I’ve run into a handful of cases lately where distinct IP Address counts have been offered as evidence that Comscore’s numbers underrepresented an audience. These are very easy cases to address, but their increasing frequency made me wonder if the “Myth of Static IP” – that IPs will become equivalent to Users – was undergoing some sort of revival. I’d like to run through a few facts about what IPs represent with respect to Unique Visitors and show that IPs are a poor proxy for UVs even in a broadband world.
It's common knowledge that ISPs rotate IP addresses. The IP address I have today could be the one you have tomorrow. With a dial-up connection, you get a new one each time you connect. With “always-on” broadband connections, the IP addresses are more durable, possibly even static over long periods of time. As consumers began to adopt broadband technologies in earnest several years ago, a myth arose that IPs would soon become a good proxy for individual households.
We looked into the matter when we were researching cookie deletion in 2007, finding that fewer than half of US machines used the same IP during a one month period, and individual machines used an average of nearly 6 IPs. We asked around the industry to verify our findings and found very little surprise. The general tone of the response was: “You didn’t know that already?” We moved on with our cookie research.
But given the apparent renewal of interest in using “Unique IPs” as a measure of “Unique Visitors,” I thought it was time to update the IT research and make a few facts known. All figures are built from Comscore’s research panel activity in June 2008.
US IP Address Counting Inflates counts by more than 5 Times
In June 2008, nearly 60% of all machines used a single IP address, which is slightly more than when we initially saw in early 2007. However on average, US machines used 5.7 distinct IP addresses in the month, which means that 40% of machines that change IP addresses during the month do it with great frequency. What’s most striking about these distributions is how the multiple-IP machines drown out Single IP machines. Of the distinct IPs used by US machines in the panel, just 10% came from Single IP machines. Only 1 in 10 observed IPs satisfies the Myth of Static IP.
When you break the results down by location of use, its clear that the problem is significantly less at Work where broadband is the norm. But even at work, where 75% of machines use a single IP, these single IP users account for just 28% of the total IPs that a publisher would see. If we use work to represent how IP counting would perform in a broadband world, its pretty clear that IP Counting will continue to be too fraught with issues to represent Unique Visitors.
(Note: we are setting aside for the moment the core workplace IPs problem: Multiple users appear with the same IP address to publishers. This situation presents the opposite problem for people counting IPs. A single IP represents many individuals. We’ve heard this fact cited as evidence of why IP counting is actually conservative. If anything this fact renders the problem unsolvable for publishers – there is error in both directions, but you don’t know which direction is any given case.)
Broadband Is Not Static IP
The root of the Myth of Static IP is the dawning of the broadband network with “always-on” connections. It was expected that with more broadband adoption, we would get more reliable IP counting. This assumption also falls short. When we break downthe U.S. Home total by connection type, it’s clear that only Cable approaches the static IP levels seen at work. Dial-up predictably uses piles of IPs for each machine, and DSL falls in between. Only 30% of DSL machines used a Single IP address during the month, and the average DSL machine used more than 10 IPs. Of these residential technologies, only cable is even close to living up to the Myth. But even Cable suffers from the core inflation issues, averaging 2.7 IPs per machine.
IP Inflation Issue is Amplified Globally
If you pull back and look at the IP Counting issue on a global scale, it becomes clear that the Myth of Static IP is simply not true. Outside of North America, average number of IPs per machine in a month are roughly double those in North America, ranging from 10 in Western Europe to 15 in the Middle East and Africa. Worldwide, only 4 in 100 IPs observed in June came from a Static IP machine.