Talk:Usage share of web browsers: Difference between revisions
→Apples, oranges, ravens and office desks: r - how much simpler does it need to be for this to be understood? |
Undid revision 465653324 by VsevolodKrolikov (talk)self-revert |
||
Line 360: | Line 360: | ||
::::::::::There is also a different way of defining "usage". Is it a unique visit (NA, W3counter), or is it every page hit? This matters because it views different browsing habits differently. |
::::::::::There is also a different way of defining "usage". Is it a unique visit (NA, W3counter), or is it every page hit? This matters because it views different browsing habits differently. |
||
::::::::::I'll stress again - real world statistics is not like stats at school. When you ask a question like "which browser is used the most", there is not one clear way of defining that question quantitatively. Operationalising concepts for measurement is not a straighforward hand-wavy procedure.[[User:VsevolodKrolikov|VsevolodKrolikov]] ([[User talk:VsevolodKrolikov|talk]]) 15:29, 13 December 2011 (UTC) |
::::::::::I'll stress again - real world statistics is not like stats at school. When you ask a question like "which browser is used the most", there is not one clear way of defining that question quantitatively. Operationalising concepts for measurement is not a straighforward hand-wavy procedure.[[User:VsevolodKrolikov|VsevolodKrolikov]] ([[User talk:VsevolodKrolikov|talk]]) 15:29, 13 December 2011 (UTC) |
||
:::::::::::Well, we must try not to do ''any'' statistics (real world or school as you say) here, because of OR. We must try not to weight any sources based on our own judgement because of OR. The only judgement available to us is what the RS say. As for the populations, I'll rephrase: in the article we analyze ''global population'' of browser share. Each of the stats engine represents a biased ''subpopulation'' of the global population. Can you provide any statistically sound method how we can select the biased subpopulation to present? How will this be less OR than to present all subpopulations and let the reader decide? [[User:1exec1|1exec1]] ([[User talk:1exec1|talk]]) 15:49, 13 December 2011 (UTC) |
|||
::::::::::::We weight sources according to their prevalence in RS. This is how Wikipedia works. It's quite simple.[[User:VsevolodKrolikov|VsevolodKrolikov]] ([[User talk:VsevolodKrolikov|talk]]) 15:53, 13 December 2011 (UTC) |
Revision as of 15:54, 13 December 2011
This is the talk page for discussing improvements to the Usage share of web browsers article. This is not a forum for general discussion of the article's subject. |
Article policies
|
Find sources: Google (books · news · scholar · free images · WP refs) · FENS · JSTOR · TWL |
Archives: 1, 2, 3, 4, 5, 6, 7Auto-archiving period: 90 days ![]() |
This is the talk page for discussing improvements to the Usage share of web browsers article. This is not a forum for general discussion of the article's subject. |
Article policies
|
Find sources: Google (books · news · scholar · free images · WP refs) · FENS · JSTOR · TWL |
Archives: 1, 2, 3, 4, 5, 6, 7Auto-archiving period: 90 days ![]() |
![]() | This article is of interest to the following WikiProjects: | ||||||||||||||||||||||||||||||||||||||||||
Please add the quality rating to the {{WikiProject banner shell}} template instead of this project banner. See WP:PIQA for details.
Please add the quality rating to the {{WikiProject banner shell}} template instead of this project banner. See WP:PIQA for details.
Please add the quality rating to the {{WikiProject banner shell}} template instead of this project banner. See WP:PIQA for details.
|
Statcounter and ipod
On Sun, Feb 20, 2011 at 5:13 AM, StatCounter Global Stats <globalstats@...com> wrote:
Hi Daniel,
6.3% refers to the iPod Touch and *not* to the iPad.
The iPad does *not* meet our definition of a "mobile device" and is *not* therefore included in our Mobile Browser stats at all.
The iPod and iPhone *do* meet our definition and are both included in our Mobile Browser stats.
StatCounter Global Stats
http://gs.statcounter.com/ http://twitter.com/statcountergs
-----Original Message----- From: "Daniel Cardenas" <daniel...@gmail.com> Sent: Saturday, 19 February, 2011 04:32 To: "StatCounter Global Stats" <globalstats@....com> Subject: Ipod touch 6.3% of usage? Perhaps it is more likely ipad?
http://gs.statcounter.com/#mobile_browser-ww-monthly-201102-201102-bar
Hi,
Ipod touch 6.3% of usage? Perhaps it is more likely ipad?
Thanks, Daniel
Jan 2011 Bar chart does not match the key
a little consistency please :) key says IE ~43% char is 46%
Opera market share error
I think it should be noted that Opera 11.50 (June 28, 2011) had 34 million downloads within 8 days ( http://www.linuxnov.com/opera-12-pre-alpha-wahoo-released-changelog/ ), whereas Firefox 4 (March 22, 2011) had around 40 million downloads total (despite being regarded by many as the biggest update in Firefox's history. — Preceding unsigned comment added by 174.95.13.144 (talk) 12:31, 26 August 2011 (UTC)
- This article is about how often a web browser is used to browse the web. Not how often it is downloaded. Daniel.Cardenas (talk) 23:58, 26 August 2011 (UTC)
- Would it not follow that browsers that are downloaded often are more likely to be used (especially since a major build of FF that would have had more people downloading it than a normal build as people wanted to check it out was barely downloaded more than a minor change to Opera)? It is suspected that Opera's usage is grossly under reported as it often masks as IE. — Preceding unsigned comment added by 174.91.96.178 (talk) 04:37, 29 August 2011 (UTC)
- Would be good if you had a reference for that. Daniel.Cardenas (talk) 04:19, 3 September 2011 (UTC)
- Not sure, but this might be what he is talking about http://www.opera.com/support/kb/view/843/ Personally, mine identifies to all sites as IE unless I specifically state otherwise. Charwinger21 (talk) 05:50, 12 September 2011 (UTC)
- Would be good if you had a reference for that. Daniel.Cardenas (talk) 04:19, 3 September 2011 (UTC)
- Would it not follow that browsers that are downloaded often are more likely to be used (especially since a major build of FF that would have had more people downloading it than a normal build as people wanted to check it out was barely downloaded more than a minor change to Opera)? It is suspected that Opera's usage is grossly under reported as it often masks as IE. — Preceding unsigned comment added by 174.91.96.178 (talk) 04:37, 29 August 2011 (UTC)
incorrect data on data statics for site w3counter
just was comparing the growth of different browsers across different sites and noticed for some reason some of the overall stats seemed to be different then what is on the websites referenced. not sure if there was a reason or not but the w3counter site shows a higher usage for chrome and firefox then is indicated here and a lot lower usage of Internet explorer. im not sure what other stats may be wrong but Im certain the stats on there are wrong. — Preceding unsigned comment added by 66.75.85.118 (talk) 11:28, 29 August 2011 (UTC)
- The W3Counter table on wikipedia and the numbers on W3Counter's website are identical. 108.16.219.54 (talk) 22:06, 5 October 2011 (UTC)Dan Grossman
Wikimedia table complexity
I have just updated the Wikimedia table, adding July's figures. I found it a very intricate and complex task, involving mental arithmetic to get the 'Total's for Opera and Safari. I was frustrated by the time I got near the end and found the column labelled 'Mobile, Other'. Other than what? By opening a spreadsheet, a copy of the article unedited, and a copy of the Wikimedia stats for June, I was able to reverse-engineer that it is the Mobile figure (given in our next column, with a spurious final zero added in the second decimal place) minus the mobile figures for Opera and Safari as well as the figure for Android. I followed the pattern dutifully.
