The Backlash, Chapter 1

Chris Anderson
The Long Tail

July 25, 2006

I'll take it as a compliment that I now warrant a proper Wall Street Journal takedown for crimes of...well, I'm not quite sure what the crimes are. But Lee Gomes has tried mightily to find flaws with the Long Tail theory and deserves a response of some sort. I have no doubt that there are many parts of my analysis and data that could be improved. Unfortunately, Gomes, in his haste to find them, stumbles over statistics and more, and in the end simply makes a muddle of what might have been an interesting debate over the magnitude of the Long Tail effect.

He writes [ http://online.wsj.com/public/article/SB115387606762117314-Inp5lUxHwVDwS_SJv5zaQShPXlE_20070726.html ]:

In the book's main sections, Mr. Anderson writes that as things move online, sales of misses will increase -- so much so that they can equal or exceed the sales of hits. The latter is the book's showstopper proposition; it's mentioned twice on the book's jacket.

I was thus a little surprised when Mr. Anderson told me that he didn't have any examples of this actually occurring.

First, the book doesn't claim that there are any cases where sales of products not available in the dominant bricks-and-mortar retailer in a sector (my definition of "tail") are larger than the sales of products that are available in that retailer ("head").

What it does say is that the current data at Rhapsody, Netflix and Amazon show that the tail amounts to between 21% and 40% of the market, with the head accounting for the rest. Although I don't discuss this in detail in the book, in the case of Rhapsody, the trend data suggests that the tail (as defined above) actually will equal the head within five years. Which is why the language Gomes cites from the book jacket is actually all phrased in the future conditional tense ("What happens when the combined value of all the millions of items that may sell only a few copies equals or exceeds the value of a few items that sell millions each?"). I asked him to quote the jacket copy in full context, but it apparently wasn't convenient to his thesis to do so, so he didn't.

Gomes continues:

Mr. Anderson told me the lack of an example of misses outselling hits doesn't diminish his basic point, which he said is simply that the role of the tail "is big and getting bigger."

By Mr. Anderson's calculation, 25% of Amazon's sales are from its tail, as they involve books you can't find at a traditional retailer. But using another analysis of those numbers -- an analysis that Mr. Anderson argues isn't meaningful -- you can show that 2.7% of Amazon's titles produce a whopping 75% of its revenues. Not quite as impressive.

Sigh. Gomes was determined to make this point, even after I and others pointed out the statistical fallacy at the core of it. As I wrote in this post [ http://www.thelongtail.com/the_long_tail/2006/04/an_apparent_lon.html ], trying to define "head" and "tail" in percentage terms is meaningless in a market with unlimited inventory, because the denominator can grow infinitely large. Let me give you an example of why this doesn't work:

Let's say you have 1,000 items and the top 100 (10%) account for 50% of the sales. Then you add another 99,000 items to the catalog, and the sales of that top 100 fall to just 25% of the total, while it takes another 900 items to make up the next 25%. I would say that demand has shifted down the tail, because those top 100 items have dropped from half the market to just a quarter of it and the rest of the demand is spread over more items.

But by Gomes' math, we've gone from a market where 10% of products make 50% of the revenues to one where 1% of the products make 50% of the revenues--in other words, it's become more hit-centric. I think this is simply a misunderstanding of basic statistics, and I'm disappointed that Gomes, despite many emails from me and at least one economist to him on this point, chose to simply say that I don't agree with that approach (but not why).

Finally, a very annoying point. Gomes writes:

Other economists, of course, are looking into these same questions, though some seem to be reaching far more restrained conclusions. Harvard's Anita Elberse, whom Mr. Anderson said was a consultant during his two-year research project, studies the video sales market, both online and off.

She said in an email that her work to date shows a "slight shift" toward the tail. But she also noted "a rapidly increasing number of titles that never, or very rarely, sell," which suggests "it is difficult for content providers to profit from the 'tail.' "

As Professor Elberse told Gomes, she was only describing Nielsen VideoScan data, which is almost entirely taken from bricks-and-mortar sources. The Netflix data, which was the basis of the Long Tail analysis that she and I worked on together, tells a very different story (Elberse's terms of data access don't allow her to share that data; my terms allowed me to share what I published in the book). We both urged Gomes to make clear that the "slight shift" measured didn't refer to the Netflix data that was at the core of the book's conclusions. But he chose to make the point he wanted to make.

