Why Does My Alexa Rank Jump Around? A: The Long Tail

By now we have all heard of the long tail, which has been used to metaphorically describe just about everything from income distribution to traffic on the Web. Originally coined by Chris Anderson in a 2004 Wired article, it refers to a frequency distribution with a long tail, sometimes also referred to as a power law graph. On the Web we have a very long tail, and it means that there are relatively few sites with a lot of traffic and endless numbers of sites with low traffic... the long tail.

While most of us can claim that we understand the meaning of the long tail, it is still often hard to comprehend how this applies to us. I refer specifically to Web site owners with low traffic who see their traffic rank jump around a lot. I was poking around the blogosphere today and ran into this blog entry, the Alexa Experiment. Over the years I have seen a few others like it. Her complaint is that her Alexa rank jumps around quite a bit... in his case from 3.5 million on day 1 of her experiment to 2.8 million on day 13. She uses this as evidence that the Alexa Rank can't be relied upon.

Forgetting about the relative merit of the Alexa Rankings for the moment, a perfect ranking system, one with perfect information about all sites, would tend to behave in the same way. Why? The long tail.

In any ordinal ranking system, like the Alexa Rank, sites out on the long tail will experience massive changes in rank regardless of their actual number of visits, visitors and pageviews. The reason for the fluctuation is because the farther you go out onto the tail the flatter it gets.

I'll use a non-Web example to explain this principle in action. Let's take every person in the United States of America and rank them based on income. That gives us 300 million people ranked from 1 to 300 million, with the person ranked at #1 earning somewhere in the hundreds of millions of dollars per year, and the person ranked #300 million earning nothing, with the rest of us somewhere in between. Like all long tail distributions there are vastly more people on the tail, earning little or no money, than there people at the head of the graph earning hundreds of millions.

If we examine the person ranked at #50 million, let's postulate that she earned $50,000 per year last year, and that she will earn $50,000 again next year. Question: Will she still be ranked at #50 million next year? No. In a shrinking economy her ranking is going to improve because millions of people earned less. Conversely, in a growing economy her position will fall as millions of other workers earning improve. Her rank jumps around wildly, even though her actual earnings have remained unchanged.

What if the economy stayed steady-state, and our $50,000 earner got a raise of exactly $1, and now earns $55,001 per year. What will that do to her rankings? Will her ranking move up by 1 to #49,999,999? No. The long tail distribution tells us that the farther we go out on the tail the more likely it is that there are others earning the exact same amount as her. In her case it could be hundreds or thousands of people. Earning just one dollar more per year can vault her position in the rankings much farther than you may expect.

The point is that the farther you go out on the tail, the less is required to move up an ordinal rank. In a system with a distribution like traffic on the Web this is especially true. If you are out on the tail and you improve your traffic a modest amount it could improve your rank by a million places or more.

That's the nature of the long tail. It is very flat and moving horizontally is not all that hard until you begin to approach the head. For the folks who are running an Alexa Experiment, I wish them the best of luck. But your time would be better spent finding ways to increase your visitors, visits and pageviews. Your Alexa Rank will follow.