When I got to adding the latest Wikimedia figures to the summary table at the top, I understood the relevance of the mental arithmetic regarding Opera and Safari - this is necessary to get totals for the summary - but I have not found any use for the 'Mobile, Other' figure. In order to encourage others to make the effort and help keep this page up to date, I propose removing the 'Mobile, Other' column from the Wikimadia table.
Finally, I wonder why we now have four summary tables in the Summary table section. Would three be enough, with a rolling deletion of the fourth when the first is complete? --Nigelj (talk) 17:46, 31 August 2011 (UTC)
- If you'd rather not enter the "other mobile", then you could leave it blank. I find the data useful and I suspect readers do too. Thanks for entering the rest of the info. Daniel.Cardenas (talk) 19:03, 31 August 2011 (UTC)
- Ok, agree its best to delete "other mobile" now. Trying to minimize the work. Daniel.Cardenas (talk) 18:16, 8 October 2011 (UTC)
"multiply the desktop percent from "mobile vs. desktop" by each desktop browser"
Apparently, the note under the StatCounter table is very important to understand our in-house maths. However, I do not understand the phrase above: I don't mean to be unnecessarily pedantic, but I don't know how to multiply a "percent" by a "desktop browser". Or why. --Nigelj (talk) 08:26, 9 September 2011 (UTC)
- I didn't know how to explain it, but this is for statcounter and Netapp. Lets say that desktop and Mobile browsers account for 50% of the browser share each. Then desktop browser share is reported as 30% for red, 25% for yellow, and 40% for green. Mobile browser share is reported as 33% for orange, blue, and purple. It is somewhat wrong to put these on the same row and say they add up to 200%. So what you have to do is multiply by the desktop or mobile share and then put them in the same row. Wikipedia reports correct percentage so no manipulation needed there. What do you think? Thanks for addressing this issue. Daniel.Cardenas (talk) 16:52, 9 September 2011 (UTC)
- To use the actual numbers from August:
- Desktop is 100%-7.12% = 92.88%. So multiply the desktop numbers by 92.88%
- Mobile number is 7.12%. Multiply the mobile percent by 7.12%.
- Daniel.Cardenas (talk) 23:03, 9 September 2011 (UTC)
- To use the actual numbers from August:
I added desktop and mobile for statcounter. The numbers add up to:
38.91% + 25.53% + 21.51% + 6.19% + 3.09% + 1.4% + 2.8% = 99.43%
The small percentage that is missing is because we don't report desktop other.
Daniel.Cardenas (talk) 18:21, 11 September 2011 (UTC)
- I could use some feedback for motivation. Do you guys like it? Do you hate it? Should I continue, stop, change direction? Thanks, Daniel.Cardenas (talk) 03:08, 24 September 2011 (UTC)
- OK, I understand what you're doing now. I think we could improve the explanation in the note a little more. How about changing
- To get the desktop number multiply what Statcounter reports by desktop% as reported in desktop vs. mobile. To get the mobile number multiply Statcounter browser percentage by mobile %
- into
- For consistency, each StatCounter desktop browser share has been reduced by multiplying it by the current overall desktop share. Similarly, mobile browser shares have each been multiplied by the overall mobile percentage.[1].
- This assumes we can use a formatted 'ref' within a 'note'. I haven't tested that, but if it fails we can use the existing 'bare link' format on the italicised terms (in which case they probably don't need to be italicised)--Nigelj (talk) 11:28, 24 September 2011 (UTC)
- OK, I understand what you're doing now. I think we could improve the explanation in the note a little more. How about changing
- I looked at new tables for August (for statcounter and netapp), and they look visually confusing. IMO clamping Desktop/Mobile as a subheader makes it harder to read.Wikiolap (talk) 04:36, 25 September 2011 (UTC)
- Are you suggesting keeping desktop and mobile separate so that table is easier to read? Daniel.Cardenas (talk) 05:26, 25 September 2011 (UTC)
- Yes, I think keeping desktop and mobile stats separate will make it easier to comprehend. Also, dynamics of desktop and mobile are very different - the dominant player in desktop (IE) is nowhere found in mobile, and vice versa (Safari).Wikiolap (talk) 02:15, 26 September 2011 (UTC)
- Perhaps the right thing to do is have different tables. One with the numbers combined, and one set with them apart. I'll create the combined one, which to me represents the true share. Other's can create the separated tables. Daniel.Cardenas (talk) 14:34, 8 October 2011 (UTC)
- Yes, I think keeping desktop and mobile stats separate will make it easier to comprehend. Also, dynamics of desktop and mobile are very different - the dominant player in desktop (IE) is nowhere found in mobile, and vice versa (Safari).Wikiolap (talk) 02:15, 26 September 2011 (UTC)
- Are you suggesting keeping desktop and mobile separate so that table is easier to read? Daniel.Cardenas (talk) 05:26, 25 September 2011 (UTC)
- I looked at new tables for August (for statcounter and netapp), and they look visually confusing. IMO clamping Desktop/Mobile as a subheader makes it harder to read.Wikiolap (talk) 04:36, 25 September 2011 (UTC)
I'm thinking the "desktop" statcounter table shouldn't be normalized for mobile. http://en.wikipedia.org/wiki/Usage_share_of_web_browsers#StatCounter_.28July_2008_to_present.29 Do you agree? Daniel.Cardenas (talk) 06:04, 17 October 2011 (UTC)
- I reverted those changes. Daniel.Cardenas (talk) 04:17, 6 November 2011 (UTC)
Broken StatCounter tables
The StatCounter and StatCounter Mobile tables appear to be completely broken (tested in IE, Chrome and FF, so it's not a browser rendering issue)? The tables are unreadable as-is, I presume this isn't deliberate? Psdie (talk) 14:21, 7 October 2011 (UTC)
- Thanks for pointing that out. I removed the sortable attribute which seems to have resolved the issue. What do you think about the discussion above? Daniel.Cardenas (talk) 17:15, 7 October 2011 (UTC)
September calculations
FYI: here is the spreadsheet I used for september calcs: https://docs.google.com/spreadsheet/ccc?key=0AoB_APc2D9MsdDdad3Z5Z3BwenpsSzN4NFdtTU9XUlE&hl=en_US# . Daniel.Cardenas (talk) 18:31, 8 October 2011 (UTC)
Message from w3counter about "other"
Subject: Why doesn't w3counter's global stats report add up to 100%?
On Wed, Oct 5, 2011 at 3:03 PM, Dan Grossman <dan@w3counter...> wrote:
There is no other information available. Sure you can post my response.
On 10/5/2011 6:02 PM, Daniel Cardenas wrote:
Are the other browsers info available publicly? Can I post your response at the discussion list of wikipedia usage share of web browsers?
On 10/5/11, Dan Grossman<dan@w3counter...> wrote:
Because there are more than 5 web browsers. There are several hundred different families each with less than 1% market share.
Kind regards, Dan Grossman w3counter.com
On 10/5/2011 5:14 PM, Daniel Cardenas wrote:
August: 35.4% 26.7% 20.2% 6.2% 2.3% 9.2% missing
Thanks, Daniel
160 million users for opera vs +200 million for Chrome
I find it interesting that Opera claims 160 million users while google claims more than 200 million users for chrome. Although the usage stats have chrome at 5 times more usage. Daniel.Cardenas (talk) 17:57, 17 October 2011 (UTC)
- Opera's users are mostly mobile users ("the vast majority using Opera Mini"). Mobile users don't go to as many sites as desktop users, so the usage share for mobile browsers is not proportional to the number of users. Mozilla claims over 400 million users, so I would say Chrome has about 400 million also. -- Schapel (talk) 21:26, 17 October 2011 (UTC)
Web should be capitalized?
