William Collins Noise

I desperately wanted to like this book, but every page turn felt like dull slap in the face. Humans make poor, inconsistent decisions and are easily swayed. The end. Save yourself £16 and move on with your life.

For many Kahneman is a God: and one that was awarded the Nobel Memorial Prize in Economic Sciences for his groundbreaking work in applying psychological insights to economic theory, particularly in the areas of judgment and decision-making under uncertainty. Thinking, Fast and Slow was the book that everybody had to have but, as was the case with Hawking's A Brief History of Time, one suspects that many copies went unread. Kahneman himself, in a move akin to God saying that he had difficulty sticking to the Ten Commandments said in a recent interview that "my own experience of how little this knowledge has changed the quality of my own judgement can be sobering." There were two problems with Thinking, Fast and Slow - firstly, the transition from fast to slow was unquantifiable and second, it seemed to propose an ability on the part of the average punter to tap into their unconscious. Moreover, many of those that read the book were great at spotting biases in other people rather than in themselves.And so to Noise, a book, we are told that is designed to offer suggestions for the improvement of human judgement. As for Noise itself we are told in the book that that noise is about statistical thinking. We are also told that noise is a distinct source of error and that "the scatter in the forecasts is noise" and, that whenever we observe noise we should work to reduce it. However, we are also told that noise is invisible and embarrassing.Noise occurs because people are idiosyncratic; they inhabit different psychological spaces; their moods are triggered by a unique set of contexts - they see and respond to the evidence in different ways. Not to mention their unconscious response to particular cues. (In many respects - seemingly the same things that trigger biases, and we are told rather confusingly that "psychological biases create system noise when many people differ in their biases.") We enter a convoluted vortex - biases cause noise - where there is noise (invisible) there will surely also be more biases at work - the two, it seems, exist in relationship that is characterised by their mutual and continuous interruption of each other. And there is actually no clear sense given as to how one should go about unpicking them.Surprise surprise the authors pay passing homage to prediction markets, of which they say; "much of the time prediction markets have been found to do very well.") Prediction markets, in the wild (outisde of organisations) have not actually performed very well at all - because they lack insiders and do nothing more than aggregate noise. Their record on political events over the past ten years has been terrible (In the recent Chesham and Amersham By-Election in the UK, for example, the Tories were trading at 1.17 on the Betfair Betting Exchange as Polls opened - they lost). A better example, in the context of noise would have been horse racing betting markets - which contain lots of noise and bias, but which display a consistent ability to be predictive - because of the presence of insiders, who cancel out the noise.Sadly it seems that we have gone back twenty years, to the notion of the jar of sweets and the benefits of aggregating independent judgements. In a nutshell, this book is about 380 pages too long.

An average book with a lot of hype .. common knowledge packaged to sell books .. effort to monopolize thousands of years of human wisdom as one's own invention, with some colorful words

I really have no idea who the intended audience was for this book: the authors really, really dumb it down, to the point of explaining what variance is over several pages of prose. We did not all fail high school.At the same time, they bring into the discussion some serious tools you won’t even meet until you get to graduate school in statistics, like the “percentage concordant,” which is not some type of supersonic airplane, but a rank correlation type of measure, and even provide a mini-table to move you from percentage concordant (PC) to correlation. The table, by the way, is bogus in the absence of context, as percentage concordant is a construct that I’m willing to bet relies heavily on assumptions that go unmentioned here.The chapters end with summaries, which was OK for Thinking Fast and Slow, but a bit of an insult when the subject matter is so plain.The style is pompous and paternalistic.System A and System B are parachuted in, but (i) they’re barely explained (ii) that’s a theory to explain bias rather than noise (and invite a celebrity author to the proceedings)Most annoyingly, terribly little ground is covered in this weighty tome. Gun to my head, I could probably get it all down to one page. Let me try:1. Noise is just as bad as bias in terms of messing up your results2. A good way to measure how bad your results are is the mean square error3. Composition of Mean Square Error:• Mean square error is made up of Bias and Noise• Noise is made up of Level Noise and Pattern Noise• Pattern Noise is made up of Stable Pattern Noise and Occasion Noise• Level Noise is the kind of noise that comes from the fact that some judges are harsh and some are lenient, so two guys who did the same crime could get very different punishment.• Pattern Noise is the kind of noise that comes from the fact that a judge may have a daughter, making him less harsh on young women that remind him of his daughter. He could be a harsh judge who is less harsh on young women who remind him of his daughter; or he could be a lenient judge who is extra lenient on young women who remind him of his daughter.• Occasion Noise is the kind of noise that comes from the fact that judges are harsher right before lunch. Same judge, same crime, same perpetrator, different outcome, because it was a different occasion4. If you ask people to measure something independently from one another, the more the merrier; but if they talk to each other first, then they will amplify errors for a variety of reasons that lead to groupthink5. Machines beat people when it comes to cutting noise6. In the quest to limit noise, people can fight back by sticking to simple rules7. We humans like to build stories after the fact to explain what happened; they’re usually bogus: statistical explanations beat causal explanations8. Bias can be the source of noise: inconsistency in bias is noise9. Noise can arise when you’re told to rank things on a scale; to cut noise, it’s better to go ordinal than cardinal10. To improve judgements you need (i) better judges (ii) a decision process that aggregates in a way that maintains independence among the judges (iii) guidelines (iv) relative rather than absolute judgements11. There is a place for intuition: it’s got to be brought in at the very end, after all the mechanical work has finished12. There actually is a place for noise: when people are bound to game the systemRead something else!

It is always a warning sign, when a household name (Kahneman) joins forces with others (Sibony and Sustein) to publish a new book. Especially so, if the new book is a look-alike of the old, very successful Thinking, Fast and Slow. Letter type, size and chapter structure are all the same.Yet the result is disappointing. The household name sold us a book that, as a stand alone publication, would be a likely (or very likely) failure.The book is far too long; arguments are repeated from Thinking, Fast and Slow; fresh evidence/new ideas are far and few.Very sad. Hope this triumviratus will NOT write any more books.