Wikipedia talk:Web statistics tool

From Wikipedia, the free encyclopedia


Contents


A Wikipedia article traffic statistics tool at http://stats.grok.se has been recently quoted in a number of WP:RM discussions.

I've created this talk page as a central point for discussion as to its merits, and how it should be used. I'll put some basic information in the project page as well, but this talk page is the more important one for now. Andrewa (talk) 01:09, 15 March 2008 (UTC)

I've turned off sinebot for this page... Please sign comments, but I don't think you need to sign entries in the list of requested moves nelow. Andrewa (talk) 14:40, 15 March 2008 (UTC)

[edit] Requested moves

In order to evaluate the usefulness of the tool for the purposes of WP:RM, I've started a list of discussions in which the tool has featured. It's not exhaustive; Feel free to add to it. We might keep it alphabetical for now. Andrewa (talk) 01:21, 15 March 2008 (UTC)

Results are in the format closed support-oppose result, ratio of page hits as measured by the web statistics tool to next most viewed page. Results of open discussions are not shown.

[edit] Article importance

[edit] Known anomalies

Situations where the counts have been obviously in error, possibly because of a program accessing an article a large multiple number of times, and not reflecting actual users.

  • The stats for the 3rd and 4th of March missed some hours entirely, so those two days are omitted.
  • Canine reproduction went from a steady 450 views a day to a steady 170,000 views per day for about two weeks, then dropped to 8,800 views per day.
  • Newton went from a steady 1.5 to 3 k views per day to over 300,000 views for one day only.
  • Resolution normally gets viewed about 500 to 1,000 times a day, was hit 300,000 times a day for three days (same program that hit Newton?).
  • Lambda calculus normally gets viewed about 400 to 800 times a day, was hit 130,000 times on 31 March/1st April.

There are also examples where an article actually receives a large number of views on one day because of media attention. Examples of that are Superdelegate and Julie Dubela, and most weekly TV shows such as American Idol. Heath Ledger went from 4,000 views to 2 million the day after his death and has been steadily tapering off since. Articles that are featured on the Main page clearly get a lot of views because of that exposure such as Xenon which was a featured article on February 10th. Valentine's day got 1.1 million views on, well, Valentine's day.

[edit] Other uses

Is anybody clicking through and reading any of these sub-articles?

[edit] How helpful is the tool for WP:RM decisions?

From Talk:Willie Johnson#Survey:

The stat tool should be the only evidence that matters. It tells you who is viewing what pages based on what words they enter to get to them. —Preceding unsigned comment added by TonyTheTiger (talk • contribs)

From http://stats.grok.se/about :

I wouldn't base any important decisions on these stats.

More discussion is required before we adopt Tony's recommendation, IMO! Andrewa (talk) 02:42, 15 March 2008 (UTC)

Hi Andrew. One instance where the tool might be useful: if someone was arguing that something was "primary use", whereas I thought there was no primary, then the results of the stat tool might help me make my case. On the other hand, the tool should probably not be used as the sole argument that something is primary use. Sam Staton (talk) 11:35, 15 March 2008 (UTC)
Agree. Andrewa (talk) 14:41, 15 March 2008 (UTC)
Page views can be subject to deliberate and unintended false readings. One example is documented at Louis Pasteur, which got 27,000 views on one day and less than a thousand a day the rest of the month, another at Canine reproduction which got 500 views a day except for a two week period where it got 150,000 views a day. There are other considerations to use other than just more page views when establishing primary usage. For example, George W. Bush gets twice as many hits as George H. W. Bush, but since they are both presidents there is no way that anyone would propose to choose one as the primary usage over the other, although if they did, it would be senior, and not junior that was used, as was done in the case of John Adams and John Q. Adams. --Gonezales (talk) 17:48, 15 March 2008 (UTC)
I think that relying on a tool that attempts to measure Wikipedia page views as the sole basis for determining primary topic is inherently problematic for a variety of reasons. The tool itself is still new and not exactly proven to be accurate with a high degree of reliability. Second, it is rather solipsistic. It might sometimes give an indication for when page titles could be better arranged, but by itself the sample is far too limited. IMO, unless the case is so lopsidedly obvious, determination of primary topic should be based on external sources and not simply on internal page views. older ≠ wiser 19:15, 15 March 2008 (UTC)
It seems to me that, at the very least, we have a rough consensus rejecting the claim that the stat tool is the only evidence that matters. A proposal along these lines seems to have no chance of becoming Wikipedia policy.
There also seems to be a rough(er) consensus that the tool should never be accepted as evidence supporting a particular primary usage, however unbalanced the statistics may be in one direction. And this is consistent with the results in WP:RM recently, where several proposals claiming that the stat tool supported a particular primary usage have been rejected.
But there's also the interesting suggestion that the tool may be useful to support a claim that there is no primary usage. Andrewa (talk) 05:53, 17 March 2008 (UTC)

