Jump to content

Wikipedia talk:Mediation Cabal/Cases/13 November 2011/Usage share of operating systems

Page contents not supported in other languages.
From Wikipedia, the free encyclopedia

Discussion moved from main page

[edit]
Initial discussion

I believe the statement of the problem above is a reasonable one. It is a straightforward content dispute with disagreement over application of some basic policy. There has been no conduct dispute.

I believe the median line is OR and straightforwardly against WP:CALC. I don't see it as a good or necessary thing to put in to improve the encyclopaedia. 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". However for the image there are alternatives which don't look quite so good but wouldn't involve any OR. I don't believe a median can be counted as a summary in Wikipedia terms. I believe a summary in Wikipedia is an abbreviated form or précis of the contents and not summary information.

On another note I believe it would be good if Wikipedia had a policy about actual summaries where we infer something and only cite the article but we don't have such a policy at the moment, anything like that is counted as OR. Even if we had such a policy I'm not sure this median would count as it involves disparate data. Dmcq (talk) 23:12, 13 November 2011 (UTC)[reply]

With all my respect, if the Median will be deleted, the image will follow. I don't see any need to discuss it here. — Dmitrij D. Czarkoff (talk) 00:27, 14 November 2011 (UTC)[reply]

To summarise my position on the question stated above (each question is answered at respective number):

  1. If Median line was considered a statistical study on the table data (see 5), it would clearly violate WP:SYNTH as reasoned above.
  2. The choice of means of statistical research would need a professional knowledge in the corresponding field, until the Median line is considered a mere summary with no statistical properties. Apart from that median is calculated trivially, without violating WP:CALC.
  3. The Median line neither formally nor factually doesn't constitute WP:OR, as each number is referenced, the choice of number among the referenced numbers is done according to the trivial calculation (algorithm) which is in turn well established, described in many sources and does even have an article on Wikipedia (as wikilinked above).
  4. As per policies named in the explanation of the question, the statistical data presented on the page may not be kept without summary. As the statistical data effectively constitutes the scope of article, the page with average hit rate >30000 per month will need to be deleted. Furthermore, as most readers are unfamiliar with many of statistical concepts, the majority of them will misinterpret the statistical information presented without summary, which clearly violates the goals of Wikipedia.
  5. As per referenced essays the Median line should be considered a summary and (due to the reasons described in 4) WP:WIARM should be applied to excuse it from WP:CALC, as no better way to summarise the statistical data in the article was worked out.

Dmitrij D. Czarkoff (talk) 00:27, 14 November 2011 (UTC)[reply]

The median helps these articles by summarizing the content.  No author has suggested a viable way to summarize tabular data with out statistics.  From the strictest point of view adding a median does violate wp:or, although I suggest wp:ignore should apply here.   Thanks, Daniel.Cardenas (talk) 01:08, 14 November 2011 (UTC)[reply]

Not quite true: I have suggested an alternative graph which does not need to compute any medians, which does not need to "summarize" but which can convey truthfully the differences between the sources while still allowing the reader to get a sense of overall patterns and variation. Illustrated here as thumb. Another editor has suggested an alternative graph as well. Useerup (talk) 18:40, 14 November 2011 (UTC)[reply]
This graph is worse then absence of summary: it summarises nothing, only clutters the article giving the same info as the table, but without details. — Dmitrij D. Czarkoff (talk) 18:45, 14 November 2011 (UTC)[reply]

To summarize my opinion: The median calculation violates a number of WP:NOR rules:

  1. The median calculation involves sorting, selection and sometimes mean calculation. It is not a routine calculation anywhere close to the examples given in WP:CALC which are all straightforward and uncontroversial. Especially the applicability is questionable. Compare "calculating a persons age from his birth date" to "calculating the median of usage share percentages". Median is a statistical function which may have a well-described definition, but the applicability of the median is a lot more involved.
  2. WP:CALC contains a provision that there must exist consensus among editors for a calculation to be allowed - even if it is considered "routine". There is provably a lack of such consensus.
  3. The median (as used in the articles) is being calculated combining multiple sources and thus also violates WP:SYNTHESIS. The new position being advanced by this synthesis is a number assigned as "the median usage share" of each operating system/browser, a number not supported by any of the sources and the methodology not supported by any source.

