I don't remember how I came across Wisdom of Crowds by James Surowiecki but the moment I read the introduction I knew I had to have the book.
The book starts with the story of a famous scientist Francis Galton stumbling upon a weight-judging competition at a Stock Fair. An ox was put on display and passer-bys were asked to look at the ox and give their best estimates of its weight when "slaughtered and dressed". The result was astonishing - while individual estimates varied a lot, the average of the 787 estimates came to be only 1 pound off the real weight. In statistical terms that means that human ox-weight-estimation ability is unbiased - it has no systematic error.
So what is so exciting about research on human estimates and their properties? They may shed some light on some important things - most notably the institute of democracy. In principle, democracy has any chance of functioning only if people can give correct estimates in areas that they are not experts in.
So it would seem to me that the book on wisdom of crowds should investigate the question of systematic errors in human judgement to its fullest. To my utmost frustration, it did not. The term "systematic error" is not mentioned even once in the book. To be fair, the book talks about other conditions conducive to good group estimates and decisions - diversity and independence but the very basic issue of systematic errors is completely skipped over.
Galton's experiment notwithstanding, people do make systematic errors, in particular, when voting on economic issues - see new book by Bryan Caplan The Myth of the Rational Voter: Why Democracies Choose Bad Policies. Some of his papers are also available online.
Subscribe to:
Post Comments (Atom)
3 comments:
Илюшка, так может он не делает выводов просто потому, что их нет?
I am frustrated with the book not because it doesn't provide the ultimate conclusion on whether human estimates (of anything) are biased. There can't be any such general conclusion. It should be a question of emperical research and probably will show different levels of bias in different areas and estimation tasks.
Rather I hated that individual human perception was treated as something mystical that just somehow delivers good estimates. I prefer to think about each estimation task as excersise of a particular human information-processing mechanism with its own statistical properties. The most important property if we want to use individual estimates in group averages is "systematic error". So when I didn't find any discussion on this in the book I had to go find it somewhere else. And to my satisfaction I found some better treatment in Coplan's work.
Thast is why LJ would be so much better. I would know right away that you replied by receiving an email. And your post would appear in My Friend's page. Lets move you?
Post a Comment