One editor appears to think that the tool is the only criteria that should be used for establishing primary usage. --Gonezales (talk) 03:36, 19 March 2008 (UTC)

Yes, exactly. And here this editor has the opportunity to give reasons for this belief, as have any others of this opinion. So far the silence (;-> is deafening. Andrewa (talk) 17:27, 11 April 2008 (UTC)
I noticed that. Well I think it should be the primary tool for answering questions about primary usage, but that it should be backed up by the number of links. I'm trying to get a number on how dominant a topic should be for it to be called primary, but so far the old school "it's dominant if I think it's dominant" have resisted using it at all, but I think that in time it will become more accepted. Bear in mind though that there actually is no requirement that anyone defend their position - arguments stand or fall on their own merits. 199.125.109.76 (talk) 01:34, 12 April 2008 (UTC)

[edit] Suggestions?

Hi, seeing that a page seems to have been created to discuss this I thought that I'd use it to gather suggestions. If you have any suggestions or thoughts on new features you'd like to see in the statistics tool, I'd be interested to hear about it.

Some of the thoughts I've had so far is to be able to view a ranked list of all articles in a category, and to compare several articles directly. Some of the other suggestions have been an API, being able to view most accessed non-existent pages, fixing case insensitivity, and top lists for all languages. I'd be willing to be open to letting people here guide the development if we can come up with a ranked list of things you'd like me to start with. What would be most useful? henrik•talk 22:01, 17 March 2008 (UTC)

I'll add my voice to the chorus of thanks. Whilst day-by-day charts are fun, what I really want is to get stats on lots of articles (a Project can have anywhere from 1000-50,000 articles), with short downlad times - but the stats don't need to be timely, and they don't need to be at higher resolution than 1 month. So for instance, if there's a source of data that updates in April reporting a single number for hits-in-January, that's absolutely fine, just so long as I'm not having to wait 5 seconds to download the stat for just one article - multiply that by 50,000 articles and that becomes a lot of seconds! I just need to get a rough idea of what articles are hot and what not at some time in the vagely recent past. Then armed with that knowledge we can set priorities in a Project - for instance which Top importance articles are really the priorities to get to GA first, or which Low importance articles are actually of more interest than the Project thinks. I've just had an assessment bot approved which among other things assesses town articles on the basis of their population - obviously small but world-famous places like St Tropez have to be picked out manually, and easy access to page hits would be a useful way to highlight potential misassessments. With my very Project-centric view of things, ideally the stats should be integrated into the WP 1.0 Bot assessment pages - just run it once a month to update with "last months" stats, or give WP 1.0 Bot access to the data so that any articles added to the Project get the stats when WP1B adds them to that list. I guess when I've got some time http://stats.grok.se/~henrik/wikistats/pagecounts_en_20080201_to_20080223_full_sorted.gz would probably serve me quite well, just pick out the articles I want from that, but it would be nice (for both of us) if I didn't have to download 450MB.... Just a web interface where I can send a text list and out comes page hits in January would be fine. Thanks again. FlagSteward (talk) 02:14, 22 March 2008 (UTC)

I think the tool itself is great. What needs development is the methodology for its use... some guidelines, based on existing guidelines and of course policy.

At the very least, the tool is useful to suggest paths of enquiry, for example suggesting what might be a primary usage. Evidence as to whether or not this is true can then be gathered from other sources. At the very most, IMO its usefulness falls short of the proposal that it is (or should be) the only evidence that matters in deciding primary usage.

If so, these are boundaries. They're very conservative IMO, and deliberately so, as a starting point. How can we improve on them? Or can we at least agree on them, as the first step to refining them? Andrewa (talk) 17:52, 11 April 2008 (UTC)

I think I am pretty much in agreement with Andrewa on this. I think the tool is potentially very useful -- but only as an indicator that requires interpretation -- not as results to be mechanistically implemented according to arbitrary thresholds. IMO, the most useful potential lies in providing evidence that a title should be a disambiguation page rather than a specific article. It seems to me that is easier to demonstrate by comparing page views than it is to conclusively demonstrate that one particular topic merits primary topic status. older ≠ wiser 18:04, 11 April 2008 (UTC)
No one would disagree that it has to be interpreted. There is no reason for example for working on a page just because it was mentioned on the Main page a few months ago and got a spuriously high count because of that. Or for jumping on a page of someone who had an untimely death a couple of months ago. Or deciding in April that Valentine's day was the most important article to work on because it got a million page views two months ago. Does the horse leaving the barn ring a bell? 199.125.109.76 (talk) 01:41, 12 April 2008 (UTC)