The median calculation violates all of the above, and any single one of them should suffice to dismiss the median. Unfortunately the median as it is being used in these articles has not only not been published/reviewed according to WP:NOR and WP:SOURCE, it also abuses statistical methodology:

  1. The median is chosen by the editors because they want to arrive at some number describing a perceived central tendency (which can then be plotted in a graph). But the applicability of such statistical analysis depends on the samples having been randomly selected. In this case Wikipedia editors have selected the sources in an anything but random way. The individual sources may have followed sound statistical procedures when selecting the samples on which they report, but the samples upon which the median is calculated have not been randomly selected and do not constitute a complete population.
  2. For the median to be meaningful the sampling must draw from the same population. But the sources draw from a number of different populations with different biases: Demographic (commercial oriented stat counters), geographic, cultural (statistics for select native languages). This amounts to computing the "median weight" from the mean weights of apples, oranges, bananas and watermelons. What does the result say? That the median weight of "fruits" is X? Why then, were the peaches left out? And does it matter that only 3 oranges were behind the orange mean weight but 300 apples were behind the apple mean weight?
  3. The statistics counters use different sampling methodologies, leading to different biases even for the same populations: Some count only unique visitors (each visitor weights as much as any other regardless of activity), others count page impressions. This leads to further incompatibility between the numbers being used to calculate the median.
  4. The sources do not break out the same set of observations. Some sources ignore mobile usage and report only shares of desktop operating system usage, others report the shares for all operating systems/browsers. The sources without mobile shares thus exhibit relatively higher shares for the desktop operating systems, skewing the median calculation. This fact was not lost on the editors maintaining the median, and led them to "correct" the shares reported by those sources with factors derived from the other sources. The dilemma was: Skew the median or "adjust" (synthesize) the numbers.
  5. The median is presented as if it somehow represents the operating system or browser share in some summarized form. But it does not add up, literally. The total of usage share medians for browsers exceed 102%, the total of usage share medians of operating systems fall well short of 100%. The sum of usage share medians for individual Windows versions arrives at a different number compared to the median for all Windows versions. This is not due to rounding errors, it is an effect of how inappropriately the median is being used.

Each of the above points demonstrates how the basic principles of statistical analysis are being violated by the way the median is being applied. It is unprofessional and this is the very reason why WP:NOR is so important: To keep questionable research out of Wikipedia. Useerup (talk) 18:33, 14 November 2011 (UTC)[reply]

No principles of statistical analysis are being violated as no statistical analysis is conducted. Don't mislead the others. — Dmitrij D. Czarkoff (talk) 18:48, 14 November 2011 (UTC)[reply]
Which median is it you calculate if not the statistical median? Here is what Wikipedia says about the Median: "In probability theory and statistics, a median is described as the numerical value separating the higher half of a sample, a population, or a probability distribution, from the lower half.". Is that not the one used in the articles? Useerup (talk) 19:56, 14 November 2011 (UTC)[reply]
Don't accuse other editors of trying to mislead Useerup (talk) 19:59, 14 November 2011 (UTC)[reply]
Per Your logic every person using a hammer is hammering nail. If median is used that doesn't mean that statistical research is carried on. It would be statistical research, if there was a statement that this median constitutes a real web client usage share, but no such statement. Instead, everyone insists it isn't the case. — Dmitrij D. Czarkoff (talk) 21:10, 14 November 2011 (UTC)[reply]

Good evening;

I have assigned myself to this case, as you can see from teh infobox above. I've undertaken some preliminary study of your case, with the relevant policies to hand. I'm going to issue some preliminary rulings: 1. WP:CALC is blown out of the window - its rules have been breached. The statistical operations referred to in the articles which are at the centrepoint of this discussion would require a computer scientist to complete them accurately. Therefore, they cannot be performed by the ordinary layperson (which is the point of that entire policy - to ensure that calculations suitable for the ordinary man are posted).

2) due to the fact that there is a self-confession that none of the sources refer to the calculations mentioned, the rule of WP:OR has also been broken.

I shall undertake a more detailed study tomorrow, and issue a fuller judgement tomorrow or Wednesday. --Thehistorian10 (talk) 20:46, 14 November 2011 (UTC)[reply]

Leaving aside the question about whether "issuing rulings" is really compatible with the concept of mediation, I am curious about your definition of "the ordinary layperson". In the USA, identifying the median from a given set of data is a basic mathematical skill that students are supposed to master in grades 3 through 5—in other words, around the age of 10. Is your "ordinary layperson" less capable than an ordinary American ten year old? WhatamIdoing (talk) 21:18, 14 November 2011 (UTC)[reply]
If kids 10 years old can do this, it is quite unfathomable that the articles manage to get it so wrong. Maybe the considerations on when the median is appropriate are not so simple? But it really doesn't matter, because the WP:CALC is pretty clear, and as the recent discussion on changing the policy showed, there is broad consensus for not changing the policy to allow for statistics.--Useerup (talk) 22:44, 14 November 2011 (UTC)[reply]

