What do you think?
Rate this book
Errors, Blunders, and Lies: How to Tell the Difference
We live in a world that is not quite "right." The central tenet of statistical inquiry is that Observation = Truth + Error because even the most careful of scientific investigations have always been bedeviled by uncertainty. Our attempts to measure things are plagued with small errors. Our attempts to understand our world are blocked by blunders. And, unfortunately, in some cases, people have been known to lie.
In this long-awaited follow-up to his well-regarded bestseller, The Lady Tasting Tea, David Salsburg opens a door to the amazing widespread use of statistical methods by looking at historical examples of errors, blunders and lies from areas as diverse as archeology, law, economics, medicine, psychology, sociology, Biblical studies, history, and war-time espionage. In doing so, he shows how, upon closer statistical investigation, errors and blunders often lead to useful information. And how statistical methods have been used to uncover falsified data.
Beginning with Edmund Halley’s examination of the Transit of Venus and ending with a discussion of how many tanks Rommel had during the Second World War, the author invites the reader to come along on this easily accessible and fascinating journey of how to identify the nature of errors, minimize the effects of blunders, and figure out who the liars are.
In this long-awaited follow-up to his well-regarded bestseller, The Lady Tasting Tea, David Salsburg opens a door to the amazing widespread use of statistical methods by looking at historical examples of errors, blunders and lies from areas as diverse as archeology, law, economics, medicine, psychology, sociology, Biblical studies, history, and war-time espionage. In doing so, he shows how, upon closer statistical investigation, errors and blunders often lead to useful information. And how statistical methods have been used to uncover falsified data.
Beginning with Edmund Halley’s examination of the Transit of Venus and ending with a discussion of how many tanks Rommel had during the Second World War, the author invites the reader to come along on this easily accessible and fascinating journey of how to identify the nature of errors, minimize the effects of blunders, and figure out who the liars are.
169 pages, Kindle Edition
Published May 18, 2017
7 people are currently reading
76 people want to read
Ratings & Reviews
Friends & Following
Create a free account to discover what your friends think of this book!
Community Reviews
Displaying 1 - 3 of 3 reviews
August 16, 2017
This book is an easy-to-read collection of statistical stories, with the theme "Errors, Blunders, and Lies". By errors, Salsburg means that our statistical models aren't always right, so we always write them in the form (Observation = Truth + Error). By blunders, we mean that sometimes data is measured that comes from a different distribution than we are interested in. By lies, Salsburg means data that has been falsified.
The errors section is the longest and (in my opinion) most interesting. Salsburg's structure would amount to something like a first course in statistical modeling and regression. We start from Poisson and Normal statistical models, move to linear regression, multiple linear regression, logistic regression, causation vs. correlation, and conclude with "Big Data". (which here specifically means situations where the number of variables (p) is larger than the number of observations (N)). Salsburg illustrates these concepts, and many others (including consistency and minimum variance) using examples from statistical problems.
The blunders section is an insightful look into situations where measured data comes from a different distribution.The example I remember the most is a severely over-weight Rat substantially driving up the "average" weight. Here, we see how the "median" is a better guess as the average rat-weight. There is a particularly interesting example of the John Tukey lead Princeton Robustness Study, and several good references.
The lies section looks into detecting falsified data. For instance, we know what a "random" sequence of numbers looks like, and we can tell a certain random sequence has been falsified based on how far it is from the uniform distribution. A similar argument can be made for counts of people, specifically that the last number should follow a particular distribution. Finally, there is an interesting story, which was debunked, about how surveyors can tell where results have been falsified (also called curb-stoning).
Overall, the book was delightful, and it is possible to read in 4/5 hours on a vacation day (which is what I did). I only have three gripes. First, I think the R^2 controversy could have been avoided, especially since part of the big-data chapter shows that R^2 can be perfect with enough variables. Second, I thought that instead of referring to the variance, I wonder if an effort early on could have been made to explain the standard deviation, and use this instead. Finally, I think the (observation = truth + error) is a little misleading. For statisticians, we often model the phenomenon as (statistical model + error), knowing that our statistical models are always wrong, but some are useful.
The errors section is the longest and (in my opinion) most interesting. Salsburg's structure would amount to something like a first course in statistical modeling and regression. We start from Poisson and Normal statistical models, move to linear regression, multiple linear regression, logistic regression, causation vs. correlation, and conclude with "Big Data". (which here specifically means situations where the number of variables (p) is larger than the number of observations (N)). Salsburg illustrates these concepts, and many others (including consistency and minimum variance) using examples from statistical problems.
The blunders section is an insightful look into situations where measured data comes from a different distribution.The example I remember the most is a severely over-weight Rat substantially driving up the "average" weight. Here, we see how the "median" is a better guess as the average rat-weight. There is a particularly interesting example of the John Tukey lead Princeton Robustness Study, and several good references.
The lies section looks into detecting falsified data. For instance, we know what a "random" sequence of numbers looks like, and we can tell a certain random sequence has been falsified based on how far it is from the uniform distribution. A similar argument can be made for counts of people, specifically that the last number should follow a particular distribution. Finally, there is an interesting story, which was debunked, about how surveyors can tell where results have been falsified (also called curb-stoning).
Overall, the book was delightful, and it is possible to read in 4/5 hours on a vacation day (which is what I did). I only have three gripes. First, I think the R^2 controversy could have been avoided, especially since part of the big-data chapter shows that R^2 can be perfect with enough variables. Second, I thought that instead of referring to the variance, I wonder if an effort early on could have been made to explain the standard deviation, and use this instead. Finally, I think the (observation = truth + error) is a little misleading. For statisticians, we often model the phenomenon as (statistical model + error), knowing that our statistical models are always wrong, but some are useful.
October 10, 2017
It is an easy-to-read book that describes the essential statistical concepts with great examples. What I like most is that the author did not stop at the surface of the concept, instead, provide more insight and the statistical reasoning underlying the concept itself. It is a great book for people interested in statistics and how it can be used.
Read
November 9, 2021short stories for statisticians
Each chapter contains a brief application of statistics, followed by a bibliography. I wished the author included a short list of statistical methods or tests that relevant to the example.
Each chapter contains a brief application of statistics, followed by a bibliography. I wished the author included a short list of statistical methods or tests that relevant to the example.
Displaying 1 - 3 of 3 reviews