Usage share of Web browsers -- with Web capitalized because it's a noun? In the phrase web browser, web is an adjective. What kind of browser? A web browser. -- Schapel (talk) 23:19, 21 October 2011 (UTC)
Undercounting privacy-conscious users
The following is original research, but surely could be sourced by someone with the time to do it?
Many users block data-miners and stat-counters. One common method is by blocking the JavaScript often used by such. So these users don't get counted at all.
Various browsers have varied capability in JS-blocking. Probably the most *selective* (enabling the site you're on, while blocking third-parties such as stat counters, is NoScript, supported by the Mozilla-based browsers Firefox and SeaMonkey.
So it seems intuitively that Firefox, especially, is going to be under-counted by approximately the number of NoScript users. NoScript runs surrogate script that returns no data to these counters, but satisfies a page's requirement to run these counters. (SeaMonkey's market share is much lower, and probably wouldn't affect the overall picture.)
IE and Chrome won't block these, especially since Chrome's owner, Google, makes a living by selling (mostly-targeted) advertising, and also by selling data.
Add all this up, and what one finds in many of the article's charts are the most popular browsers used by *non-privacy-conscious* users. The steady decline of Firefox in the charts might be partly or substantially due to the increase in NoScript users.
I wish I had the time to do all the proper sourcing and citations for this. Just browsed to the article, and realized this issue. I hope someone who does have the time will raise this issue in the article. It's a very plausible explanation for significant under-counting of Firefox and other Mozilla browsers. Thanks. Unimaginative Username (talk) 11:41, 17 November 2011 (UTC)
I feel like removing the wp:or tag because of this comment. Let me know what you think. Daniel.Cardenas (talk) 04:14, 18 November 2011 (UTC)
No, the dispute is ongoing. The comment does indeed support that the median can be considered "simple" - but that's a professional opinion not an opinion on Wikipedia policies. Useerup (talk) 06:03, 18 November 2011 (UTC)
- Please see Wikipedia:Mediation_Cabal/Cases/13_November_2011/Usage_share_of_operating_systems#Removing_the_median_until_dispute_has_been_resolved. The median must be deleted until dispute is resolved. Do *not* re-add the median Useerup (talk) 06:34, 25 November 2011 (UTC)
Summary table proposal: Only use stat sites that break out mobile usage
There is a proposal on cabal to only show the stat sites on the summary table that break out mobile browsers. Let us know what you think of that. Thanks, Daniel.Cardenas (talk) 04:55, 19 November 2011 (UTC)
Pie Chart
The pie chart is not nice as 3d ones are especially hard to read. there was a discussion of this long ago which is why it ended up as a bar chart. Can we switch back for readabilities sake - Thanks — Preceding unsigned comment added by 87.74.74.94 (talk) 17:17, 2 December 2011 (UTC)
Limit summary table to those stat sites that break out mobile
There was a proposal here to only list stats in the summary that break out mobile. I plan to remove the sites that don't have mobile unless there is significant concerns and or objections raised here. Thanks, Daniel.Cardenas (talk) 00:43, 4 December 2011 (UTC)
Consensus on median in summary
wp:No original research states that a wp:routine calculation should have wp:consensus. Would like to poll and discuss to see if there is consensus. Also note that wikipedia doesn't wp:vote on issues.
Issue | Arguments For | Arguments Against |
---|---|---|
Simple and routine |
The following was written by Richard D. Gill on the Cabal page: |
Each of these issues will be discussed in separate rows:
|
1: Not simple because the median of a set of percentages may not add up to 100%. | It is still the median value of the reported share. That it may not add to 100% is not an overriding concern. It is not possible to get a number outside of the reported usage. |
A percentage means out of 100. With this method if the results are rescaled to 100 it is entirely possible to get a figure outside the range of its sources. |
2: Numbers not comparable because they are derived by different methods and can mean different things. | Basically getting median of numbers of apples and grapes and calling it the median number of fruit. | |
3: Wikipedia is not in the business of approximations. The median is an approximation of usage share since it is the central tendency of the reported sources. |
Policy WP:OR basically says we should not make up our own results.
The median is only an approximation as far as the sources are representative. The sources do not sample the same population (and they use different methodology) and any attempt to present the median as the approximation or central tendency of "the population" is a misrepresentation. There is no population. | |
4: Misquoted as actual usage share other WP articles. | Fix the misquotes and add notes to median. | Plus any newspaper that cares to quote us. |
5. Wikipedia is an encyclopaedia, the content policy of WP:NOT#Content says 'In any encyclopaedia, information cannot be included solely for being true or useful." | The examples for WP:NOT#Content do not apply here. WP:LEAD does. | Richard Gill's idea about providing the figures because they might be easy and useful though they have little intrinsic meaning is not in line with the basic aim of providing verifiable information. |
6. The graphics based on these do not show the variability and imply spurious accuracy. | Useerup should add his graphic to the article. | Useerup showed a graphic with all the figures in and it showed the variability between the different surveys much better. This sort of thing shows how the results were compiled by different methods and show different thing. |
7. The median is needed because people are too stupid and we need to spoon feed them. Anyway that's my paraphrase of the main argument I've seen for keeping the median, feel free to improve. | Absurd issue / comment. WP:NOTSTATSBOOK |
Feel free to edit the table above with pro and con arguments and or enter your comments below. Daniel.Cardenas (talk) 01:26, 4 December 2011 (UTC)
- Added issues 5,6,7 Dmcq (talk) 13:02, 4 December 2011 (UTC)
Comments
I see both proposed views (mean and median) as providing useful information to users, why can we not have both a "median" and "mean" row at the bottom (possibly with a slightly off gray bgcolor to separate them from readings) and give the reader complete coverage, surely this can't complicate things too much nor is it hard for the reader to comprehend. 72.241.135.189 (talk) 03:44, 4 December 2011 (UTC)
- We're not in the business of making up things even if we think it might be useful, see the very first sentence of WP:NOT#Content. Dmcq (talk) 12:51, 4 December 2011 (UTC)
- Wikipedia editors are required to summarize content, see wp:lead, required to paraphrase, etc... The median is a way of summarizing the content. Daniel.Cardenas (talk) 18:32, 4 December 2011 (UTC)
- We're not in the business of making up things even if we think it might be useful, see the very first sentence of WP:NOT#Content. Dmcq (talk) 12:51, 4 December 2011 (UTC)
- The article doesn't summarize any of this in the lead anyway so I don't see why you quote that, not that a lead should use invented information any more tan the body of an article should. Dmcq (talk) 09:48, 7 December 2011 (UTC)
- Summarize by selection not by synthesis. Median is not summarizing; it is a calculation and creates a new viewpoint not supported by any single source. That is the very definition of original research. You are creating a logical fallacy when you try to re-label median as "summarizing". There is a reason for WP:CALC Useerup (talk) 19:02, 7 December 2011 (UTC)
Support Median
- Support - Its a very simple and straitforward calc. Daniel.Cardenas (talk) 18:34, 4 December 2011 (UTC)
- Support. The calculation is intended as a summary, to free the readers from the need to do some calculation themselves in order to understand what the sources say. It is like a kind of illustration just in text form, so WP:OR doesn't apply. 1exec1 (talk) 22:28, 4 December 2011 (UTC)
- Support. --188.10.90.52 (talk) 14:38, 6 December 2011 (UTC)
- Support. What was the outcome to the OS usage share debate on the same issue? I stopped following it once it started to be dispersed on several pages. Jdm64 (talk) 20:08, 7 December 2011 (UTC)
- I believe the two issues are tied together. Thanks, Daniel.Cardenas (talk) 22:24, 7 December 2011 (UTC)
- Have you alerted the original participants that you are performing a poll?--Useerup (talk) 06:01, 8 December 2011 (UTC)