Question I am fairly unfamiliar with this case, but how is the median being calculated? As I recall there was some choice to be made when the median landed in between two elements of the list? As I recall any way you choose to average the two elements to get a number in between is considered equally valid. I am uncertain in this case but I personally feel that say calculations of variance are not appropriate on WP, as you must decide if you want to divide by N or N-1, and both formulas are commonly used. And how would the outcome of this effect more detailed statistics, such as the inter quartile range, which is more or less the same operation? Thenub314 (talk) 21:54, 14 November 2011 (UTC)[reply]

Offhand, interquartile ranges would probably fall into the category of "simple descriptive statistics", which the {{supplement}} to NOR approves of (see the WP:NOTOR#Simple_calculations section). The range (statistics) and median are even simpler examples from this category. WhatamIdoing (talk) 21:59, 14 November 2011 (UTC)[reply]
As the articles prove, it can be quite difficult to apply statistics correctly. The NOR policy is intended to avoid situations such as these where wp editors select a bunch of numbers and start calculating statistics on them. Calculating a median is somewhat simple. But if it has to be meaningful at all, the sample over which it is calculated must meet certain criterias. For instance, the samples must have been randomly selected from the same population. The medians here violate both. And that is just some of the problems. Please also observe that the essay linked to by user:WhatamIdoing is not an official policy, and that in case of conflicts (as is the case here) you are referred to the core policy. user:WhatamIdoing also forgot to disclose that she has been editing the very section on calculations which is in conflict with the core wp:nor policy. --[User:Useerup|Useerup]] (talk) 22:21, 14 November 2011 (UTC)[reply]
This article proves only one thing: the usage share of operating systems is unknown and isn't likely to be known ever. But this discussion shows a way more things:
  1. You can't see difference between using median in statistical research and in summarising.
  2. Nearly everybody who actually maintained the article for years routinely calculate it regardless of their professional skills.
  3. You failed to provide any other mean to make a valid summary of statistical data, instead asking to adopt Your graph which does not summarise anything.
  4. You decline to even comment on application of WP:WIARM and issues with WP:NOTSTATSBOOK.
  5. You pushed Harumphy out of maintaining the monthly changing article.
Please, keep calm and show the respect to the due procedure and its participants. WhatamIdoing is no way obliged to disclose anything, and use of "the" in front of username is offending.
Dmitrij D. Czarkoff (talk) 00:34, 15 November 2011 (UTC)[reply]
Ad 1: You are correct, I can't see the difference between using median in statistical research and in summarising. You are once again introducing a euphemism, re-labeling the application of the statistical function median into summarizing. You can call it what you want, it is still statistical analysis. Perhaps you use this statistical analysis because you want to summarize some data. For good reasons that is not allowed on Wikipedia.
Ad 2: Yes, even a ten year old can calculate a median. Have you totally ignored the point of apples, oranges and grapes? Have you totally ignored the point about sampling and populations? That editors have done it for years does not make it more correct (in fact - it makes it worse!).
Ad 3: I am under no obligation to provide other means to make a summary of the statistical data, yet I have suggested a graph which does not perform any WP:OR (it is an illustration). You don't like it? That's ok. But you are once again trying to disguise analysis under summary, because you have observed that summary is allowed but analysis is not. If you want to do a proper summary you can write statements like Windows is by far the most used operating system for accessing the Internet and In recent years usage share of Mac OS X has seen an uptick while Linux usage share has remained flat. Write descriptive statements which can be straightforward derived from the statistics and supported by all of the sources. Summary is meant to be text and possibly illustrations, not analysis and not original research.
Ad 4: No I don't decline, I just don't think they are relevant. But I will say this: WP:WIARM is an essay. Read the disclaimer at the top. WP:NOR is a core content policy. Under no circumstances should anything like WP:WIARM be used to introduce erroneous and misapplied statistics into an article. If - after we remove the median - you believe the article falls under WP:NOTSTATSBOOK you can AfD it. Personally, I think the topic is notable and the article contains a good deal more than the misapplied median.
Ad 5: I really cannot answer to Harumphy reasons for "taking his ball and go home". I did consider whether it was an attempt at creating undue anti-sentiment, but I chose to ignore it at the time and instead focus on the topic. It really is Harumphys own decision and I did not push Harumphy to do anything. I suggest you refrain from making such accusations.
I do not understand why you feel it necessary to tell me to keep calm. I am in no way agitated. WhatamIdoing was referring a newcomer to an essay (i.e. not a core content policy), specifically to a section which is in contradiction to the core content policy. She has been contributing to that very section. So she directs an editor to read a contradiction of a core policy. I just thought that disclosing the fact that she had been contributing to that very section would have been the proper thing to do. I assure you that I did not mean to use "the" in any offensive way. I have edited it away. Useerup (talk) 07:12, 15 November 2011 (UTC)[reply]
