 10 November 2011 (UTC)
6. Support It may not be mathematically perfect to use the median - but I don't see how anyone could be fooled into thinking these figures are precise in the first place. I think a median is a sensible way to present the data. If we use a mean we are far worse off, if we don't provide a summary then it makes the article harder to read. Also I would like to express my astonishment that people are making a fuss about using this - it seems totally innocuous to include a median. If wikipedia policy forbids it, wikipedia policy should be changed. Dilaudid (talk) 09:39, 10 November 2011 (UTC)
7. Support   The median shows a central tendency. Yes, it is far from perfect. The point is: Is it helpful to the readers? Does anyone disagree that it shows a central tendency? In summary I see the argument as practical versus being overly caught up in rules and not using common sense. Daniel.Cardenas (talk) 04:22, 11 November 2011 (UTC)

   Yes, I strongly disagree that it shows a central tendency. It shows nothing of the sort. It's actually more like a vote on which language speakers have the most surveys taken about them. The numbers used in calculating the median come from website surveys done in different geographical areas - which in turn (see arguments below) means that each has a strong skew in favor of reporting results from people who speak the same language. So you're calculating the median of numbers that mean different things - and that makes it a totally invalid result. Let me try to reduce this to a simplified and much exaggerated example to make it clear why this process is broken. Suppose you have ten results for the percentage of Linux users of various web sites: six calculated from English-language web sites and four for Chinese-language web sites (English and Chinese together make up 50% of all Internet traffic - 27% English, 24% Chinese). Not many English speakers speak Chinese - so very few of them hit on Chinese web sites - and vice-versa. Let's suppose (for argument's sake) that there was a MUCH higher percentage of Linux use in China (let's say it's ~95%) than amongst the Brits, Americans, Australians, Canadians, etc (let's say those are ~5% Linux users). The numbers you'd get from those ten surveys (allowing for small error margins) might be 4%,4%,5%,5%,6%,6%,94%,95%,95%,96%. The median of those numbers is 6% - so that's what our article would 'summarize' as the worldwide median Linux use. But worldwide, 27% of Internet accesses are for English sites and 24% for Chinese sites. So the correct way to estimate the results of these surveys is to say that ~5% of 27% of the users run Linux and ~95% of another 24% of users run Linux - so on average (5%x0.27+95%x0.24)/(0.27+0.24)=47%...which means that on the basis of these numbers, we might guess that 47% of users worldwide use Linux. A median is not an "average" - not a reasonable representation of the number of users world-wide. A median is more like a vote on which countries have the most surveys taken about them - then we're quoting one of those surveys as if it represents the entire world. If (hypothetically) 20 studies has been done on users in the Vatican City (A country with only 800 inhabitants and a highly skewed demographic) and 10 studies on various other countries in the world - then our table would reflect only the operating system usage amongst 800 Catholics! Statistically, this is complete and utter bullshit. Now, it might be that we could get lucky and kinda-sorta imagine that the operating system distribution isn't that different from place to place - but that is (without doubt) an original-researchish kind of decision that we can't back up from reliable sources. SteveBaker (talk) 14:55, 17 November 2011 (UTC)

   I recommend removing stats that don't proclaim a worldview. Daniel.Cardenas (talk) 02:06, 5 December 2011 (UTC)