- I believe the two issues are tied together. Thanks, Daniel.Cardenas (talk) 22:24, 7 December 2011 (UTC)
- Support. This is a trivial 'calculation' that provides a useful summary for both the web browser figures here, and the OS figures elsewhere. The row of medians is not a data row in its own right (it is a collection of the medians of each data column); it does not add up to 100%, and so care must be taken not to treat it as if it were another data row and further summarise the summaries, e.g. they should not be displayed in a pie chart. It could be displayed as a bar chart, but it is probably safer to display charts of actual data from individual sources, labelled as such. Glancing across the medians may help editors decide which of the sources provides or is closest to key medians, for promotion into the lede (as a chart, complete with identifying caption) on any given month, if a lede image is desirable. --Nigelj (talk) 20:29, 8 December 2011 (UTC)
- See WP:USEFUL in arguments to avoid in deletion discussions. Plus even safer is not to produce badly flawed figures in the first place. Dmcq (talk) 08:45, 9 December 2011 (UTC)
- Support. The teaching of medians and means is a standard part of the British maths curriculum (I cannot comment about other countries). Provided that the figures are collected in a reliable manner, I see no harm in publishing the median value - in many cases it is more valuable than the mean value, while in others publishing the mean and median alongside each other gives the astute reader a feel for the skewness of the distribution. Martinvl (talk) 09:09, 9 December 2011 (UTC)
- Support. An incredibly basic and standard bit of mathematics. It is in no way original research. Bilrand (talk) 23:49, 10 December 2011 (UTC)
- Support. Agree this is pretty basic. I don't understand why this is such a controversial topic. scottjduffy (talk) 11:00, 12 December 2011 (UTC)
- Support. Although I understand the "fear" that this don't show a global stat, I don't believe that it harms Wikipedia. The fact is: most pages don't cite Wikipedia - they cite netapplications and thus the whole discussion is a bit out of focus. mabdul 17:54, 12 December 2011 (UTC)
- If most pages cite Net Applications, we should cite Net Applications. (It's a toss-up between Net Applications and Statcounter). NPOV asks us to reflect majority usage. Clicky, w3counter and Wikimedia are rarely (with Wikimedia possibly never) used counts. VsevolodKrolikov (talk) 17:58, 12 December 2011 (UTC)
- Can someone show actual evidence of the fact that NetApplications is cited most widely? Even if this proves true, as far as I can remember, there are at least several sufficiently reliable pages citing each of the sources we currently use. So, per WP:DUE we should at least mention them all. 1exec1 (talk) 19:06, 12 December 2011 (UTC)
- It's disappointing that you didn't try to look yourself first. Google News results for measure plus "browser":
- Net Applications gets 3540
- statcounter gets 1670
- and the rest get bugger all.
- I haven't heard from anyone supporting the median why we should treat each of these measures as equally valid given that some are clearly more accepted than others, nor anyone tackle the apples and oranges problem. The calculation itself is very simple - but that's really not the point. One doesn't discover the median height of a human by taking the average height of the US, the average height of the US and Europe, the average height of the US and Europe when jumping four times over, and the average height in the world dependent on CIA sources and your own range of busines clients, and seeing which height is in the middle of that.VsevolodKrolikov (talk) 19:31, 12 December 2011 (UTC)
- I haven't seen anyone suggesting that each of the measures is equally valid. In my comment about the wisdom of the crowd, I explain why it actually can make sense to take the median of measurements from different populations and consider that an estimate of the median of an actual value. Can you provide a source and detailed explanation of why your example won't result in the median height of a human? All I see is people claiming it without evidence, whereas I provide examples of where taking the median of lots of poor guesses works very well in practice. Taking the median of fairly accurate measurements should work even better. -- Schapel (talk) 20:16, 12 December 2011 (UTC)
- By following these links I get the opposite: 57 for "Net applications" and 289 for "Statcounter". Maybe you receive highly personalized results or have additional settings enabled (I don't supply cookies to google)? Could someone else recheck the links? 1exec1 (talk) 20:43, 12 December 2011 (UTC)
- For Schapel - do you really think that taking a median from an unjustified set of measures, most of which are heavily weighted towards the two tallest continents in the world we get either a good picture of global human height or one that is commonly cited? Sounds like bad stats to me. Furthermore, would you really want to use how high one can jump as a good proxy for relative height? Over half of all Americans are overweight, with a third obese - this would screw up the data. Again, it would be bad stats. One has to look not at generalised descriptions of data (such as "browser usage" and "around the world") but how such concepts are operationalised.
- As for wanting a link - the only example I can think of someone trying to aggregate poor data in order to improve it is the "race realist" J. Philippe Rushton, who uses such an approach as part of his attempts to show that black people are stupid and have too many kids with no thought of tomorrow. You'll find criticism of this approach at Race,_Evolution,_and_Behavior#Validity_of_the_data_and_the_methodology_of_aggregating_the_data. I certainly don't mean to associate you with his views; on the contrary, once you look at data which you're very probably rather less attached to, perhaps you'll see the problem - and also how serious academics have savaged the approach.VsevolodKrolikov (talk) 01:54, 13 December 2011 (UTC)
- For 1exec1 - my apologies - in trying to cut away the crap from the web address the google search was producing, I took out the code for "archives". Set the google news date search to "archives" (left hand side of the page) and you'll get the full results. I'm on a different network from yesterday and I get the same results. People here seriously really need to read WP:NPOV. The first line says "Editing from a neutral point of view (NPOV) means representing fairly, proportionately, and as far as possible without bias, all significant views that have been published by reliable sources." This means we should not give undue weight to browser counts that few or none use. The Wikimedia stats are interesting given that this is Wikipedia, but they should not be part of any representation of global stats, as very probably no independent reliable source refers to them as a measure of general web usage. Only Statcounter and Net Applications turn up on google news. A Google trends measure reveals the same over the whole of google's internet reach. NPOV is not a trivial consideration - it's one of the core principles of the project.VsevolodKrolikov (talk) 02:03, 13 December 2011 (UTC)
- No problem, later I found that archives produce your data. However, I'm not convinced that WP:RS mostly refer to NetApplication for the following reasons:
- GoogleNews produce vastly different results for different timespans. For example 57 for "Net applications" and 289 for "Statcounter" for the last month would imply that "Statcounter" is a better source for us, especially since we use most recent figures in the article.
- GoogleNews not necessarily represents reliable sources well (it only considers news sources, thus a bias is apparent). Checking regular Google search again produces very different results: 20.1 millions for "Statcounter" statistics, 4.2 millions for "Net applications" statistics, 2.9 millions for "W3counter" statistics, 51.9 millions for "wikimedia" statistics, 0.4 millions for "statowl" statistics, 66.9 millions for "clicky" statistics (I also get similar ratios for browser statistics instead of statistics). I have no reasons to believe that across such large sample sizes, WP:RS shouldn't be distributed in similar ratios. Thus I think that it would be a bigger WP:OR to select one method over another in order to select one source over the others, than by showing several sources, providing all sources with notices what they measure (to address WP:NPOV) and a median for illustrative purposes. 1exec1 (talk) 14:09, 13 December 2011 (UTC)
- First of all, I'm glad you've grasped the point of NPOV - reflecting prevalence in RS, rather than giving equal space to minority and obscure views. We're almost on the same page.