  1. Please, stop calling euphemism everything that doesn't help You to delete the median. As You claim to be unable to understand this difference, I'll draw a simpler example: the matches are tools for starting fire. This doesn't necessarily mean that a person showing geometry trick with matches (a fairly common use in Russia) is going to burn anything. The purpose of the instrument doesn't limit the instrument usage. Is this clear?
  2. I replied to these points several times. They would be relevant in statistical study, which is out of the scope here, as nobody performs such in the discussed article.
  3. Effectively this one would fall under WP:OR and WP:SYNTH. Not to say it's inaccurate. This proves my point — it's really hard to give a proper summary, and You were not even thinking about that. You follow Your narrow target without considering side effects. That's the secondary reason why WP:WIARM was linked to the top of WP:OR.
  4. The essays here are used to clarify the usage of policies. That's why this essay is linked in topmost section of WP:OR.
  5. I don't make accusations, I report an established fact You are already aware of.
  6. user:WhatamIdoing was refering to the essay clarifying the stubby wording of WP:CALC in this case. Until the contradiction is found, this essay should be considered as a supplement to WP:OR. The contradiction was not found, as the referenced parts merely describe what kind of operations are not covered with WP:CALC, while WP:CALC itself doesn't limit its scope.
There's no need to disclose anyone's participation, as there is a page history there, where one can see that some of the participants (including me) did several contributions to the section. This is a thing one doesn't need to stress, specifically by using such inappropriate wording as to disclose.
Dmitrij D. Czarkoff (talk) 10:21, 15 November 2011 (UTC)[reply]
If interquartile range is clearly beyond the spirit of something a "reasonably educated person" could be able calculate and understand. And I think is not be covered by WP:NOTOR#Simple_calculations. Mostly there are problems with simply pointing at the NCTM and saying that every 5th graders should no this, so everyone should know this. First the NCTM guidlines are not used in all school districts. Secondly at the level of a high school diploma there is a substantial number (if not the fast majority) of people that cannot distinguish between the concepts of median, mode and mean. At least in my anecdotal experience. Finally the NCTM guidelines did not exist for a large number of our readers. The concept that elementary statistics is important in elementary school mathematics education and it would not be correct to assume that this was true for readers who are middle age or older. Thenub314 (talk) 22:45, 14 November 2011 (UTC)[reply]
Dear god, talk about making a mountain out of a molehill. I'm not a computer scientist and I have been calculating the medians using a pencil and paper each month. It's very easy: cross out the highest and lowest values in the column, rinse and repeat until there are no more than two left. If there are two, take the mean. If there's just one, don't, because you've already got the answer. --Harumphy (talk) 23:21, 14 November 2011 (UTC)[reply]
And what's the median of 5 apples 3 oranges and 10 grapes? Dmcq (talk) 23:28, 14 November 2011 (UTC)[reply]
To make the same point a somewhat different way. Even if the calculation is easy, who says the median of the numbers at this page is at all a meaningful or reasonable number to look at. We do, because we calculate it and we include it. But the issue is that that there are situations where medians are not useful and/or meaningful. By including the medians in the table we are implicitly conveying the idea that this particular measure is meaningful. But why not include a column for the Geometric mean, which is also relatively easy to compute. The answer is there is no reason to look at the geometric mean for this data set. So why is there a reason to look at the median? I think this is really where synthesis comes in. Thenub314 (talk) 23:57, 14 November 2011 (UTC)[reply]
You're both conflating two separate issues: (a) whether or not median is an appropriate calculation, (b) whether or not it's a simple calculation. My comment aimed to address only the latter.--Harumphy (talk) 00:08, 15 November 2011 (UTC)[reply]
I did not mean to cause confusion. I fully grant that the calculation we are doing is not difficult. But we are making a choice to do this calculation, which seems a bit dubious. Thenub314 (talk) 00:22, 15 November 2011 (UTC)[reply]
The main problem here is that we have to make this choice, as the raw statistical data is (1) of no use to most readers and (2) is not an encyclopaedic subject. Though the actual matter this statistical data is supposed to explain is an encyclopaedic subject, so there must be some summary the readers could make use of. So the question is not whether to remove median or not, but whether there is something we can replace it with. — Dmitrij D. Czarkoff (talk) 00:45, 15 November 2011 (UTC)[reply]
About (1), the positions for having the median line seem contradictory. It is simultaneously a simple enough calculation as to be easily verified by any reader but too difficult a calculation that the raw data does not allow them to calculate this on their own if they want it? About (2), you seem to be saying the data by itself shouldn't be part of the article, then why do we have it there? If there is nothing useful about it why not include only the summary statistic? I suspect the answer is because we could not source that data, so the column makes it look sourced, but it is still OR. Thenub314 (talk) 02:49, 15 November 2011 (UTC)[reply]
It appears I should explain these points in more detail:
  1. It is a simple enough to be calculated by any reader, and something of a kind will be calculated by most of readers (this is the way the people comprehend the data), but then it would be a quick glance comprehension that would generally be even less accurate.
    The point of of no use to most readers statement was about raw statistical data on its own, without summary, as such data can be easily found on internet.
  2. Exactly! That's why the questions 4 and 5 were posed. There is no doubt that this line (to some extent) does constitute original research (in terms of WP:OR, as opposed to statistical research; I don't want to run another round with Useerup) and synthesis. The question is whether these violations should be excused per WP:WIARM as they are fairly minor and vital for the article's goals.
Dmitrij D. Czarkoff (talk) 12:05, 15 November 2011 (UTC)[reply]
You question the logic of median application as if we discussed a statistical research here. But no one does want to do such research. We just try to find a way to make the statistical data properly summarised, so that the question You pose would not be an open question to each and every Wikipedia reader, who opens the article in question. — Dmitrij D. Czarkoff (talk) 00:34, 15 November 2011 (UTC)[reply]
I cannot follow you, could you explain more? Why would my question need to occur to each and every reader? Thenub314 (talk) 01:22, 15 November 2011 (UTC)[reply]
The problem of comparing apples to grapes is not a problem of the median line — it's a problem of the data in the table. Removing median line won't fix the issue of the sources that don't give any reliable analysis of the operating systems' usage shares. — Dmitrij D. Czarkoff (talk) 01:59, 15 November 2011 (UTC)[reply]
Do you agree then that comparing the different rows is like comparing apples to grapes? Thenub314 (talk) 02:49, 15 November 2011 (UTC)[reply]
Anywhere the median is applied there is a bit of apples and grapes. Perhaps the data samples are different because where the data was taken, or who was taking the measurements, or what time the data was taken. If there were no differences then there would be no point in taking the median because the data would be the same. For usage share of web browsers each professional statistics site is conveying what they believe the "usage share" is. The median tells us what is the central tendency of this data amongst the different ways of measuring. People can ignore it and look at the raw data if they choose. Daniel.Cardenas (talk) 04:44, 15 November 2011 (UTC)[reply]
That is simply not true. But it does drive the point home why Wikipedia editors should not perform statistical analysis like this. Read Sampling (statistics)#Population_definition. It is correct that the median (and other similar techniques) can be used to reduce the risk of sampling errors skewing the result. But that absolutely requires that the samples have been drawn from the same population. In a situation like this where samples are drawn from multiple more or less disjoint as well as overlapping populations it is an error to perform this type of analysis on them. Useerup (talk) 06:12, 15 November 2011 (UTC)[reply]
There is a science of statistics used to solve problems or optimize processes. This does not apply here. We simply are wanting to know what the central tendency of the data is and the median works well. Yes, if I were trying to optimize the formula for a batch of wafers in a fab process it would be crazy to use disjoint data, but again that doesn't apply here. Daniel.Cardenas (talk) 06:41, 15 November 2011 (UTC)[reply]
Do you mind telling us which of the |medians you are calculating, then? --Useerup (talk) 16:54, 15 November 2011 (UTC)[reply]
Yes. The problem is that the scope of article is fruits, not apples and grapes. And in the lack of adequately complete information comparing apples to grapes does the job. — Dmitrij D. Czarkoff (talk) 10:21, 15 November 2011 (UTC)[reply]
No. I don't understand why apples and oranges is a valid analogy here at all. We're not calculating a median of different things, we're calculating a median of similar things: OS percentage share data from various published sources. The methodology of each source is different, but that doesn't make their product a different category of thing. Fruit farms may grow apples in different ways, but they're still apples. --Harumphy (talk) 14:08, 15 November 2011 (UTC)[reply]
Yes, you are calculating median of different things. Usage share in US is one population. Usage share from German language sites is another statistical population. Each population has a different number of users, each population has its own bias. The median is not a valid calculation when the samples are drawn from different populations. This really is very basic statistics, and your refusal to "get it" is all the more reason why we don't trust editors to "summarize" using statistics. You are using it wrong. Please go read Median again --Useerup (talk) 16:52, 15 November 2011 (UTC)[reply]
The US and German-language populations are both subsets of the overall web user population, and the web sites monitored by each source all are subsets of whole web. AFAICS there's no overlap - i.e. no usage is monitored by more than one stats source. Therefore the effect of taking a median will be, if anything, to mitigate against bias. It doesn't make the median representative of the whole user population or of all web sites, but such representation has been neither claimed nor implied. So, I still don't get why "the median is not a valid calculation when the samples are drawn from different populations", when those populations are non-overlapping subsets of the same, single global population. You keep repeating this opinion, blandly asserting it's "very basic statistics". Well, years ago I did a basic statistics course and your assertion doesn't make sense to me. So stop patronising us and justify your assertion properly.--Harumphy (talk) 00:42, 16 November 2011 (UTC)[reply]