Oppose

1. Oppose use of median line in the table, fails WP:CALC and it is not an illustration or example. I believe however the graphic is okay as a reasonable illustration of the figures as per WP:OI if we make it obvious it is a summary generated by some Wikipedia editors, instead of "Usage share of web client operating systems. (Source: Median values from Usage share of operating systems for August 2011.)" perhaps something like "Summary of usage share figures using the median of the sources in the table". Dmcq (talk) 21:31, 9 November 2011 (UTC)
2. Oppose because the median is unimportant and not a good representation of the data.Jasper Deng (talk) 23:02, 9 November 2011 (UTC)
3. Oppose. I do find medians very useful, and I think they improve the article, but it doesn't take away the fact that they are WP:OR. To allow them would be stepping on a slippery slope, in fact, after first we allowed them, next change was to use desktop/mobile percentage from one source in order to compute desktop share of another source, which is definitely WP:OR.Wikiolap (talk) 05:23, 10 November 2011 (UTC)
4. Oppose stats, including medians, are based on sample sets and sample sets are derived from conscious choices. The median measure is used to estimate a central tendency. Both sample sets and central tendency are by definition terms relevant only to the forming of conclusions and predictions to some "real" as approximated by some "sample". So despite the deceptively simple arithmetic, the whole point to a median is to help derive conclusions and predictions, ie "original research", about something that can't be otherwise be pinned down. What's to be in the sample set is a choice. What measurements to factor and where to get them are choices. If calcs, such as median, are disputed then a good rule of thumb would be that they're not "routine" calculations. As in, a "median" of cell phone numbers would be both simple and arithmetically valid but not routine. Professor marginalia (talk) 05:39, 10 November 2011 (UTC)
5. Oppose use of median line. It is not a routine calculation anywhere near the examples of WP:CALC; it is calculated across multiple sources selected by WP editors. The WP:OR is compounded by the fact that WP editors has felt it necessary to "correct" (WP:SYN) the numbers of some of the sources before the calculation. Furthermore, it is not clear what the median results express: They do not express shares of anything as the total of those shares may exceed or fall below 100% and the sources represent different demographics and geographic/cultural usage patterns which has not been shown as neither complete nor representative for a "summary". If the article after deletion falls into WP:STAT it should be AfD'ed. However, I believe the topic is notable and that the sources are ok. I prefer a fix where sources are still set up in a table with proper notes as to possible bias. Another graph could be produced: Horizontal stacked bars, each 100% with each slice an OS or version, each bar a source. Useerup (talk) 06:52, 10 November 2011 (UTC)
6. Oppose. Of course we should not calculate a median of figures that were collected in different ways. Itsmejudith (talk) 08:07, 10 November 2011 (UTC)
7. Oppose. Let us consider the bullets at the top of this thread:
   • median may not be a routine calculation (and thus original research) - WP:CALC says "This policy allows routine mathematical calculations, such as adding numbers, converting units, or calculating a person's age, provided there is consensus among editors that the arithmetic and its application correctly reflect the sources" - so it doesn't disallow any specific kind of calculation but merely demands that there is consensus to use it. IMHO, this is a routine calculation - all you do is line up all of the numbers from smallest to largest and pick the one in the middle of that set of numbers. This is a truly trivial 'calculation'. The median of (for example) 1,5,2,6,4 is calculated by rearranging them into order 1,2,4,5,6 and picking the middle one - which is '4'. This could not be easier and does not (as a calculation) constitute anything as difficult as computing age from birthdate (which requires a sophisticated understanding of leap years and leap centuries and such). So if this were the only issue, I'd be supporting this.
   • the specific application of a median creates a synthesis over multiple sources - WP:SYNTH says "Do not combine material from multiple sources to reach or imply a conclusion not explicitly stated by any of the sources." - but the 'median' calculation merely picks one of the sources (the one in the middle of the range of opinions) and uses the result it claims. In fact, multiple sources are not being quoted here - they are merely used to decide which of the sources is to be chosen. If this were the only issue, I'd be supporting this.
   • the specific application of a median is not statistically safe as the sources have known (and declared in article) geographical and demographical biases - I'm not sure I'd use the word "biases" here. The problem is that computing the median of (say) the fuel consumption of all of the cars in the world is a useful number - it tells you something about the nature of fuel consumption in general. However, computing the median of the mean fuel consumption figures for each of the individual countries of the world tells you nothing about fuel consumption world-wide. The result isn't a median value anymore. That's just not statistically defensible. So on this bullet - I have to voice a strong oppose.
   • the sources over which the median is calculated are selected by Wikipedia editors, the selection not being supported by any source - In such matters as selection of which reliable sources to use, we have to trust our editors. It's perfectly possible to write a biassed article by ignoring half of the reliable sources. That doesn't mean that we're not allowed to write any articles - it means that we have to work hard to include a balanced set of sources to maintain a neutral point of view...this is the same deal. We can rely on editors to select an unbiased set of sources from which to compute the median. You can argue whether the set we currently have is fair and unbiassed or not - but that's not a reason to exclude the median in the table - it's a reason to improve how that median is calculated. So on this ground, I'd support retaining the median.
   • some of the sources have been adjusted by Wikipedia editors to allow for the cross-source calculation because the sources do not break out the numbers in the same way (mobile usage share). - Same deal as the previous point. This is a reason to improve how we select sources - not a reason to not do it at all.
   • medians calculated for operating systems individually yields a set of usage share percentages the total of which may exceed 100% - Again, same deal. How you do this is a matter for debate - but that doesn't impinge on whether you should do it.