I'm actually quite an admirer of Gomes' work and he certainly did do a lot of research for this piece. But he started off with the wrong end of the stick (looking at the market in percentage terms, which doesn't work because the definition of "head" keeps changing) and sadly wouldn't let it go. As an editor, I've seen this happen before and we try to watch out for it. But sometimes the lure of the gotcha is too much to resist.

Posted by John Dodds

July 26, 2006

In your example, you add 99,000 items and they collectively take 25% of the sales - that's a big assumption backed up by nothing and I think shows the difficulty of dealing with hypothetical examples. Gomes's strongest points arise from his research into actual figures for Rhapsody and Ecstat which, if correct, seem to me to be powerful rebuttals to specific claims.

Having said that, it doesn't deny the importance of the Long Tail, it just alters the economics of potential businesses therein and I'm sorry British railways conspired to prevent me attending the London geek meeting enable me to understand it better.

01:45 PM

Posted by Anita Elberse [ aelberse@hbs.edu ]

July 26, 2006

Chris,

I have no intention of inserting myself into this discussion ;-), but I have one small comment:

You say "Nielsen VideoScan data (...) is almost entirely taken from bricks-and-mortar sources." I don't think this is entirely correct. The VideoScan data reflect both offline and online sales, and actually break them down by channel. The breakdown is not as detailed as one might wish in an ideal world, but they do allow one to track whether, say, the share of offline sales go down over time. Therefore, I do think the fact that my colleague and I only observe a "slight" shift is meaningful.

02:26 PM

Posted by Sean [ oneillsoneills@hotmail.com ]

July 26, 2006

Chris, I agree with the earlier comment that:
"Gomes's strongest points arise from his research into actual figures for Rhapsody and Ecstat which, if correct, seem to me to be powerful rebuttals to specific claims."
Your failure to address them in your post is unnerving, as was the failure to post on your blog or include in your book updated numbers from these companies whose surprising statistics launched your quest.

03:00 PM

Posted by Chris Anderson [ canderson@wiredmag.com ]

July 26, 2006

Sean,

I'm not sure what you mean. My research *was* based on the actual Rhapsody figures, of which I have several years' worth. Have you read my book? It's all based on hard data. Indeed, Lee got some of his data from *me*.

Chris

03:22 PM

Posted by Sean [ oneillsoneills@hotmail.com ]

July 26, 2006

Chris,
I refer to the figures for Ecast, a company that told you years ago that 98% of its catalog gets played at least once a quarter.
The WSJ says that the latest numbers show this phenomenon is becoming less pronounced, not more pronounced. Ecast's "quarterly no-play rate has risen from 2% to 12%."
Considering that the Ecast example helped launched your Long Tail quest, I think it's an oversight that you didn't bring to light this trend and discuss it on your blog and elsewhere.
I concede that there are other examples that suggest a long tail phenomenon exists, such as at NetFlix and Slate.com. But I feel the WSJ has brought up a fair point with one of your showcase examples. Am I really being unfair, as a reader, to be disappointed in your lack of curiosity about what has been happening at Ecast, when two years ago you thought the numbers from that company were worth discussing?
Sean

06:36 PM

Posted by Chris Anderson [ canderson@wiredmag.com ]

July 26, 2006

Sean,

To be honest, I wasn't aware that Ecast was still in business. That single data point back in early 2004 simply started my research, which evolved into the Long Tail research that has been the focus of my last two years. This more recent research has been in bigger companies, such as Netflix and Amazon. And, as it happens, Netflix does give the relevent statistic: 95% of its 60,000 DVD are rented at least once a quarter (http://www.netflix.com/MediaCenter?id=1005&hnjr=8). I sent Gomes that information but--surprise--he chose not to use it. So to answer your question, although my research was focused elsewhere, the companies I was following seemed to confirm the original observation that there's at least some demand for nearly everything. You might want to ask Gomes why he didn't mention the Netflix data.

Chris

08:40 PM