- As for the "bias" in google news: you may not be aware of this, but the internet in general is not considered a reliable means of assessing prevalence in RS on wikipedia. Google News, scholar and books do all have a "bias": towards featuring reliable, oversighted sources over promotional, fanboi and fringe sites, unmoderated blogs and discussion sites and so on. That's why on Wikipedia we actively prefer using those google services rather than the general one to establish things like notability, article titles and due weight.
- Your search terms aren't particularly efficient. An alarm bell really should have rung when "Wikimedia" statistics came out top - I trust you are open to hearing alarm bells, and not simply pretending to look at the evidence. Try the same searches with "browser share" (in quotes) and you'll see Statcounter and Net Applications come out far and away the most used compared to the others. I'm open to your suggestions of other, more efficient search terms for cutting out irrelevant hits. As I commented below, perhaps it is better that we have both Net Applications and Stat Counter in the lede, as it illustrates the results from different approaches to the usage share question. It doesn't mislead the reader into thinking there's a single approach.VsevolodKrolikov (talk) 14:50, 13 December 2011 (UTC)
- That is not an issue IMO. The issue that selecting one method is OR. Can you provide any evidence that selecting one of "XX statistics", "XX browser statistics", "XX "browser share"" is not OR? Why not "XX "browser market share""? 1exec1 (talk) 14:59, 13 December 2011 (UTC)
- Could you instead cite which part of WP:OR it would contravene? It's not exactly logical to ask someone to disprove something that you haven't even taken the time to describe. I'm surprised that you don't think search terms are an issue. I thought you knew about the internet - my mistake.VsevolodKrolikov (talk) 15:45, 13 December 2011 (UTC)
- That is not an issue IMO. The issue that selecting one method is OR. Can you provide any evidence that selecting one of "XX statistics", "XX browser statistics", "XX "browser share"" is not OR? Why not "XX "browser market share""? 1exec1 (talk) 14:59, 13 December 2011 (UTC)
- No problem, later I found that archives produce your data. However, I'm not convinced that WP:RS mostly refer to NetApplication for the following reasons:
- It's disappointing that you didn't try to look yourself first. Google News results for measure plus "browser":
- Can someone show actual evidence of the fact that NetApplications is cited most widely? Even if this proves true, as far as I can remember, there are at least several sufficiently reliable pages citing each of the sources we currently use. So, per WP:DUE we should at least mention them all. 1exec1 (talk) 19:06, 12 December 2011 (UTC)
- If most pages cite Net Applications, we should cite Net Applications. (It's a toss-up between Net Applications and Statcounter). NPOV asks us to reflect majority usage. Clicky, w3counter and Wikimedia are rarely (with Wikimedia possibly never) used counts. VsevolodKrolikov (talk) 17:58, 12 December 2011 (UTC)
Oppose Median
- Oppose. To me this is an OR issue. First the median is one of an arbitrary number of possible measures, and the first tell tale sign we are doing research is that we are making a choice about which measure is the most appropriate. Similarly, the data sets are not comparable, they are collected in different ways and measure different things. (For example, Some use CIA adjusted data while other entries use raw data, etc). Thus it doesn't make a lot of sense to compare them in this way. Thenub314 (talk) 05:47, 4 December 2011 (UTC)
- Oppose We should not be making up our own results just reporting what others have said. The median does not add anything meaningful and there's better alternatives for the graphic. We should leave it to people out there to do something if it is useful and then report it rather than make up our own 'useful' figures. Dmcq (talk) 12:58, 4 December 2011 (UTC)
- Oppose. Median is WP:OR both under WP:CALC and under WP:SYN (synthesis over multiple sources). Median numbers are misleading (sum of medians will never add up to 100% and thus cannot be usefully thought of as "shares" of anything). Median numbers are being quoted by other articles as the usage share - which they are not. Medians conveys the idea that a central tendency is relevant, even though the sources are picked by WP editors and do not represent global usage share or any other useful "tendency". --Useerup (talk) 21:04, 4 December 2011 (UTC)
- Oppose. (here from NOR board) While calculating a median would - in my view - be acceptable under WP:CALC, this is a misapplication of its use that is WP:OR and generally very problematic. The OR comes in our selection of measures from which we draw the median. Why these particular measures? Why do we - by using a median - give them equal value? Are they all seen in RS as equally valid and reliable, or are some preferred to others? (How many RS use Wikimedia stats, for example, compared to Net Applications or Stat Counter?) A general problem also comes in these not measuring the same populations, nor the same things. For example, Wikimedia is not the whole internet, nor is it a random sample of the internet, and there is no reason to suppose that users to Wikimedia organisations are representative of the internet. Two sources measure unique hits, others measure visits. One measure is adjusted for estimated country internet use, while others are not. My strong preference for the lede would be to choose the most commonly used source, with a footnote that other measures are also used. Taking the median makes no sense to me here.VsevolodKrolikov (talk) 10:10, 9 December 2011 (UTC)
- So to you it would make more sense to choose one of the stat sites that doesn't random sample and manipulates the numbers? Daniel.Cardenas (talk) 00:24, 11 December 2011 (UTC)
- Er, no. My criterion was quite clear: "the most commonly used source". This meets NPOV. But it's good to see you now understand that these sites are not in practice measuring the same thing. In that light, perhaps you might review your support for this misapplication of statistics? VsevolodKrolikov (talk) 01:20, 11 December 2011 (UTC)
- You fail to understand that none of the sites practice random sampling. Perhaps you might review your support for this misapplication of statistics? Daniel.Cardenas (talk) 01:58, 11 December 2011 (UTC)
- The point about "random sample" is not that the stats should all be random samples. You're taking the Wikimedia stats as representative of general Internet use - a very limited sample of Internet traffic - but based on what principle? In what way are you establishing representativeness? (other stats do this by taking a broad inclusive range of high volume sites) It's clearly a biased sample. You call my support for using one source only a "misapplication of statistics". If you think each of the most commonly used major stats sources misapplies statistics, bring that to the table, rather than take an average of sources you apparently consider to be seriously flawed. VsevolodKrolikov (talk) 02:37, 11 December 2011 (UTC)
- "broad inclusive range of high volume sites" - hmmm, yeah right. Where did you pull that from? All the stats sites are bias. Daniel.Cardenas (talk) 03:37, 11 December 2011 (UTC)
- All the stats sites are bias. Exactly! And the median - unlike what someone here naively assumes - will not erase that bias. For the median to express anything useful about a population (internet usage), the samples must be drawn from the same population (internet usage). In that case the median can be used to erase sampling error. But none of the sources sample the same population and they use very different methodologies at that. The only thing a median could express is that it is the median of the sources. But then it loses every value it could have to readers - who will almost all of them assume that the median express the central tendency of the internet usage population. The ignorance displayed here about these basic statistical principles is the very reason that WP:CALC requires consensus for a calculation to be allowed in. While the calculation itself is simple, the selection of sources, it's applicability and not least the interpretation of the result is anything but straightforward. The sources are all biased and they sample different populations. --Useerup (talk) 09:07, 11 December 2011 (UTC)
- Can you cite a source and explain in more detail this idea of taking the median from different populations not erasing the bias? I have heard that when you ask a crowd of people to estimate the number of gumballs in a container, most individual people give inaccurate (biased) answers. However, by taking the median of these biased guesses, you get a very accurate result. Here is an anecdote that confirms this story. It seems to me that if we have a bunch of measurements of the usage share of browsers, taking the median will result in a more accurate answer than picking a random one, or even the most cited one. Does the fact that one measurement is biased towards large sites, and one is biased towards European sites, and the others are biased in other ways, mean that the median will not become close to the actual value as the number of measurements increases? If so, what is the explanation for the wisdom of the crowd? -- Schapel (talk) 14:56, 11 December 2011 (UTC)
