This review has been edited to correct some misstatements pointed out by the author. I was working from a prepublication version that did not have all the end-notes, nor a reference to the website. Moreover, the author's comment to this review adds some useful material. On the basis of that, I would raise my rating from three stars to three and a half if that were allowed, but my basic opinion has not changed.

This book is likely to teach you some fascinating and useful material, but I can't recommend it wholeheartedly because it may drive you crazy as well. The basic idea is simple. Experts know a lot, but are bad at making predictions about human affairs. Simple models based on quantitative game theory are more accurate, and even when they're incorrect they expand your thinking in useful ways. Moreover, these models allow you to simulate alternatives and generate outcomes as good or better than the best human strategists can achieve.

To evaluate this book, it's useful to separate that claim into two parts. I'm a quant, and therefore I think it's pretty well established that you make better decisions by asking experts what they know and letting a computer trace the logical implications than by following the experts' recommendations. I also accept that simple quantitative models do remarkably good jobs, and are only rarely surpassed by complex qualitative analysis. Anyway, if you don't accept those positions, there's no point in even opening this book. So to me and probably to you if you're still reading, asking experts simple questions with answers on a scale from 0 to 100 and combining the results in a reasonable way, is an excellent approach to most decisions. Call this the basic quant position.

The author goes further than the basic quant position in three respects. First, he makes much stronger claims for the superiority of his approach. Second, he advocates specific game theory analysis that involves complex modeling, as opposed to simple rules such as guessing that the outcome will be somewhere near the average opinion weighted by salience and power. He doesn't justify his methods as practical shortcuts that seem to work, he repeatedly claims that they are backed by science and logic, unlike alternatives. Finally, he goes beyond prediction to use his model to engineer outcomes. This is quant-on-steroids. It made me recall that John Nash, the most important game theorist with respect to this kind of work, maintained he was Emperor of Antarctica.

To paraphrase Dizzy Dean, it ain't megalomania if you can back it up. This is where it starts to get irritating. In the preface, before you get to page 1, he writes, "I've been predicting future events for three decades, often in print before the fact, and mostly getting them right. . . . I have made hundreds, even thousands, of predictions--a great many of them in print, ready to be scrutinized by any naysayer." This guy makes extraordinarily bold claims about his quantitative prediction ability, and he doesn't keep track of his record? He can't even recall within a factor of ten how many predictions he has made? Who decided he was "mostly right?" And why is anyone who wants to know the record a "naysayer?" Even if you did all the work, unearthed all the printed predictions (he only references a few in the book) and found he was zero for 200, he could just say he got thousands of other ones right, ones you didn't find. This sounds more like a Nostradamus defender than the "science" he's always claiming.

To bookend that frustration, by the end of the book, despite frequent promises, he has not revealed his model! He does have a version on-line that allows you to play around with it, and has more details on how it works, but still no clear, top-down description. In the book, you get hints and bits and pieces, but no clear explanation of how he arrives at his predictions. And this is not the only broken promise, there are frequent comments that he will "go into this further" or "provide more details," later; I can't find one example where he follows through. On the other hand, there are a few facts that get repeated far too many times.

Those two things would be enough for a lot of people to conclude he's a fraud. But there's an awful lot of good, clear, insightful analysis in between. He gives examples of political and diplomatic predictions he has made, discussing the inputs and basic form of the analysis. There are accounts of corporate and legal struggles where he maneuvered to an outcome favorable to his clients. He also applies the methods to history, to ask what might have happened. This is all fascinating stuff, and the data and conclusions speak for themselves. They show plausibly that this approach could work, it is practical to implement and it leads to conclusions that are surprising, but can shown to be logical. Without a lot more details these stories don't prove the model works, but they represent a coherent claim that it does.

However, this brings us to another problem. Some of the accounts are not credible. An account of how he maneuvered a fifth-choice candidate into a CEO job requires us to believe the board of directors split into five groups of three that agreed among themselves on the exact preference order for the five candidates, that these generated a cyclic preference order that included all five candidates (a mathematically possible but unlikely result only previously observed in game theory textbooks), that all this information was known with certainty beforehand and that none of the board members was smart enough to consider coalition-building or voting a second choice when it was clear a first choice vote would be wasted. Moreover, the Rube Goldberg scheme that worked seems far less promising than simple politicking to either build a coalition or change a slight preference.

This is the least credible account (unless you include the million dollars he was offered by Libya to engineer the removal of Anwar Sadat from power in Egypt, or the 10% of Zaire dictator Mobutu's external wealth offered to keep him in power), but none of the stories include basic information to allow fact-checking. In some cases the need for confidentiality is clear, but why no names of the government officials who hired him, or the partners at Arthur Andersen? Why is he protecting the agents of Libya and Mobutu? Does the brokerage firm that bought his advice in 1992 still insist on remaining anonymous? (And does it even still exist?) Some discretion is understandable, but this book reminds me of the fictional spies who have all labels removed from their clothing and possessions.

The final irritation is only one account of a missed prediction is given, and it is explained implausibly. The author predicted Hillary Clinton's healthcare plan would pass in 1994. He claims that the outcome was changed by the Rostenkowski scandal. That's hard to believe, since Rostenkowski resigned from his leadership position before the first bill came to Congress. Rostenkowski was not a strong supporter of healthcare reform. There's no doubt that his political skills would have been useful, had he chosen to push the plan, and the scandal did weaken the Democrats in general, but many other things happened that seemed to be at least as important. So either the prediction was dependent on lots of unpredictable events, and therefore should have been given as a probability distribution instead of a point estimate, or Rostenkowski was special, in which case the prediction should have been healthcare will pass or fail based on how Rostenkowski does. And why was the prediction not updated as the scandal worsened?

I know this is a long review, but I've only covered some of the bigger irritations. If you're an easy-going, tolerant sort who wants to learn some important practical and theoretical aspects of prediction, by all means read this book. If not, you might want a blood pressure check before you attempt it.