Just to clarify, in England, no kid does statistics of that calibre until they are (at a minimum) 16. Even then, some kids do not go on to study statistics (in the British curriculum, it is an optional course). If you want to study that at ages 17 and 18 (what we call "A Levels", you need a grade B in maths as a minimum. That's why I'm saying that the ordinary layperson would not (in some cases) have even learned about quartiles, and interquartile ranges. Although, they will have done some stuff about the mean.

The median is not a complex or uncommon measure. It is commonly used in the media in England. Whenever the "average income" is talked about, it most often a median figure that is used. The dictionary definition of "average", a common English word, gives median as an example. I don't think there's any issue with whether people know how to calculate it - many british school-leavers have difficulties with basic arithmetic, but that does not prevent wikipedia editors calculating ages. Dilaudid (talk) 11:15, 15 November 2011 (UTC)[reply]

Please see the Administrative Notes section, above. — TransporterMan (TALK) 15:29, 15 November 2011 (UTC)[reply]

Editor Useerup contacted me on my talk page about this issue (I'm a university professor of mathematical statistics). I see nothing terribly wrong in giving the median of a collection of numbers as a simple (easily understood) summary statistic of central location. 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. Half of the sources have smaller numbers, half have larger. Easy to understand, unpretentious. On the other hand, I prefer Useerup's own graphic representation of the whole table, which gives simultaneously an overall impression of general tendency as well as a picture of the large variation between different sources. (By the way, the relative sizes of the total areas of different colours in that graphic corresponds to another summary statistic: the mean.) Richard Gill (talk) 01:07, 16 November 2011 (UTC)[reply]

Professor Richard, Thanks so much for sharing your thoughts on this issue!   :)   Daniel.Cardenas (talk) 03:47, 16 November 2011 (UTC)[reply]
Professor Richard, I really appreciate you taking time to help us out. I know this must seem trivial not terribly challenging to you, but I'm sure your help will be highly welcomed here! So what you are saying is that the median calculated in this table cannot represent the median of any "population" but that it is useful for readers in it own right? That does it for me as far as the median goes. I would like the caveat about "not the population median" to be reflected in the notes, though. We still have the problem of some of the sources not breaking mobile OSes out and editors "correcting" the numbers with those from another source (synthesis). Is that a proper methodology, and if not how do we solve that? Useerup (talk) 04:18, 16 November 2011 (UTC)[reply]
Please just address me as Richard. Actually, the problem of how to summarize such a table of numbers *is* challenging. And doing it seriously certainly would require a lot of own research. An intelligent summary would at the same time be an evaluation and an interpretation. It seems to me that (a) probably many readers would appreciate a summary statistic (b) any reader who knows what the median is could look at the numbers and write down the medians themselves since, as someone said above, it is just a question of crossing off largest and smallest repeatedly till only one or two numbers remain (and in the latter case one conventionally averages the last two). This is not even arithmetic. Richard Gill (talk) 09:55, 16 November 2011 (UTC)[reply]
Useerup, it seems that the question of "correcting" mobile data is out of dispute — there seems too be consensus: extending source data is plain evil. I would suggest removing mobile-related data from this table and recalculating shares, though such sollution is pretty controversial on its own. Harumphy suggested to remove the data with no mobile results at all - which may seem reasonable, as there is curretly only one source without such split. — Dmitrij D. Czarkoff (talk) 17:08, 16 November 2011 (UTC)[reply]
Being a Professor myself (admittedly without tenure as of yet, but I'm young give me time :) ) I would disagree with Prof. Richard, at least in so far as the appropriateness on WP. The question still stands, why not the goemetric average, or the mean, etc. We are choosing this measure as the one that is useful for understanding this dataset, which is where I fell OR comes in. Thenub314 (talk) 05:29, 16 November 2011 (UTC)[reply]
The median can be written down just by looking at the numbers. No arithmetic required.

