Take a guess: What percentage of US frequencies are controlled by the Federal government (represented by the NTIA), and what percentage is shared with non-Federal  users, who are under FCC jurisdiction? And what’s the remainder, devoted solely to non-Federal users?
My intuition, for what it's worth, was completely wrong. I thought the Fed/non-Fed split was roughly 50/50, with a bit (say 10%) being shared. As I pointed out in my recent post about partitioning Fed and non-Fed allocations, the amount of sharing should be easy enough to establish. It turns out that Peter Tenhula of Shared Spectrum Company has done a lot of work on this , and he pointed me to the FCC’s spectrum dashboard where one can download an XML snapshot of the allocation database (currently the API only covers the range 225-3700 MHz).
The answer? It depends on the frequency range and how one counts , but very roughly 10% is Federal, 40% shared, and 50% non-Federal (i.e. FCC) only.
Here's a picture (click it to enlarge); an Excel file with my analysis is here.
Update, 21 October 2011: Perhaps the flaw didn't lie with my intuition, but with my interpretation of the data. A senior FCC person has pointed out to me that many supposedly "shared" allocations are, to all intents and purposes, controlled by Federal agencies, and non-Federal (i.e. FCC-managed) services are present only on sufferance, if at all (e.g. 220-2290 MHz); or "sharing" only occurs both Federal and non-Federal entities used the same service (e.g. air traffic or maritime radar). So the question still stands, pending further digging: how much sharing (for various values of "sharing") is really going on?
The implications for my partition proposal is that there’s a lot more opportunity – or, seen another way, a lot more work to be done! – in untangling shared frequencies. The jurisdiction problem means that the intensity of use of frequencies shared by the FCC and NTIA is probably very low: since there’s no-one to arbitrate a fight between them, it’s better not to get into a fight in the first place. The best way to avoid conflict is to be over-conservative about protection rules for services sharing a band.
One might argue that the fact of such extensive shared jurisdiction means that partition isn’t necessary; it’s worked so far, so why change it? I’d respond that we haven’t really seen much real sharing, because utilization has been low, and guard bands large. It’s like a kid on rollerblades and a couple of seniors sharing the boardwalk at two o’clock in the morning; there’s so little traffic, they might as well each be there on their own. As everyone starts using radios for everything, we’re heading into an era analogous to the kid and the couple crowding onto the boardwalk along with thousands of others on a sunny Saturday evening.
The extensive sharing of Federal with non-Federal users also points to an incentive for both communities. Untangling use will give the Feds the capacity for bandwidth-intensive new applications like live streaming video, and will give the FCC the authority to optimize sharing between commercial players without “people will die if you do this” trump card always lurking up the NTIA’s sleeve.
 This is sometimes referred to as “commercial use spectrum,” but that’s a misnomer since the non-Federal frequencies managed by the FCC includes public safety uses.
 Peter has compiled an excellent list of sources:
Karl Nebbia, Director, NTIA Office of Spectrum Management, presentation to the Commerce Spectrum Management Advisory Committee (CSMAC), December 9, 2009, available at http://www.ntia.doc.gov/advisory/spectrum/meeting_files/225_3700MHzPresentation.pptx.
Lawrence E. Strickling, Assistant Secretary of Commerce for Communications and Information, keynote speech delivered at the 11th Annual International Symposium on Advanced Radio Technologies (ISART), Boulder, Colorado, July 28, 2010, available at http://www.ntia.doc.gov/presentations/2010/ISARTkeynote_07292010.html.
Spectrum Policy for the 21st Century – The President’s Spectrum Policy Initiative: Report 1 Recommendations of the Federal Government Spectrum Task Force and Spectrum Policy for the 21st Century, National Telecommunications and Information Administration, U.S. Department of Commerce (June 2004), available at http://www.ntia.doc.gov/reports/specpolini/presspecpolini_report1_06242004.htm#_Toc74447274.
Press Release: U.S. Department of Commerce Takes Major Step Towards Unleashing the Wireless Broadband Revolution, available at http://www.ntia.doc.gov/press/2010/SpectrumReports_11152010.html.
Factsheet, available at http://www.ntia.doc.gov/reports/2010/SpectrumFactSheet_11152010.pdf
P.C. Roosa, Jr., Basic Elements of Spectrum Management: How the Spectrum is Shared, August 1992, available at http://www.ntia.doc.gov/osmhome/roosa5.html.
 A hertz isn’t a hertz isn’t a hertz; lower frequencies are usually valued more highly, since they offer better propagation and thus coverage. Further, 500 MHz covers pretty much the entire UHF TV band of 300-700 MHz, whereas it’s only a fraction of the 60 GHz band (57-64 GHz, i.e. 7,000 MHz) that’s about to be opened up for multi-gigabit wireless systems. It might thus make sense to weight a hertz at low frequencies more highly than one at high frequencies. I use a "frequency-adjusted bandwidth" of bandwidth divided by center frequency, as used in the 2005 Ofcom Spectrum Framework Review (PDF), "a 1MHz allocation at 100MHz is given equal weighting to a 10MHz allocation at 1GHz". I multiply the ratio by 1,000 for ease of interpretation, so that 1 MHz at 1 GHz counts at 1, not 0.001. For frequencies lower than 1 GHz, a MHz is "counts for more," and “counts for less" above 1 GHz. So for example, 10 MHz at 500 MHz counts for 20 = 1000*10/500, whereas at 5 GHz it counts for only 2 = 1000*10/5000. This is equivalent to plotting the bands on a logarithmic scale, as is done in the allocation chart.