- There is ignorance in those that don't believe renowned professor of statistics on the usefulness. Daniel.Cardenas (talk) 19:17, 11 December 2011 (UTC)
- Can you cite a source and explain in more detail this idea of taking the median from different populations not erasing the bias? I have heard that when you ask a crowd of people to estimate the number of gumballs in a container, most individual people give inaccurate (biased) answers. However, by taking the median of these biased guesses, you get a very accurate result. Here is an anecdote that confirms this story. It seems to me that if we have a bunch of measurements of the usage share of browsers, taking the median will result in a more accurate answer than picking a random one, or even the most cited one. Does the fact that one measurement is biased towards large sites, and one is biased towards European sites, and the others are biased in other ways, mean that the median will not become close to the actual value as the number of measurements increases? If so, what is the explanation for the wisdom of the crowd? -- Schapel (talk) 14:56, 11 December 2011 (UTC)
- All the stats sites are bias. Exactly! And the median - unlike what someone here naively assumes - will not erase that bias. For the median to express anything useful about a population (internet usage), the samples must be drawn from the same population (internet usage). In that case the median can be used to erase sampling error. But none of the sources sample the same population and they use very different methodologies at that. The only thing a median could express is that it is the median of the sources. But then it loses every value it could have to readers - who will almost all of them assume that the median express the central tendency of the internet usage population. The ignorance displayed here about these basic statistical principles is the very reason that WP:CALC requires consensus for a calculation to be allowed in. While the calculation itself is simple, the selection of sources, it's applicability and not least the interpretation of the result is anything but straightforward. The sources are all biased and they sample different populations. --Useerup (talk) 09:07, 11 December 2011 (UTC)
- "broad inclusive range of high volume sites" - hmmm, yeah right. Where did you pull that from? All the stats sites are bias. Daniel.Cardenas (talk) 03:37, 11 December 2011 (UTC)
- The point about "random sample" is not that the stats should all be random samples. You're taking the Wikimedia stats as representative of general Internet use - a very limited sample of Internet traffic - but based on what principle? In what way are you establishing representativeness? (other stats do this by taking a broad inclusive range of high volume sites) It's clearly a biased sample. You call my support for using one source only a "misapplication of statistics". If you think each of the most commonly used major stats sources misapplies statistics, bring that to the table, rather than take an average of sources you apparently consider to be seriously flawed. VsevolodKrolikov (talk) 02:37, 11 December 2011 (UTC)
- You fail to understand that none of the sites practice random sampling. Perhaps you might review your support for this misapplication of statistics? Daniel.Cardenas (talk) 01:58, 11 December 2011 (UTC)
- Er, no. My criterion was quite clear: "the most commonly used source". This meets NPOV. But it's good to see you now understand that these sites are not in practice measuring the same thing. In that light, perhaps you might review your support for this misapplication of statistics? VsevolodKrolikov (talk) 01:20, 11 December 2011 (UTC)
- So to you it would make more sense to choose one of the stat sites that doesn't random sample and manipulates the numbers? Daniel.Cardenas (talk) 00:24, 11 December 2011 (UTC)
- Oppose per VsevolodKrolikov. -- Scjessey (talk) 12:41, 9 December 2011 (UTC)
Conclusion: No consensus
This was the opening statement of this poll:
- wp:No original research states that a wp:routine calculation should have wp:consensus. Would like to poll and discuss to see if there is consensus. Also note that wikipedia doesn't wp:vote on issues.
It is obvious from the above discussion that there is no consensus on the issue. — Preceding unsigned comment added by Useerup (talk • contribs) 06:19, 12 December 2011 (UTC)
- I don't think we should have a section where people express their opinions while the poll is still open. In other words, I'd prefer if this section was blanked, including this comment, until the poll is closed. --Nigelj (talk) 16:40, 12 December 2011 (UTC)
- No, nothing is obvious yet. There is no conclusion. Please wait until the poll ends 1exec1 (talk) 23:21, 12 December 2011 (UTC)
- I think the premise of this poll is flawed. Routine calculations, that is, arithmetic, requires consensus according to WP:NOR. According to Richard Gill, finding the median does not involve arithmetic. If this is the case, then no consensus is even necessary. -- Schapel (talk) 23:33, 12 December 2011 (UTC)
- Also I think the above poll is good enough consensus. Someone external to the conversation should summarize. From wp:consensus:
- "... "Consensus" on Wikipedia does not mean that decisions must be unanimous (which, although an ideal result, is not always achievable); it is not a vote either. It means, rather, that the decision-making process involves an active effort to incorporate all editors' legitimate concerns, while respecting community norms. ..."
- Daniel.Cardenas (talk) 00:20, 13 December 2011 (UTC)
- You don't appear to be making an active effort to take into account other editors' legitimate concerns. If you could address that oversight, it might help things.VsevolodKrolikov (talk) 01:43, 13 December 2011 (UTC)
- Which concerns? Or is this another of your oversights? Daniel.Cardenas (talk) 03:17, 13 December 2011 (UTC)
- Have you ever read our policy on NPOV, Daniel?VsevolodKrolikov (talk) 03:30, 13 December 2011 (UTC)
- He may be not involved into the discussion, but your concerns are reasonably addressed. Please see above posts. 1exec1 (talk) 14:13, 13 December 2011 (UTC)
- Which concerns? Or is this another of your oversights? Daniel.Cardenas (talk) 03:17, 13 December 2011 (UTC)
- You don't appear to be making an active effort to take into account other editors' legitimate concerns. If you could address that oversight, it might help things.VsevolodKrolikov (talk) 01:43, 13 December 2011 (UTC)
- Also I think the above poll is good enough consensus. Someone external to the conversation should summarize. From wp:consensus:
- I think the premise of this poll is flawed. Routine calculations, that is, arithmetic, requires consensus according to WP:NOR. According to Richard Gill, finding the median does not involve arithmetic. If this is the case, then no consensus is even necessary. -- Schapel (talk) 23:33, 12 December 2011 (UTC)
Suggested solution to the median issue
Statcounter seems to be the most commonly cited measure. It is also a measure which closely matches the "median" calculated here. How about we choose Statcounter for the lede? The reader sees more or less the same figures. VsevolodKrolikov (talk) 16:45, 12 December 2011 (UTC)
- ...at the moment. And how did you work that out? Oh, by looking at the median row. QED. That is exactly the reason I gave for keeping it in my !vote above. --Nigelj (talk) 16:50, 12 December 2011 (UTC)
- The reasons for not liking this "median" are not because people don't want information to be easy to access. Using Statcounter or Net Applications (or a table featuring both) would meet NPOV better than a "median" comprising these two frequently cited sources plus three that are largely off the RS radar. Statcounter would be closest to those favouring the "median", so it seemed like a good compromise. But if you're wedded to this calculation, not much is going to persuade you. VsevolodKrolikov (talk) 16:59, 12 December 2011 (UTC)
- Choosing 5 sources, calculating the medians, finding which of the 5 sources supplies most of the medians, then erasing three of the sources, and keeping the one that is closest to the medians, and one that serves mainly the US market and so is far away from many of the medians. That sounds the worst of all possible solutions. How would you explain that methodology in a footnote? --Nigelj (talk) 19:04, 12 December 2011 (UTC)
- eh? What? Where do you get that from? My personal preference is to use either Statcounter or Net Applications or both, because they are the most commonly cited and I value NPOV. However, there is a group of editors, of which you are one, who want to emphasise (in my view) the use of rarely or never cited measures w3counter, clicky and Wikimedia as having equal status with Statcounter and net applications, and calculate a "median" of these measures. I think this approach stinks much like I'm sure you think my rejection of your approach stinks, but at the same time I notice that one outcome I would be happy with - the use of Statcounter in the lede as one of the most cited measures, broadly coincides with the result you favour. Using statcounter overcomes any WP:V, WP:CALC and WP:OR objections. I genuinely don't understand why you think it's such a terrible idea.VsevolodKrolikov (talk) 19:14, 12 December 2011 (UTC)
