Tuesday, August 27, 2019

Dept. of Physicists Can Do Stuff: Brexit!

Dominic Cummings on how Vote Leave won the Brexit campaign. (Video should start at 13m30s.)

Dom hired a team of physicists and data scientists who

1. Studied the literature on elections (i.e., entered a well-established subject, studying it de novo, applying real horsepower), figured out which results/beliefs in that field were likely correct (much found was incorrect)

2. Adapted results to the job at hand (Brexit referendum) and invented new techniques for applying them

3. Built a new platform, wrote new code, executed in real time, and won a huge electoral victory against all odds.

Of course, this is the age old story of physicists invading/creating other fields: early computing, electrical engineering, molecular biology, computational biology, quantitative finance, high frequency trading, etc.

This victory will have historical reverberations that are still playing out.
13m35s: ... we had to take risks and we had to do things in a slightly new way so one of the basic things that I did was I brought in a team of physicists who essentially looked at campaigning from complete first principles and what they did was they went they simply scanned around the world and they said what studies have been done on issues of turnout and persuasion that actually have good maths behind them to support and have been replicated and we can actually have confidence in and they basically filtered all when through filtered them all out and came back to me in the team and said here is a small selection of things actually high quality or reasonable quality work which you can rely upon and here are the principles that you can see in these studies that have been replicated with randomised controlled trials and whatnot in the States

we basically created a checklist of what these things were and we built the communications team around trying to exploit each of these elements which the physicists found they also constructed models to help direct resources on the ground campaigns to wedge where to send your activists and the digital campaign how do you actually do that in a in a scientific way and essentially you had streams of data coming in from all sorts of different ways the website email on the ground canvassing a social media blah blah all of this stuff could be traditional polling all of the stuff coming in and you had the data science people sitting at the heart of the operation and essentially taking our core messages and just running experimentally a whole bunch of different things on Facebook and elsewhere and then figuring out what what things and what things don't work and we started off with relatively small amounts of money just to run this experimental process

another thing which which I'll go into a little bit of detail because it's from perhaps of interest regarding this election is we did a new kind of polling so I'm sure all of you know the polling methodology used throughout the world essentially the same system that was invented in the late 1930s and the idea of it is yo you take roughly speaking a thousand person sample and if it's random a representative then you can rely on the mathematics of the normal distribution and the famous bell curve and you that should give you a pretty accurate picture of what people think for various reasons that is becoming harder and harder to do happy to answer questions about why that is but leaving that aside what the physicists said was this is actually not the way that you would invent polling if you were going to invent polling now the way actually to do it is take massive samples of hundreds of thousands of people ideally actually millions of people but say hundreds thousands people and then use machine learning and you will actually have a system which is faster cheaper more accurate and never has another great advantage which we exploited which is that if you do these very large sample surveys you then have sub sample you can define the demographics that you interrogate yourself and what we did was we basically use the exact same categories infer demographics that Facebook uses for its digital advertising platform so we sucked in data on the precise same basis that Facebook marketing allows and then we had therefore large sub samples of the overall polling samples which you could actually rely on and then you could take that data and plug it straight back into Facebook so you could say for example we will target women between 35 and 45 who live in these particular geographical entities who don't have a degree or who do have a degree or whatever it's after cetera

and because you've got very large samples you can actually get useful information on those kind of relatively small breakdowns so we did all this and we as I said we essentially ran a whole series of experiments based on what we found at the conventional polling in the focus groups out in the digital world and then filtered what worked and then we held back almost all of our budget and then we basically dumped the entire budget or in the last ten days...
See also Dept. of Physicists Can Do Stuff: Gene Sequencing, Harold Brown, Ashton Carter.

More Dom.

How Brexit was won, and the unreasonable effectiveness of physicists:
The scale of ... triumph cannot be exaggerated. He ... had brought about a complete transformation of the European international order. He had told those who would listen what he intended to do, how he intended to do it, and he did it. He achieved this incredible feat without commanding an army, and without the ability to give an order to the humblest common soldier, without control of a large party, without public support, indeed, in the face of almost universal hostility, without a majority in parliament, without control of his cabinet, and without a loyal following in the bureaucracy.


On the eve and day of Brexit I happened to be staying at the estate of a billionaire hedge fund manager, which hosted a meeting of elite capital allocators. At breakfast, more than half of these titans of capital were in shock ... Markets were down 8% or more and my host asked for my view. It will play out over years, I said. No one knows where this is going to go. The market is oversold and it's a buying opportunity. So it was.

No comments:

Blog Archive