Here are nine numbers. The middle one is ... The next thing I'ld like to know is the smallest and the largest. At some point I'd like to talk to experts in the field to see if one can gain some understanding of the differences between the sources. Is any one clearly different from the rest? Are there clear groupings? I think it's a matter of taste where one has crossed the boundary between merely reporting what is out there, and adding to it oneself. So I'd rather ask other questions: is it useful? Ot is it dangerous? Since the nine numbers are all there and anyone could find the medians in their head, e.g. by guessing and checking and then adjusting the guess, it seems to me to be pretty harmless. Trouble is, there is not only a table but also a graphic (bar chart). Next thing will be that journalists will start quoting the wikipedia numbers. Difficult. Richard Gill (talk) 09:55, 16 November 2011 (UTC)[reply]

I thought perhaps the graphic might be okay whatever about the actual line in the table but I hadn't considered there might be some damned lazy journalists or others looking at it. That's tricky, quite a few of our images are generated by editors and need only be accepted as looking like what they represent but otherwise are caveat emptor - or more like cave freebe perhaps. Dmcq (talk) 11:24, 16 November 2011 (UTC)[reply]
If Prof. Richard is still reading this, how do you propose we deal with the fact that the sum of the medians does not come to 100% (it may even exceed). A pie chart would be the natural choice to illustrate "shares". But a pie chart assumes that there is a well-defined "whole" - represented by the entire disc.--Useerup (talk) 06:30, 16 November 2011 (UTC)[reply]
This is a bug which could be taken instead as a feature. There is no pretence to come up with some combined improved estimate of market shares. Richard Gill (talk) 09:55, 16 November 2011 (UTC)[reply]
The figures are medians of shares, not shares of medians, if you see what I mean, so they shouldn't be expected to add up to 100%. This is why we switched ages ago from a pie chart to a bar chart for the graphic. If we wanted the medians to add up to 100% then we'd have to multiply them by a weighting factor to make them do so, but that would be another level of synthesis that would no doubt invite yet another kerfuffle.--Harumphy (talk) 14:52, 16 November 2011 (UTC)[reply]
That certainly would be to compound the trouble. It is perfectly possible for one or more of such rescaled medians to fall outside the min-max range of the original figures! If the columns are treated independently and the results have to add up to 100% you'd have to use the mean to always avoid that. Not that I think any such average should be used, Useerup's graphic is a perfectly good illustration without all this messing around. Poll of polls stuff is the sort of harmless foolery popular newspapers and televisions go in for but I really don't think we should be doing it or pushing for it. Doing it properly is useful and very worthwhile for assessing the effects of drugs for instance but it requires great skill and can still go wrong. Here it doesn't matter much but if so why is it worth compromising Wikipedia to do it? Dmcq (talk) 22:58, 16 November 2011 (UTC)[reply]
I should have said that I do agree with Prof. Richard that Useerup's graphic works quite well. Thenub314 (talk) 07:15, 16 November 2011 (UTC)[reply]
And I don't like Useerup's graphic, as it is too busy and cluttered. Effectively, it requires studying it even more then the data in the table does, as it blurs the overall standings and makes comparing disconnected sections more difficult. — Dmitrij D. Czarkoff (talk) 17:08, 16 November 2011 (UTC)[reply]


On the usage share of web browsers we don't use a pie chart, we use a bar chart of the results. Daniel.Cardenas (talk) 07:57, 16 November 2011 (UTC)[reply]
I've read in multiple sources that pie charts are considered the bad means of data representation, as the human eye is not suited for judging shares in discs. I could give references if needed. — Dmitrij D. Czarkoff (talk) 17:08, 16 November 2011 (UTC)[reply]
The current graph gives WP:UNDUE weight to non-Windows OSes in that it compares individual Windows versions against all versions of the other operating systems. You would not have that problem with a pie chart as it would be trivial to recognize the combined slice of Windows versions against the other OSes, and it would still allow the reader to recognize the individual versions.--Useerup (talk) 17:31, 16 November 2011 (UTC)[reply]
Well, the Windows bars are all in similar colours, so the amount of these colour gives the impression of windows 80-some to 90-some percent domination from the first glance. Also there is a beneficial side effect here: most XP users didn't upgrade their systems as they reject Vista and 7. So, having a slight distinction between Windows versions is a good idea. At the same time, visually joining those Windows versions more would give the wrong impression that Microsoft OSs' usage share is limited to those three, meanwhile Windows 3.x, 9x and 2000 are in other column, where they constitute nearly 2% altogether (as per StatCounter). And even without these issues pie charts are really less informative. — Dmitrij D. Czarkoff (talk) 18:18, 16 November 2011 (UTC)[reply]
Nearly 2%? Are you seriously concerned about 2% "other" when we are discussing knowingly introducing errors in the 3% range for the sake of illustration? Really?--Useerup (talk) 19:13, 16 November 2011 (UTC)[reply]
As they reject Vista and 7? I assume you can read that from the statistics as well. Is that where we are going now? Useerup (talk) 19:21, 16 November 2011 (UTC)[reply]
After some more thinking I can say, that's it's really WP:DUE weight, as for many users (including me) the choice between XP, 7, Linux and OSX gives more sense then the choice between Windows, Linux and OSX. Eg., I will never install XP on a real hardware, while the choice among three other system would depend on multiple factors. — Dmitrij D. Czarkoff (talk) 18:23, 16 November 2011 (UTC)[reply]
Bogus argument. It matters more which Linux distribution you chose than Windows version. The software repositories vary enormously in size and to avoid dependency hell, software is being packaged not only per distribution but also per distribution version. Are you going to break out the Linux distros? Useerup (talk) 19:17, 16 November 2011 (UTC)[reply]
I don't agree with both of Your comments (especially the difference between Linux distributions as compared to Windows versions — as I maintain several Windows and Linux installations, I have absolutely opposite observations), but that's our POVs thing. The more objective thing is that pie chart are less informative then bar plots. — Dmitrij D. Czarkoff (talk) 19:26, 16 November 2011 (UTC)[reply]
Sure, that's why all of the sources break out Windows versions against all versions of other OSes. Oh wait, which of the sources does that? --Useerup (talk) 19:39, 16 November 2011 (UTC)[reply]
All counters (except for Chitika) split Windows. AT Internet also splits OSX per hardware platform (Windows' and Linux' alt.platform web clients shares are close to zero, so it's a plain hardware split), Clicky and Wikimedia split everything. Not sure about StatOwl, I don't have flash. So, Windows is split by 8/9, other platforms by 2/9. I'm in line with 6/9 sources. — Dmitrij D. Czarkoff (talk) 21:42, 16 November 2011 (UTC)[reply]

