Web metrics

Web metrics

Many ways to skin a cat
Nov 29th 2007 | SAN FRANCISCO
From The Economist print edition


Such a lot of data, so little information

IMAGINE you are an advertiser, you want to place your banners on the most popular website, and you want to know how much to pay. Globally, the leading site is Google, which has the most “page views”. Or is it Microsoft, whose various sites have, in the jargon, the most “time spent”? Or should you go by unique users, duration, hits, click-through, impressions, queries, sessions, streams, or engagement? Whether or not there is truth in advertising, there is certainly none for online advertisers, at least none that is immediately obvious and simple. Bob Ivins, who is in charge of all non-American business for comScore, a big web-measurement firm, says that the web produces data as “a fire-hose shoots water”, and that working out what those data mean is rather like “putting a straw into the fire hose to take a sip”. Get the angle even slightly wrong, and you are blown away.






Take, for instance, the case of page views, the most widely used measure for much of the past decade. It is the number of times web surfers call up web pages on a given site. Page views became popular in the late 1990s, because they were far superior to the existing measure of “hits”, also known as “file requests”. Hits are confusing because every graphic on a page, as well as the page itself, counts as a hit. If a site owner puts more graphics on his pages, he gets more hits, even if visitors, clicks and everything else stay the same. “We produced hits numbers because we could, not because it was useful,” says one old-timer in the industry.
By comparison, page views do actually mean something, and are easy to comprehend by analogy to the offline world to boot—many advertisers are still used to counting pagination in magazines. So everybody started paying attention to page views. But then something odd happened. Page views at certain kinds of websites, especially the more sophisticated sort, began to decline, even though the site appeared otherwise healthy and popular.
The explanation has to do with “Web 2.0”, and more specifically with a constituent technology called “asynchronous JavaScript and XML”, or AJAX. This is a method that lets web pages update parts of themselves—a share-price ticker or an e-mail inbox, say—without having to reload and redraw the rest of the page, resulting in web pages that behave less like documents and more like pieces of software. But this means that a user of an AJAX page, such as Yahoo! Mail or Yahoo! Finance, can spend the entire day working on the same page, and this activity counts as only a single page view.
Perhaps advertisers should therefore ditch page views in favour of “user sessions”, since that promises to count actual people, and show how many of them use a site. Except that it doesn't, because this measure counts browsers rather than humans. So 2m sessions could mean, theoretically, that 2m people visited a site once, that 1m people visited twice, or that one astonishing individual visited 2m times. People tend to check their favourite pages in the office, at home, and even from their mobile phones, which leads to an overestimate of the number of users. Conversely, sometimes several people watch YouTube clips when gathered around the same screen, which leads to an underestimate of the number of users. Nobody looking at user sessions would ever know.
As websites, and especially those using AJAX, become more interactive, advertisers are therefore interested in other measures. “Duration” and “time spent”, for instance, suggest how long one or more people were interacting with a page, which in turn hints at how “engaged”, or alert, they were. Using these criteria, social-networking sites such as Facebook and MySpace (part of Fox Interactive Media in the chart) suddenly look attractive.
In the old days of traditional media, measures may have been simpler, but they were also dumber, says Randall Rothenberg, the boss of the Interactive Advertising Bureau, a trade association. In broadcast television or radio, firms such as Nielsen or Arbitron traditionally give gadgets to samples of volunteers that measure what their televisions or radios are tuned to; or they ask people to fill out diaries describing their reading, listening and viewing habits. Both methods produce notoriously unreliable estimates.
The web is an open book compared with those old media. Search advertisements, the text links on the results pages of search engines, charge advertisers only when a user actually clicks, thus expressing an interest. “Pre-roll” advertisements that run at the start of a web video report back exactly how many times they were viewed. But when it comes to banner advertising, says Mr Rothenberg, advertisers just have to consider all these new measures as they would a pointillist painting by George Seurat: looking at one dot is no fun; taking them all in can be rewarding.