- Choosing 5 sources, calculating the medians, finding which of the 5 sources supplies most of the medians, then erasing three of the sources, and keeping the one that is closest to the medians, and one that serves mainly the US market and so is far away from many of the medians. That sounds the worst of all possible solutions. How would you explain that methodology in a footnote? --Nigelj (talk) 19:04, 12 December 2011 (UTC)
- The reasons for not liking this "median" are not because people don't want information to be easy to access. Using Statcounter or Net Applications (or a table featuring both) would meet NPOV better than a "median" comprising these two frequently cited sources plus three that are largely off the RS radar. Statcounter would be closest to those favouring the "median", so it seemed like a good compromise. But if you're wedded to this calculation, not much is going to persuade you. VsevolodKrolikov (talk) 16:59, 12 December 2011 (UTC)
- Support for this suggestion. Median goes and one counter is picked for the lede. As per the above poll which shows absence of consensus and WP:CALC which requires consensus to keep a calculation. Useerup (talk) 19:51, 12 December 2011 (UTC)
- Support I am not particular about which of the sites is used but going down to one representative sample seems good, we can always give links to the other cites so interested poeple can look up more data if they want. Thenub314 (talk) 23:34, 12 December 2011 (UTC)
Apples, oranges, ravens and office desks
There has been talk of such things here. I understand that we are only measuring one thing, viz how many web visitors use this, that, or the other browser. The article goes into some detail as to how this is actually very difficult to measure. Then we take all the known sources of major information and provide detailed figures going back months and years from each of them. In between, we try to summarise. We choose five major measuring sources, that each provide data based on billions of web visits in the most current month, and it is surprising by how much they differ. So we provide a summary. The summary includes medians. End of. If we had a jar of sweets and asked some Americans to estimate how many there were, then some Europeans, then some people who look up things on the Wikimedia sites, we'd get different estimates. What could we do? Summarise and show a median. If we had estimates from people interested in X, Y or Z, we could include those. It wouldn't affect the fact that all the measurements were of the same thing, just different subsets from the true, global, (unmeasurable) world population. The more we got, the better. I suppose if some of the estimates were from a society for the blind, or from the association of sweetshop assistants, we may have to think about those. I don't see any apples and oranges problem here. --Nigelj (talk) 22:37, 12 December 2011 (UTC)
- I certainly do, some of the numbers are adjusted with CIA census data to estimate users from which there are not any hits. Others such as wikimedia use raw hits, and to not try to account for regions which don't have access to their servers, or where there servers are not very well used. As with the article I linked in my !vote these kinds of difference mean effectively that your measuring different quantities. I do believe there is an apples and oranges situation here. Thenub314 (talk) 23:30, 12 December 2011 (UTC)
- Nigel, I know stats can be confusing, but you do not appear to understand the arguments made against the use of this "median". inter alia, these different measures are looking at different jars, not the same one several times - your statement that "we are only measuring one thing" is flat out wrong. The targets can be described in vague English in the same way, but that's just a distraction. Furthermore, some of these jars do not appear to be relevant in the real world. For example, how many sites independent of the Wikimedia foundation cite Wikimedia stats as representative of the Internet? It's probably close to zero. Per WP:NPOV (seriously, read it) we should not give these stats any weight at all. Clicky and w3counter seem pretty obscure too.VsevolodKrolikov (talk) 23:51, 12 December 2011 (UTC)
- The stats sources are estimating the actual usage share of browsers by measuring the usage share of a biased sample. The jar in this case is the actual usage share of browsers. The estimate is the measurement of their biased sample. In the case of a crowd of people estimating the number of jellybeans in a jar, each person is using their own mental model to determine their estimate. But just because their models are different does not mean that they are estimating different things -- they are all estimating the number of jellybeans in the jar, each in their own different and biased way. When you take the median of their estimates, it tends to approximate the actual number of jellybeans well. -- Schapel (talk) 00:08, 13 December 2011 (UTC)
- No, that's wrong. They measure different populations and have different definitions of "browser usage". Look at the dramatic difference between Net Applications, Statcounter and Wikimedia stats. The differences are not reflections of margins of error in measuring. (You're confusing an imaginary situation constructed by some of the editors in this discussion here with the real one.)VsevolodKrolikov (talk) 01:41, 13 December 2011 (UTC)
- I don't see any dramatic differences between browser stats. I see numbers that agree fairly closely. I understand that the differences are not related to errors in measuring. The differences are caused by different biases in the samples that they take. Similarly, if you ask many different people how many jelly beans are in a jar, they will report wildly different numbers, each biased by the method they used to estimate the number. If you take the median of those guesses, you get an accurate number. -- Schapel (talk) 02:02, 13 December 2011 (UTC)
- So 35% and 50% (IE) is close agreement? 6% and 11% (Safari) is close agreement? 15% and 22% (Chrome) is close agreement? You're kidding, right? These are large differences in market share. If a browser managed to increase its share by 50-90% it would be big news. Please don't pretend to engage in discussion if you've no intention of considering the evidence, as that would be acting in bad faith.VsevolodKrolikov (talk) 02:11, 13 December 2011 (UTC)
- Which one of them is correct? I feel that picking one is a bigger WP:OR than simply calculating a median for illiustrative purposes. Can we just consider the median an illustration in a text form (and use it as such)? No original research involved. 1exec1 (talk) 12:30, 13 December 2011 (UTC)
- Yes, those are pretty close. When you ask people the number of jelly beans in a jar, they give results that vary by an order or two of magnitude. So, yes, the numbers you cite are in close agreement relative to other situations where the wisdom of the crowds works well. So it seems to me that we can use it here. I am considering evidence, but I don't see you providing any. Do you have any evidence that the median of the measured usage share of browsers is not close to the actual value? All I see is people claiming that we can't do so, without evidence. -- Schapel (talk) 13:16, 13 December 2011 (UTC)
@1exec1. Choosing one of them is not original research at all, as we would not be creating new information. The policy we have to be careful of is WP:NPOV - that we are biasing the article. NPOV requires us to present viewpoints in proportion to their preponderance in reliable sources. If we are going to choose one measure for the lede, we choose the most popularly used measure (or possibly - in this case - the most popular two measures). As it stands, we're putting StatCounter and Net Applications - which are cited rather a lot judging by a rough google news archive count - on the same footing as Clicky and W3counter, which appear to be really rather more obscure, and Wikimedia foundation stats which I gravely doubt are cited by any independent source as a measure of general browser usage. In other words, we're contravening NPOV by giving equal credence to these measures in our calculation of this "median", particularly with the Wikimedia one, which no RS employs. (A graph of the Wikimedia results does no harm in the article, but it should absolutely not play any role at all in our headline descriptions of the market shares; that would break both NPOV and OR.) Look at it like this - if an editor were interested in using Wikipedia to undermine one browser or promote another, s/he could easily just keep adding or removing various measures to push the result towards the one s/he wanted. NPOV policy is a core policy for a reason, and it applies just as much in computer articles as in political ones.