   So in the end, the only reason I oppose this is because the calculation of a median of disparate things is not statistically reasonable. If you had ten sources of 'worldwide usage' demographics - then computing the mean, median and/or mode 'average' of them would be defensible - but computing the median of measurements of different things is meaningless - and for that reason only, I have to oppose this measure.

   SteveBaker (talk) 16:45, 16 November 2011 (UTC)

   I would like to clarify one matter for You. There is only one source that has known bias — Webmasterspro collect data from german language site. The other supposedly biased sources are not biased in a sense You got it. Their bias is that they serve the sites hosted in some region (eg. AT Internet counts the sites hosted in several European countries). As You might know, the place of hosting says nothing about the actual user distribution (eg. until mid-2000s most popular sites for Russian auditory were hosted in US). So, until it is explicitly stated otherwise, we may safely assume that all sources except Webmasterspro are neutral towards web client geographical distribution. As for me, this serves a better basis for excluding Webmasterspro, then Median, as the OS usage share in Germany is clearly out of this article's scope. — Dmitrij D. Czarkoff (talk) 17:33, 16 November 2011 (UTC)

   Webmaster doesn't monitor sites in Germany, it monitors German-language sites. German is the most widely-spoken language in Europe and is the main language in, IIRR, four European countries. As such Webmaster is arguably less parochial than, say, StatOwl.--Harumphy (talk) 20:45, 16 November 2011 (UTC)

   You can't tell me that the statistics of web site usage aren't skewed towards sites in the country in which the browser's computer resides. Perhaps more Russian users read US web sites but a lot more of them read Russian sites than (say) US users do. So the Russian web sites will report statistics that are more about Russian client machines than US client machines. There is no way that provides a balanced view. And if the view isn't balanced, you can't take the median of them. For the sake of argument, let's make up some hypothetical numbers for Linux usage as measured in different countries around the world...China:5%, India:5%, USA:2%, UK:3%, VaticanCity:90%, FalklandIslands:95%, Tuvalu:80%, Nauru:80%, Monaco:95%. Sort those into order and pick the middle one and the answer would appear to be that 80% of people use Linux...when in fact, the number is right around 2 to 3%. A median of this kind of set of numbers is utterly meaningless. Now, if the numbers were numbers for the entire world, gathered by different agencies without bias - then some kind of average of them would take account of experimental error and such like. SteveBaker (talk) 21:21, 16 November 2011 (UTC)

   Absolutely not. There is no correlation between place of hosting and auditory web clients' origin. So, You can regard all counters (except Webmaster Pro, which is biased towards German language, which is predominant in Germany and not elsewhere) as unbiased. — Dmitrij D. Czarkoff (talk) 21:36, 16 November 2011 (UTC)

   That makes no sense at all. 'Auditory' is an adjective that means 'relating to the sense of hearing'. Do you want to try again?--Harumphy (talk) 21:44, 16 November 2011 (UTC)

   Sorry, I'm not a native speaker, and I currently live in alien linguistic environment, so I have several languages mixed in my head. — Dmitrij D. Czarkoff (talk) 21:50, 16 November 2011 (UTC)