Compromise proposal

[edit]

In the interest of pursuing consensus I propose the following:

  1. Medians stays in table, but it is explained that they are medians of the sources; not medians of total usage shares
  2. Sources that do not break out mobile usage is removed; removing the need to "correct" their numbers.
  3. Graph changed to horizontal bars, shades of a color used to distinguish versions of an operating system, similar colors used to indicate "families" of operating system.

--Useerup (talk) 19:31, 16 November 2011 (UTC)[reply]

This is the wrong place to discuss this. This is a Mediation Cabal page to resolve one question: whether or not the median row stays in the table. Assuming that it has been settled that it does stay, further discussion should take place on the article's talk page.--Harumphy (talk) 20:49, 16 November 2011 (UTC)[reply]
It hasn't been settled. The topics are interconnected. --Useerup (talk) 21:06, 16 November 2011 (UTC)[reply]
They may be interconnected, but this page nevertheless exists to resolve a single question. The other questions should be resolved by editors seeking consensus in the usual place. Otherwise we'll end up with discussions scattered across different locations.--Harumphy (talk) 21:19, 16 November 2011 (UTC)[reply]
Sorry, but discussion led here. OK for me. Anyway, this issue still has to be settled, and here all the interested parties are now. Furthermore, we are not mediated here, — mediation is on hold — so You can think of this discussion as happening on the talk page. — Dmitrij D. Czarkoff (talk) 21:29, 16 November 2011 (UTC)[reply]

Support

[edit]

Oppose

[edit]
  • Oppose. The median row should stay, but the rest should be discussed the article's talk page. We haven't even tried to achieve consensus on the other related issues yet, so it's totally disruptive (again) for Useerup to try and bounce us into voting on something that hasn't yet been discussed on the article's talk page. (If someone else had proposed a vote he'd be the first to remind us that WP isn't a democracy.)--Harumphy (talk) 21:24, 16 November 2011 (UTC)[reply]
  • Oppose. Per User:Harumphy. We have discussed only median so far. Let's just decide whether we retain median line or no. I am not really opposing the other points, just that we discuss them here. 1exec1 (talk) 01:21, 17 November 2011 (UTC)[reply]
    I also agree with these comments. At least everyone agrees the median stays. Daniel.Cardenas (talk) 07:37, 18 November 2011 (UTC)[reply]
    No, I and others are still against the medians. I still consider them original research - especially because they involve multiple sources. But I am willing to compromise. You do not get to cherry pick from a compromise proposal. The idea is that everybody will have to make some concessions. If we cannot agree on this, at some point someone will have to put forward another proposal. Useerup (talk) 09:26, 18 November 2011 (UTC)[reply]
    Pot, kettle, black. You don't get to cherry pick a bunch of questions, put them into a package and tell the rest of us to take it or leave it.--Harumphy (talk) 09:49, 18 November 2011 (UTC)[reply]

Polling for consensus

[edit]

Please review the arguments and provide your input here: http://en.wikipedia.org/wiki/Talk:Usage_share_of_web_browsers#Consensus_on_median_in_summary   Thanks! Daniel.Cardenas (talk) 02:40, 9 December 2011 (UTC)[reply]