one would have to get a rental record from say a physical Blockbuster store that has tended to have about the same inventory size over time, and then compare that distribution with netflix [for any year].

as far as I know, there is nothing in the long tail phenomenon that says, “netflix will be more-and-more a long tail system over time”. maybe netflix was a long tail phenomenon **from the very beginning** and every year of their operation represents that same reality. I would argue that is the case. I would say the analysis on this website tends to support that.

I would say, this shows, it is tricky to formulate and measure the exactly right hypothesis. and doing so is an iterative process. it is not at all clear what exactly the “long tail phenomenon” actual implies, because it was never stated in a strictly mathematical form. I suppose the distribution diagram that ilya posts at the top of the article comes close.

it seems for someone to better articulate the long-tail phenomenon, we need a way to characterize the distribution of the long tail. and fit the distribution to the model. I am not sure what distribution the netflix data fits, it is probably a power law I would guess. [I have posted this speculation on their forums but nobody has yet confirmed it].

assuming it is a power law, but that also maybe physical stores like blockbuster also follow a power law, then what the long-tail phenomenon is probably asserting is a difference in the **exponent** involved in the power law. if so, we could say ‘x’ in some range is “long tail” and ‘x’ in some other range is “conventional”. these mathematical properties are just on the verge of being articulated.

First off, just reading the responses I am amazed how split the views are. I did not mean to try and disprove the Long Tail theory as a whole, nor did I delve into other interesting areas such as consumer demand. My disagreement with Chris lies in the very simple distinction of choosing the appropriate axis. Namely, absolute vs. relative.

To be honest, I am still not convinced that looking in percentages is a statistical fallacy. How so? If the axis is growing, then the only way to compare two distributions is to normalize each one. At least that was my reasoning.

Graham: I agree, I do not follow that argument either. Also, I took the liberty of fixing the formatting of your comment - hopefully I did not modify it’s meaning or intent. And if I did, please let me know!

But since those 100 titles now represent a smaller percentage of a much larger inventory (from 2.5% to 0.1%), it looks like the market is indeed more concentrated on the hits, even though such percentage analysis is meaningless in the real world where absolute numbers are everything.

So percentages are meaningless.

Fascinating.

This is impressive analysis, but I think you’ve fallen victim to a common statistical error in basing your conclusions on percentages of the total inventory, rather than absolute numbers (which is the only thing that matters in the real world.) I’ve responded here.