   How can you possibly claim that: "There is no correlation between place of hosting and web clients' origin."??? You are saying that the probability of someone living in Austin, Texas accessing the site: http://www.miniofaustin.com/ (a car dealership situated in Austin) is statistically similar to that of someone living in Ap Dong Binh, (Southern Vietnam)!?! Hell no! 99.9% of the traffic headed to that site will be from within a 50 mile radius from the server itself. There is absolutely a strong correlation for the vast majority of web sites in the world. Only 1.8 billion of the 7 billion people in the world speak English - so discovering which operating system is being used by people accessing an English-language site can only possibly tell you about the operating system usage numbers for ~25.7% of the world population. (Well, the numbers may be a little better than that because we might expect computer users to have better-than-average language skills - but for sure it's not going to hit 50%). But if you consider a site like Baidu (the number one Chinese search engine - with 200 million regular users and no English-language translation) - the number of non-Chinese people who use it must be utterly negligable. How can you possibly claim that measuring how many Linux users access Baidu tells you anything other than how many Chinese people use Linux? That's just crazy! So if you have some sites measuring the number of Linux users in one geographical area and other sites measuring the numbers in other areas, calculating the median of them is an utterly meaningless thing to do. SteveBaker (talk) 14:11, 17 November 2011 (UTC)

   Statements like "that's just crazy" don't really assume good faith. "I respectfully disagree" works in instances such as these. -- Scjessey (talk) 14:32, 17 November 2011 (UTC)

1. Sorry, I might have explained it in wrong way. In Your example my point is that the site will be accessed mostly by those who live in Austin, but the hosting itself may be located anywhere. I'm not sure, that this site's hosting is actually located in US. And I am sure that the counters that don't explicitly state their lingual/geographical bias are largely biased towards world-wide sites.

   P.S.: In fact, I ran traceroute and it revealed that my hit has landed in Los Angeles, which is pretty far from Austin, isn't it?

   Dmitrij D. Czarkoff (talk) 15:16, 17 November 2011 (UTC)

   OK - but at least it's still within the US. We don't expect many Chinese language sites to be hosted on servers in the US or other English-speaking countries - or many English language sites to be hosted on servers in China...so the bias is still strongly present. Companies don't tend to host too far from their target audience because they want to minimize delays in loading the site. Remember, the amount of Internet traffic from chinese language speakers is about the same (24%) as the entire English-speaking world (27%) (See Global_Internet_usage#Internet_users_by_language) - so unless your surveys are including hits on a representative percentage of chinese sites, you're missing 24% of the computers out there...and if your survey focuses only on English language sites, then it's missing 73% of the users out there. There is no evidence that English speakers have similar statistical distributions of operating system usage as other language speakers - so we're on incredibly unsteady ground here. But what makes this MUCH worse is that we're using a median score - which effectively means we're treating these results as votes for and against some choice of which number to present. If statistical differences exist between chinese and english speakers use of different operating systems - then taking the median number effectively makes this a 'first past the post' vote on whether chinese speakers or english speakers should have their usage figures reported on Wikipedia! We're presenting this (to the statistically and mathematically naive) as if these numbers were some representation of how many users use which operating systems...and that's utterly bogus. SteveBaker (talk) 16:22, 18 November 2011 (UTC)

   OK. So if we just keep the sources that explicitly state their stats are global (worldwide), it will be OK? That effectively means removing AT Internet and Webmasters Pro. — Dmitrij D. Czarkoff (talk) 22:20, 18 November 2011 (UTC)

2. Regardless of the simplicity of the calculation this is OR, in my opinion. Thenub314 (talk) 23:05, 18 November 2011 (UTC)
   • Oppose Median is a simple calculation, however application to Wikipedian editor selected sources makes its application here original research. Gerardw (talk) 13:17, 27 November 2011 (UTC)

Conclusion: No consensus

Only routine calculations are allowed per WP:CALC, and only if there is consensus among editors that the arithmetic and its application correctly reflect the sources.. Such consensus does not exist. I am deleting the median, removing the expert request. I'll leave the graph in for now, pending suggestions on how to visualize data without performing synthesis.Useerup (talk) 18:41, 10 November 2011 (UTC)

You can't do such statements until discussion ends. It didn't. Furthermore, there is no consensus on leaving statistic data without summary. — Dmitrij D. Czarkoff (talk) 18:48, 10 November 2011 (UTC)

Yes I can: WP:OR. Even if you consider it "routine calculation" WP:CALC such calculations are only acceptable when there is consensus among editors that the arithmetic and its application correctly reflect the sources. Per the policy it is WP:OR, and original research must be removed. The WP:BURDEN of evidence lies with the editor who adds or restores material (i.e. you). There is no evidence of such consensus, rather there is evidence that such consensus does not exist. Since you cannot demonstrate consensus that median is a routine calculation it is gone. So please revert your last edit which restored the median. Useerup (talk) 19:01, 10 November 2011 (UTC)

