Enrico Fermi was one of the last century’s greatest scientists and he created the world’s first nuclear reactor. One of the talents of that generation of scientists and engineers, driven by necessity, was the ability to do what we in the UK call “back of the fag packet” calculations. Of course good scientists and engineers of today should still be able to use this technique. I’m sure Fermi used these skills every day – and he appreciated the value of them as well, challenging his students with “Fermi problems” to get the cogs of their brains working. A typical Fermi question was “How many piano players are there in Chicago?” A sensible person with some rational assumptions and general knowledge can come up with a pretty good (good enough?) answer with no further data. The purpose of such an answer is to give a benchmark perhaps against an experimental result. An experimental result that’s way out, then needs reassessing or investigating – its either wrong or shows something very interesting is going on. All in all, thats a sound scientific and engineering technique and to some degree should be a a simple stage for almost every scientific and engineering to check for gross errors. With a knowledge of a few basic constants and some common sense a lot can be achieved, and inefficiencies avoided,
But the principles of this technique can also be used for intelligence analysis. Many RFIs might prompt a Fermi question. And for those of you who ever face me as a job interviewer for an analytical post at IMSL there’s a big hint there. For many of today’s analysts such a question of “How many piano tuner’s are there in City X?” would be a challenge of finding open source data – many analysts would google “piano tuner Chicago” and work from there . That would take some time, and a little effort (Try it if you must). But what should also be prompted is the context of the question. Does the person raising the RFI need to know the exact number of piano tuners in the City or is an answer within an order of magnitude what they need? In which case use your Fermi methodologies rather than create a spreadsheet and start populating numbers. The Fermi answer will then, hopefully provide you with a gauge as to the sort of answer that might be expected and prompt you to look really hard at the data if it doesn’t fit the expectations set by your preliminary assessments.
More deeply, Fermi techniques can also lead you to derive answers from related data – so, for example an estimate of car dealerships in a city might lead you to a figure for the number of new car sales in that city. At IMSL we pride ourselves on the fusion of different OS data sets to create new insights, and quite often its these sorts in inferences that can become useful.
Now there is one downside, which every knowledgable analyst will flag in that interview I mentioned, I hope. There is the potential that Fermi answers could , in some circumstances where you aren’t as clever as Prof Fermi, lead you down the path of confirmatory bias. If you are looking for answers that fit your pre-conceptions, there’s a small danger that the methodology might encourage a bias. But there’s an equal chance at least that the methodology will cause you to challenge your own biases. A good analyst might also have a look at what Wolfram Alpha has to say about the data on the piano tuner question. See you at the interview!