You Count
How many readers does Slate
have? Like other Web sites, we get asked this question all the time. Our usual
reply is: "Thank you for asking. That's a complicated question." But here is
our best effort to answer it, along with an explanation of why counting readers
on the Web is so complicated (and why you should take any claims about Web-site
traffic--except ours, of course--with a grain of salt).
Traditional print magazines know exactly how many copies they sell, though they
have no idea how many people actually read a particular article or see a
particular ad. TV networks rely on sampling by companies like Nielsen to estimate how many
viewers were watching which channel at any hour of the day. (See Slate's recent
"Dispatch" from someone selected to be a "Nielsen family.") Nielsen
and other firms such as PC
Meter, now known as Media Metrix, are scrambling to adapt this concept to
the Web. Meanwhile, though, there are hits .
The contents of a Web site are stored on high-performance
computers called servers . When you type a Web address into your browser,
or click on a link or a "favorites" or "bookmark" button, you are telling your
computer to fetch a specific set of data--text, images, sound, etc.--from some
Web site's server. Your browser then assembles these data and displays them as
a page. Each of these requests for information is a hit.
Hits (per
day, per week, per month ...) are the most common measure of Web traffic. But
they are deceptive. A single Web page can be one hit or many, depending on how
it is constructed. The more separate elements it contains (images, sound ...),
the more hits it will require. That's why Slate, and some other sites, prefer
to talk about pages served . This is a measure of how many complete pages
your server has sent out in response to requests from browsers. In recent
months, Slate has been serving an average of about 90,000 pages a day.
But the number of pages served is also a
misleading measure in some ways. A "page" is not a standard unit on the Web the
way it is in print. Depending on a Web site's design, the same amount of
content can take up a very different number of pages. Slate is designed so that
every article or feature takes up a single page. Other sites break up articles
into many pages. (We're not suggesting that people are padding their page
counts. It's a judgment call about whether readers will find having to scroll
more annoying than having to click and wait.)
Also,
sometimes a site will feed you two pages when you've only asked for one. Slate
does this, for example, with our cover and contents pages. However--Boy and
Girl Scouts that we are--we subtract this double counting in the figures we
report. We cannot guarantee that every site is so scrupulous.
Further complicating page counts is the issue of
caching . Computers at corporations and other large organizations often
access the Web through a middleman computer known as a proxy server .
Your computer requests a page from the proxy server, which then requests it
from the Web site. But proxy servers are often programmed to save, or cache,
frequently requested pages, rather than retrieving them from the Web each time
they are requested. America Online also caches pages for its customers. And
your own computer is probably set up to cache some pages you've visited
recently. (That's why, when reading Slate, it's quicker to go back to the
contents page than it was to call it up at the start.) All this caching means
pages are served more quickly, but it does become harder for a Web site to know
how often its pages have actually appeared on someone's screen.
Anyway,
100 pages served could be 100 people reading this column, or one obsessive
participant in "The Fray." What we really want to know is how many
individual readers Slate has. And we can find out, sort of. Each request
to Slate from your browser carries with it an assortment of useful information.
Your computer tells us what operating system and browser software you are
using, so that our server can return a page appropriate to your setup. We also
are told about the referring page --that is, the page you were reading
when you requested this one. (Marketers love that information.) More important,
each page request is accompanied by a return address. How else would the server
know where to send the data? Every computer connected to the Internet has an
address, known as an IP address , consisting of four numbers. For
example, 207.68.136.88 is currently the IP address of one of Slate's servers.
Depending on how you access the Internet, your computer has either a permanent
or a temporary IP address.
But for a variety of reasons, including
caching, IP addresses aren't a great way of counting individual visitors.
Therefore, there are cookies . Cookies are like the club bouncer who
stamps your hand when you leave the premises temporarily, so as to identify you
when you return. The first time you visit Slate, our computer sends yours a
piece of data, which your computer sends back every time you return, so we know
it's you. Cookies aren't perfect. They don't account for multiple people
accessing the site from one computer, or one person using different computers
(or browser software). Also, some older browsers don't understand cookies, and
newer ones allow privacy freaks to turn them off. But, using cookies, we can
get a pretty good idea of how many unique browsers are visiting our
site.
Even that, though, is a
problematic concept. Do you measure browsers per day? Per week? Per month? The
longer the time period you choose, the more individual visitors you can claim,
which is nice. On the other hand, because of repeat visitors, the number of
unique browsers in a given month is less than the total of unique browsers per
week for those four weeks, which is less than the total of unique browsers per
day over those 30 days. Choose your poison: Slate attracts about 6,000 unique
browsers a day and 80,000 a month.
More
complications. Every Friday, 20,000 people get a print-out version of Slate
delivered to them by e-mail. (Click here to sign up.) Several hundred others download the print-out version directly from our site every day. A
few hundred determined traditionalists pay $70 to get this version of Slate
printed out and mailed to them each week via the U.S. Postal
Service. Hotmail, a
Web-based e-mail service, delivers Slate's table of contents to 106,000 of its
customers every week.
As this discussion shows, measuring Web traffic is
generally less exact than measuring traditional magazine circulation. In one
way, though, it is more exact. We know that every Slate page served--and, by
the way, the ad on that page--has hit the eyeballs of one reader (and in all
probability not more than one). Print publications have no clear idea of how
many people read each copy of their publication and, conversely, how many
individual pages of any given copy go unread. In practice, they tend to make
wild claims about the former and ignore the latter.
Even this distinction,
though, is changing with the development of off-line readers: software that
automatically goes to the Web to retrieve material, and stores it on your own
computer. Slate is already part of PointCast, which puts information you select on your screensaver.
Off-line readers put Web sites--and Web advertisers--in the same position as
their counterparts in the print world (although the off-line reader built into
Microsoft's upcoming Internet Explorer 4.0 will actually send the server a log of the
user's reading habits).
So how many readers does
Slate have? Taking all these complications into effect, and being as honest as
we can, we estimate about 28 million, give or take 27,900,000 or so.
Remarkable, isn't it?