Again, You didn't prove the absence of consensus yet. I will remove this line on my own, but only after the relevant sections here become less active. Now the discussion is ongoing, so there is no proof of Your claims that consensus is not reached. — Dmitrij D. Czarkoff (talk) 19:05, 10 November 2011 (UTC)

I do not have to prove anything. It is you who are restoring material, the WP:BURDEN is on you. The support/oppose sections above do not support that there is consensus that median is allowable under WP:CALC. Therefore, it is not. I say again: Please remove the median, your revert is in direct contradiction to WP policies. Useerup (talk) 19:16, 10 November 2011 (UTC)

The process of reaching consensus is out of scope of WP:BURDEN. Your edit fails both WP:ADMINSHOP and WP:DRNC, and I reverted it as such. The problems are not solved yet, I see no obligation to undo my revert. — Dmitrij D. Czarkoff (talk) 19:29, 10 November 2011 (UTC)

Please see WP:RFC#Ending RfCs. RfC lasts until consensus has been reached. Currently there is no consensus as you admitted, so you can't close the discussion. Also, per the same policy: "[Closure of a RfC] should be done by an uninvolved editor". 1exec1 (talk) 19:27, 10 November 2011 (UTC)

Lack of consensus has clearly been demonstrated above. "Routine calculation" was the only way to avoid the median calculation being classified as WP:OR. But I can wait. Useerup (talk) 19:44, 10 November 2011 (UTC)

The first comment in the thread about consensus was posted 2011-11-09 20:31 (UTC). Your edit happened 2011-11-10 18:43 (UTC). No fact can be established in 22 hours on internet. That said, the common duration of such procedure is 30 days. — Dmitrij D. Czarkoff (talk) 19:49, 10 November 2011 (UTC)

I think we should ask for outside advice through the use of Wikipedia:Dispute resolution noticeboard. I've already typed a request up, but haven't submitted it yet. Should I submit it? Jdm64 (talk) 20:18, 10 November 2011 (UTC)

Will we need to establish a consensus on this one also? ;) Jokes aside, I think that's a good idea - it would certainly be beneficial to see some help of uninvolved people that could guide the discussion. 1exec1 (talk) 20:38, 10 November 2011 (UTC)

Do you have any reason to believe that will lead to another conclusion? IMO this is mainly policy discussion and it is hard to see how it will lead to another result. But if you believe so, then go ahead.Useerup (talk) 20:32, 10 November 2011 (UTC)

You should, as some of us are a step away from something very wrong. — Dmitrij D. Czarkoff (talk) 20:34, 10 November 2011 (UTC)

I oppose the inclusion of that median graph. I especially oppose its current position, at the top of the page. IMHO, this MISLEADS a casual reader who comes to this page. It gives an "authoritative" impression, that the percentages shown in the graph are a "good" approximation to OS market share. That is, the reader might miss the distinction between measuring "web client" and measuring the actual installed base. (Which there is no known way to do.) I do favor providing the user with a line at the bottom of the web clients table, which provides that median information. IMHO, that gives a better "context" to the information. Can such a line be automatically calculated from the contents of the table, so that when any editor updates the table, the calculation is updated? ToolmakerSteve (talk) 03:07, 11 November 2011 (UTC)

I consider the median to be a justifiable calculation because it is (a) straightforward, (b) intuitive, and (c) easily verified by any reader. I disagree with the assertion that the "selection of sources" makes it OR. ESPECIALLY if it is merely a summary line in a table. What I oppose, is placing that information in such a way that it is given more prominence than is justified; e.g. don't make it appear to summarize the entire article, don't make it appear to answer the user's question authoritatively. There is no authoritative answer; this needs to be clear. ToolmakerSteve (talk) 03:14, 11 November 2011 (UTC)

