review by
Gordon Haverland, November 2008

I disagree that this book will let you master statistics. It will most likely make you competent at reasonably simple stuff. I think some of the detail presented is sloppy in a couple of places. There are a few more mistakes in the book than I would have expected for such a large team involved. It is very much more than a single person writing a book.

All of the Head First series are written in a new style, which is supposed to be more effective at teaching. There are many different kinds of things going on in the book, and for the most part the mix seems to work well.

Critisms:

• Over the years I have seen a number of problems when people choose to work with units of "percent". I think the only good solution is to discourage people from using percent in any description or analysis.
• The term "normal distribution" has far too much baggage to be useful to anyone. I think every statistics book should endeavor to get people to avoid the term. Perhaps one incentive to get people to stay away from this term is to point out that distributions of undefined variance exist.
• There also exist many problems where the domain of the independent variable is finite, and the use of a Gaussian in those circumstances can lead to problems.
• Starting the book off with charting is as good a place to start as any. Axis origins are discussed, axis transformations aren't.
• Frequency is used in many places where a count might be better. For example, frequency is often a "rate" of doing something (in my world), which means that a sample histogram is the same as a population histogram (except for noise in the sample).
• Some uses of the word variance seem to be describing standard deviation.
• I think the goodness of fit testing on the croupiers needs to be expanded.
• Some people will think that correlation requires a straight line relation between variables, not any functional relation between variables.
• The bivariate example chosen is not a good choice for explaining regression, or at least the simple kind of regression done for this example. This example really calls for showing the difference between Y(X) regression, X(Y) regression and orthogonal regression.

Not that I am short, but I hope Julie never gets a date. :-)

That notwithstanding, if a person was only going to have a single statistics book, this would probably be a very good choice.