Department of Awful Statistics

A reader sends in this note:

I am a former tax lawyer who worked for years on pharmaceutical international transfer pricing cases. The basis for such cases revolves around what share of profits is attributable to a given country. Let me tell you, you will never get an accurate figure, because no such figure exists - all the numbers are purely notional.
As you know, pharmaceuticals are essentially a zero-marginal-cost product. However they have huge up-front costs that include the R&D, clinical trials, setting up manufacturing facilities, regulatory submissions, IP legal work, promotion with thought leaders, etc.  All this work may be done in various countries while benefiting the pharmaceutical company in all countries in which it sells the drug.  Revenues for a given country are generally easy to determine, but costs - the other necessary variable to assess profitability - depend entirely on how you attribute these costs to each country.
There are dozens of plausibly fair and logical ways to attribute the costs. In a typical transfer pricing case both the company and the tax authorities will have good arguments as to how to best attribute them (for this reason, most cases are settled), but they can differ by several billion of dollars of profit on a single drug.
For this reason, any number as to the percentage of pharma profits made in the U.S. should be treated as arbitrary and bogus. It will entirely depend on how the costs were allocated, which will differ from company to company, and may even differ from one company's financial statements to its tax returns.

I find this argument pretty compelling.  So it looks like I got taken, at least in the sense that there's probably no way to come up with an estimate that I would find acceptable.  I wouldn't have put that number in a blog post, because I would have looked for better corroboration, and I'm sorry that I used it when responding on the fly, which I did several times. 

I'm now adding this to my long list of "dark numbers", with the best available proxy being the global sales of New Chemical Entities.  Two thirds of those, not more than three quarters, occur in the United States, versus about a quarter in Europe.  You can argue about what the fixed costs are in various places, but as my correspondent implies, given how much cross border activity there is, the problem seems to be indeterminate, so I'll stick with a number we know.  This doesn't really change my assessment of the problem, since 2/3 is still pretty overwhelming, but statistics matter.

On a side note, the reason I said 80% was a hypothetical in the Washington Post chat is that . . . well, I didn't.  I forgot that conversation, and thought the commenter was referring to this post.  These are the perils of typing thousands of words a week, and also, of getting old.