A bit more about the topic of being OR. As long as it is presented as a mere "calculation aid", then it makes sense to me. A knowledgeable reader, faced with the table, would want to know the average. An unknowledgeable reader, would want to know what they should do with the information, to make sense out of it; if the "knowledgeable" answer would be "calculate the median", then do so, for them. Presumably a statistician would also calculate an "error estimate", but I hesitate to suggest doing so, because that risks being misleading. Specifically, it is only an estimate of error based on the numerical differences between the sources -- it assumes the sources are "free of errors". A distinction that would be lost on 99.9% of readers. Better to "do the simplest thing". ToolmakerSteve (talk) 03:32, 11 November 2011 (UTC)

A statistician would be more likely to produce something like in the article Summary statistic. For disparate data it is just a bit finger in the air. For this sort of stuff it wouldn't matter much outside WIkipedia but we need to follow our standards or figure out new standards. Dmcq (talk) 12:02, 11 November 2011 (UTC)

OK, I just read the rest of the discussion. Did not realize that the median was originally added as a line in the table, and that this was considered MORE controversial than the current graph. IMHO, the reasoning here is backwards. The graphic risks being noticed/perused/referenced/copied on its own; a median line in the table is "safer", it would tend to be examined with the table, not referenced on its own. I agree that such a median line would need to be clearly distinguished from raw source data. IF it is so distinguished, I completely fail to understand what is controversial about it. However, I agree that UNTIL/UNLESS consensus is achieved, it should be omitted. ToolmakerSteve (talk) 04:15, 11 November 2011 (UTC)

Ask yourself what is wikipedia's goal? To provide reliable information. Does presenting a central tendency of disparate data helpful to the readers? The alternative is to present a bunch of numbers that few want to look at or to fight with editors which one represents the "best" numbers. The central tendency from the table also known as a "summary" is used in quite a few other pages. It is a compromise between options that are not ideal. The compromise helps people understand the "usage share of web browsers". Daniel.Cardenas (talk) 05:24, 11 November 2011 (UTC)

Changing subjects somewhat. This isn't about OS market share. This is about which web browsers are used most to browse the internet. Daniel.Cardenas (talk) 05:24, 11 November 2011 (UTC)

Agree but that should be a separate section from this. Best to separate problems. That's why they have 'web client' operating systems further down but yes the title does imply something different. Dmcq (talk) 11:53, 11 November 2011 (UTC)

WP:TLDR, and the future of any controversial article on Wikipedia is: there is no consensus. ASCIIn2Bme (talk) 21:15, 13 November 2011 (UTC)

As advised by ItsZippy at Wikipedia:Dispute resolution noticeboard‎, applied to Mediation Cabal. — Dmitrij D. Czarkoff (talk) 21:51, 13 November 2011 (UTC)

Well that was a waste of time. And I really can't see the point of mediation. There hasn't been any failure of civility and they only seem to deal with conduct problems. Ah well I'll see where all this leads to or if Wikipedia is totally unable to deal with anything except conduct problems. I know that WP:CPUSH is a dreadful problem and this seems to be getting into the same rut. I hope this isn't going to be a festering sore in Wikipedia for the next few years. Dmcq (talk) 22:43, 13 November 2011 (UTC)

I've acted per resolution of Dispute resolution noticeboard‎. If that doesn't work, so be it. — Dmitrij D. Czarkoff (talk) 00:44, 14 November 2011 (UTC)

I'm not blaming you for anything, just bellyaching about the process. Dmcq (talk) 00:49, 14 November 2011 (UTC)

This process is supposed to help the parties to reach WP:CONSENSUS if those really try to. Our case is special — everyone (well, at least me) is trying to actually explain his (or her) position to the others. No luck even here, yet, as it seems.

Effectively, we've already reached consensus on one issue: the table has to be split, so that there would be no need to guess the mobile figures. Though it seems evident, the practice was established for several years, as it seems.

May be eventually we'll have some luck? If not, there is the next step - arbitrage - that has the right to enforce its decisions.

The funniest thing of all, is that our discussion now is several times longer then the article in question.

