A few days ago, I read Chris Anderson’s new piece about the aggregation and interpretation of data will make the science obsolete. I said I can’t agree with his ideas but he’s great at selling them. How fitting that this week, Harvard Business Review featured this article disproving The Long Tail. And the reason why Chris Anderson responds is because the article was written by the very person who helped him with the research for his book.
As demand shifts from off-line retailers with limited shelf space to
online channels with much larger assortments, is the tail of the sales
distribution getting longer and fatter?
My colleague Felix Oberholzer-Gee and I studied this question. In
particular, we looked at weekly sales of home videos as reported by
Nielsen VideoScan from January 2000 to August 2005, focusing on a
random sample of nearly 5,500 titles. Using econometric models that
control for a number of possible concomitant trends, we found that
sales did shift measurably into the tail: The number of titles that
sold only a few copies almost doubled for any given week from 2000 to
2005. In the same period, however, the number of titles with no sales
at all in a given week quadrupled. Thus the tail represents a rapidly
increasing number of titles that sell very rarely or never. Rather than
bulking up, the tail is becoming much longer and flatter.
Chris Anderson’s counter argument is that even though the “head” of some of these online stores are more valuable than the tail, that head is a “tail” when you compare it to a brick and mortar store that might carry a less diverse inventory. (Damn that sounded clumsy.)
I get what he’s trying to say but it’s flawed because he’s making an apples to oranges comparison. He argues that the denominator doesn’t matter when you’re comparing the catalog of the two stores. But I argue that the percentages do matter because his primary thesis was that there was more money under the long tail of the curve than the head. And if you want to support that with data, you have to compare the head of a curve to the tail of the very same curve.
If not, then the long tail argument is simply reduced to a discussion about variance/pooling resources and that’s not interesting at all.