Hi Phoebe --

Allow me to direct you to the eminent statistician George Box's aphorism: "All models are wrong but some are useful"[1] :)

Right now the Reading team models our success on three axes:

Reach: how many people read our content
Engagement: how much do they read
Retention: how often do they come back

Just as metrics for games measure "fun", we think of ours as measuring "learning" but there's a lot of nuance left on the table, particularly in the areas that you highlight above.

We're slowly building up our quantitive and qualitative understanding of our Readers. Leila and team have been doing some really interesting work on Reader intention and we've made some serious strides in our qualitative research as well. Every new project and exploration adds another dimension of understanding but clearly we need to balance complexity and effort with expressive value.


[1] https://en.wikipedia.org/wiki/All_models_are_wrong

On Fri, Mar 18, 2016 at 7:45 AM, phoebe ayers <phoebe.wiki@gmail.com> wrote:
On Fri, Mar 18, 2016 at 9:45 AM, Toby Negrin <tnegrin@wikimedia.org> wrote:
> Hi Andrew, Phoebe -- here's what the Communication department here is
> comfortable with:
>> Wikipedia and the other projects operated by the Wikimedia Foundation
>> receive hundreds of millions of unique visitors per month

Thanks Toby. I'll start using that and add it to my presentation toolkit.

>  The numbers of unique devices for enwiki are far greater than 500mm. But of
> course, people use more than one device and/or shared computers. There are
> datasets out there about average number of devices per person so we could
> potentially use this as a scaling factor to get a higher level of confidence
> but IMO the mapping from device to actual human is always going to be dicey.

Sounds like a research project! Like Andrew, for communication
purposes I'm less interested in exactitude than I am in
order-of-magnitude. (The kinds of things I use these numbers for:
comparisons against the online population, against the population
reached by libraries, etc. -- all of which is deeply qualified, of

Semi-off topic thoughts about "reach" as a metric:

I wonder if there's a qualitative project somewhere in here about
*types* of use -- e.g. if I'm using WP on my phone & my work pc is
that really equivalent use? Perhaps I am using them for different
kinds of information seeking, e.g. looking up terms related to work vs
looking up info on movie stars -- does this different kind of use
matter for how we construct and present information, or count "use"?

Can we build testable hypotheses about use patterns & needs for people
who do straight device swapping (phone to tablet to pc, for the same
purposes) versus people who have devices for different purposes (i.e.
work v. personal) versus people who share devices? (Obviously, all
this goes well beyond just Wikipedia use).

I also think there's something in here about levels of access related
to language, which relates to multilingual use of Wikipedia. Someone
who speaks a language served by a Wikipedia with 100K articles can not
access Wikipedia to the same depth or level that a person who can use
English Wikipedia with 5M articles can.

In other words, though we talk about reach, not all reach is the same.
The depth to which Wikipedia reaches me -- someone with unlimited data
on multiple devices and 24/7 device access for all purposes, who reads
English well and a couple other languages poorly -- is way different
from the depth which someone with part-time access on a mobile phone
who speaks an underserved language is reached by our projects. This
may be pretty obvious, but I hadn't thought about the implications for
claiming "we reach x millions" before.

-- phoebe

> For comparison, I worked at Yahoo for a long time and generally understand
> their tech stack -- in their 2014 annual report, Yahoo speaks to "more than
> 1 billion MAUs".[1] From my experience, I really don't know how they could
> measure this with any certainty without estimation or other statistical
> techniques because they have the same measurement issues that we do. Only
> Facebook or other sites where personalization is necessary for the site to
> work can report on reach without some sort of qualification.
> -Toby
> [1]
> http://static.tumblr.com/7drgjla/386nnw4n9/yahoo_inc._2014_annual_report.pdf
> On Fri, Mar 18, 2016 at 4:09 AM, Andrew Gray <andrew.gray@dunelm.org.uk>
> wrote:
>> On 17 March 2016 at 19:40, phoebe ayers <phoebe.wiki@gmail.com> wrote:
>> >> One of the drawbacks is that we
>> >> can't report on a single total number across all our projects.
>> >
>> > Hmm. That's unfortunate for understanding reach -- if nothing else,
>> > the idea that "half a billion people access Wikipedia" (eg from
>> > earlier comscore reports) was a PR-friendly way of giving an idea of
>> > the scale of our readership. But I can see why it would be tricky to
>> > measure. Since this is the research list: I suspect there's still lots
>> > to be done in understanding just how multilingual people use different
>> > language editions of Wikipedia, too.
>> Building on this question a little: with the information we currently
>> have, is it actively *wrong* for us to keep using the "half a billion"
>> figure as a very rough first-order estimate? (Like Phoebe, I think I
>> keep trotting it out when giving talks). Do the new figures give us
>> reason to think it's substantially higher or lower than that, or even
>> not meaningfully answerable?
>> --
>> - Andrew Gray
>>   andrew.gray@dunelm.org.uk
>> _______________________________________________
>> Wiki-research-l mailing list
>> Wiki-research-l@lists.wikimedia.org
>> https://lists.wikimedia.org/mailman/listinfo/wiki-research-l
> _______________________________________________
> Wiki-research-l mailing list
> Wiki-research-l@lists.wikimedia.org
> https://lists.wikimedia.org/mailman/listinfo/wiki-research-l

* I use this address for lists; send personal messages to phoebe.ayers
<at> gmail.com *

Wiki-research-l mailing list