- The issue is thus not "which one is correct", but "which one is most commonly used". This is how we keep our neutrality in relation to the real world. It's clearly either Net Applications or Statcounter. I wouldn't object to either; it has struck me that having both side by side in the lede might be educational as to the different results one gets from different ways of counting, sampling and adjusting.VsevolodKrolikov (talk) 14:20, 13 December 2011 (UTC)
- The problem is that deciding which source is most commonly used is itself OR. There are so many RS talking about browser share, that we need to use (statistical ?) methods such as GoogleNews, Google search, etc. Which one of them represents reliable sources? They give vastly different results (see my comment in the poll), how we can address that, WP:DUE and WP:NPOV then? Use median to combine the results of different methods that represent RS in some way? As you can see, we quickly return to the same problem we are solving, just with much more complexity and OR. Thus I do not think that picking some browser statistics engine is going to solve anything with less OR involved. 1exec1 (talk) 14:45, 13 December 2011 (UTC)
- I've answered above - use better search terms. I get Net Applications and StatCounter coming out consistently way ahead of others. For the life of me I do not understand how any of you can think Wikimedia stats have any place in a general calculation. You've all been on here a while, you should all be familiar with NPOV. VsevolodKrolikov (talk) 15:29, 13 December 2011 (UTC)
- The problem is that deciding which source is most commonly used is itself OR. There are so many RS talking about browser share, that we need to use (statistical ?) methods such as GoogleNews, Google search, etc. Which one of them represents reliable sources? They give vastly different results (see my comment in the poll), how we can address that, WP:DUE and WP:NPOV then? Use median to combine the results of different methods that represent RS in some way? As you can see, we quickly return to the same problem we are solving, just with much more complexity and OR. Thus I do not think that picking some browser statistics engine is going to solve anything with less OR involved. 1exec1 (talk) 14:45, 13 December 2011 (UTC)
@Schapel. I don't know how to put this delicately, but you don't seem to have grasped at all what the differences between these measures are. You keep talking as if there is only one population to count and one aspect of that population these measures are counting. This is simply not the case. The reason for the differences is not statistical or human error. It's because in practice they're not all trying to count the same thing. You may have what you think is a straightforward idea of "usage share" in your head, but establishing what that means in practice has resulted in varied approaches and definitions. Real world statistics are not like statistics one does in the maths classroom. (As for considering differences of 50% and more in results acceptable in statistics - that really strains credulity).VsevolodKrolikov (talk) 14:20, 13 December 2011 (UTC)
- How are they counting not the same thing? They all measure global browser statistics, just with different biased samples of the global population. You suggest that picking one biased sample (one browser stat engine) and presenting that in the article is better than presenting several biased samples. Can you provide any statistically sound method how we can select the biased sample to present? 1exec1 (talk) 14:56, 13 December 2011 (UTC)
- Woah - you haven't understood that either? OK: First of all, the populations are different. Wikimedia foundation stats are reflective only of people who use wikimedia websites. This is going to be biased towards people who edit the sites (disproportionately white, male, college/technically educated, anglophone) as well as those looking up information (likely to have a higher level of education than average). Net Applications works with 40,000 partners, but at least they have a range of commercial, corporate, content, public and other sites. StatCounter gets information from over three million sites, but only those that volunteer to carry its tracking code. Its website says "StatCounter Global Stats are based on over 15 billion hits per month, by a random sample of people worldwide, to over 3 million global websites, covering multiple interest areas and geographic locations".
- As for bias: While both statcounter and Net Applications have general bias problems in the sample (Judging by statcounter's figures, towards the US and away from places like China and Japan), only Net Applications deals with this by re-weighting their stats geographically in accordance with CIA stats on internet usage. That is, Statcounter does nothing to remove the bias. In other words, they look at different populations to begin with, and then Net Applications does something to their data which changes it a lot. So all three measures have very different samples, and only one of them does anything to get rid of the bias in their sampling. They're not all measuring "global" usage, or even representative usage in a single country.
- There is also a different way of defining "usage". Is it a unique visit (NA, W3counter), or is it every page hit? This matters because it views different browsing habits differently.
- I'll stress again - real world statistics is not like stats at school. When you ask a question like "which browser is used the most", there is not one clear way of defining that question quantitatively. Operationalising concepts for measurement is not a straighforward hand-wavy procedure.VsevolodKrolikov (talk) 15:29, 13 December 2011 (UTC)
- How are they counting not the same thing? They all measure global browser statistics, just with different biased samples of the global population. You suggest that picking one biased sample (one browser stat engine) and presenting that in the article is better than presenting several biased samples. Can you provide any statistically sound method how we can select the biased sample to present? 1exec1 (talk) 14:56, 13 December 2011 (UTC)
- So 35% and 50% (IE) is close agreement? 6% and 11% (Safari) is close agreement? 15% and 22% (Chrome) is close agreement? You're kidding, right? These are large differences in market share. If a browser managed to increase its share by 50-90% it would be big news. Please don't pretend to engage in discussion if you've no intention of considering the evidence, as that would be acting in bad faith.VsevolodKrolikov (talk) 02:11, 13 December 2011 (UTC)
- I don't see any dramatic differences between browser stats. I see numbers that agree fairly closely. I understand that the differences are not related to errors in measuring. The differences are caused by different biases in the samples that they take. Similarly, if you ask many different people how many jelly beans are in a jar, they will report wildly different numbers, each biased by the method they used to estimate the number. If you take the median of those guesses, you get an accurate number. -- Schapel (talk) 02:02, 13 December 2011 (UTC)
- No, that's wrong. They measure different populations and have different definitions of "browser usage". Look at the dramatic difference between Net Applications, Statcounter and Wikimedia stats. The differences are not reflections of margins of error in measuring. (You're confusing an imaginary situation constructed by some of the editors in this discussion here with the real one.)VsevolodKrolikov (talk) 01:41, 13 December 2011 (UTC)
- The stats sources are estimating the actual usage share of browsers by measuring the usage share of a biased sample. The jar in this case is the actual usage share of browsers. The estimate is the measurement of their biased sample. In the case of a crowd of people estimating the number of jellybeans in a jar, each person is using their own mental model to determine their estimate. But just because their models are different does not mean that they are estimating different things -- they are all estimating the number of jellybeans in the jar, each in their own different and biased way. When you take the median of their estimates, it tends to approximate the actual number of jellybeans well. -- Schapel (talk) 00:08, 13 December 2011 (UTC)
- Nigel, I know stats can be confusing, but you do not appear to understand the arguments made against the use of this "median". inter alia, these different measures are looking at different jars, not the same one several times - your statement that "we are only measuring one thing" is flat out wrong. The targets can be described in vague English in the same way, but that's just a distraction. Furthermore, some of these jars do not appear to be relevant in the real world. For example, how many sites independent of the Wikimedia foundation cite Wikimedia stats as representative of the Internet? It's probably close to zero. Per WP:NPOV (seriously, read it) we should not give these stats any weight at all. Clicky and w3counter seem pretty obscure too.VsevolodKrolikov (talk) 23:51, 12 December 2011 (UTC)
- ^ "StatCounter Global Stats: Desktop vs. Mobile". StatCounter. Retrieved 24 Sept 2011.
{{cite web}}
: Check date values in:|accessdate=
(help)
- B-Class Internet articles
- High-importance Internet articles
- WikiProject Internet articles
- B-Class software articles
- Unknown-importance software articles
- B-Class software articles of Unknown-importance
- B-Class Computing articles
- Unknown-importance Computing articles
- All Computing articles
- All Software articles