Dmitrij D. Czarkoff (talk) 00:58, 14 November 2011 (UTC)

We do not have consensus on whether median is a routine calculation

Can we agree about that? If not, please explain why you think there is consensus here Useerup (talk) 07:26, 14 November 2011 (UTC)

We can't currently declare neither consensus nor its absence. Effectively, the voting came in favour of Median (even not counting an editor who plainly disappeared, other one who said she isn't a party of the dispute and the third who clearly stated that median improves the article after calling oppose). The dispute resolution process You've started with RfC is now in process. When it all finishes, we would either have a consensus or formally resolved case. Why are You that much in a hurry? — Dmitrij D. Czarkoff (talk) 09:18, 14 November 2011 (UTC)

Either there is consensus or there is no consensus. Which one is it? Wikipedia is WP:NOTDEMOCRACY and disputes are NOT resolved through majority votes. Useerup (talk) 13:57, 14 November 2011 (UTC)

We still don't know, as we can't declare anything until the dispute resolution procedure is over. It is the third time I tell You the same exact thing. What is the point for asking the same exact question three times in a row? — Dmitrij D. Czarkoff (talk) 15:03, 14 November 2011 (UTC)

The dispute resolution procedure can go on for a long time. I'm sure you have eyed that by now. Meanwhile, there is a lack of consensus and this is actionable. I'm sure you are also aware of that. We can still go through the procedures and act accordingly when (and if) a resolution is found. If you don't want me to point out again and again that WP:NOTDEMOCRACY then stop suggesting that we can decide disputes by voting. What is the point of suggesting voting when you know that it is not an acceptable way to reach consensus? Useerup (talk) 15:42, 14 November 2011 (UTC)

Note, that the policy You linked states that polls are no more binding than any other consensus decision, while WP:CONSENSUS states that "Consensus" on Wikipedia does not mean that decisions must be unanimous (which, although an ideal result, is not always achievable). So yes, we have a consensus. You dispute it — go an prove Your opinion. We periodically aid You with different means... — Dmitrij D. Czarkoff (talk) 15:26, 14 November 2011 (UTC)

You have no evidence either for or against a consensus on the question of whether this is a "routine calculation". The question asked in the above !vote was whether or not "to include the median line". That's an entirely different question! It was not a !vote about whether the mathematical operation of computing a median is an acceptable operation under WP:CALC.

In my opinion, it's clearly a very simple calculation - much simpler than the example given in WP:CALC of computing a persons age from their birthday (think leap years and leap centuries, for example). There were half a dozen bullet points against "including the median line" - and WP:CALC was only one of them. I !voted oppose - but I opposed on entirely different grounds (that the statistical operation is invalid when you're using numbers measured in disparate ways). It's like saying "the average size of a person in America is 50 inches" and backing that up using the median of measurements of the height of Texans, the girth of female Californians, the penis length of male New Yorkers and the inside leg measurements of two year olds in Arizona. Sure, you have a number and it's very easy to calculate - but it's just meaningless. That's the problem here. SteveBaker (talk) 16:59, 16 November 2011 (UTC)

Erm... I might not have chosen the examples SteveBaker did, but he has a point. I also think the number is meaningless, and worse we give the impression it has meaning by including it. Thenub314 (talk) 23:03, 18 November 2011 (UTC)

Mobile results from table in Web clients

It seems that we reached consensus on the improper synthesis in case of the mobile/desktop split computation in the table. Now we should decide on how to solve the issue. As I see it, we have a problem here:

1. If we split the table or just remove the mobile clients with no further modification, the sums won't add up to 100%, which would effectively render the table useless.
2. Another option would be to stretch the results (multiply the values of included systems by ratio of included/total), but this may be disputable as per WP:CALC and WP:SYNTH, as the other column will content even more variable results per review.

So may be it's time to say good by to table and have the section split on per-source basis? If so, that would dissolve the problem of median once and forever, which is a fairly nice side effect, by the way. Dmitrij D. Czarkoff (talk) 11:14, 14 November 2011 (UTC)

The easiest way to resolve this is to remove the two relevant sources from the table - Clicky and StatOwl.--Harumphy (talk) 16:02, 14 November 2011 (UTC)

Which data did you take from Clicky? As far as I can see they report only day-by-day figures. Did you calculate average of the month or just take one-day data? 1exec1 (talk) 19:01, 14 November 2011 (UTC)

As its says in the footnote, it's averaged over the last seven days of the month.--Harumphy (talk) 00:18, 15 November 2011 (UTC)