Choosing between "Statistics" by Freedman et al., and "Statistical Models: Theory and Practice" by Freedman I'm not a statistician, but I'm very interested in statistics and I'd like to buy a book to keep as a reference. I have a few books on specific subjects (such as The Elements of Statistical Learning for machine learning or Bayesian Data Analysis for...well, Bayesian Data Analysis :) I was also looking for a more generic book.
Freedman's books are often well-considered here:
Advanced statistics books recommendation
What book would you recommend for non-statistician scientists?
Statistics by Freedman, Pisani and  Purves (A) is the chosen answer for the latter question, and I was going to buy that. However, I found out about Statistical Models: Theory and Practice (B). The two books seem similar (for what I can tell: Amazon restricts me even from reading the full ToCs...I don't know why). The publication dates are very close. However:


*

*B is considerably cheaper. I could get A used, though, so if A is clearly better than B, I'm willing to go for A.  

*A is longer, but it seems to me that the main chapters missing from B are related to probability. I don't need that part, so if that's the only difference or the main difference, I'd rather buy the cheaper and more transportable 
B :)


Which book would you suggest I buy?
 A: I am a statistician, taught it for 40 years, mainly to biologists.  The Nick Cox answer above is dead on.  In my opinion, "FPP" is still by far the best intro book on statistics.  Strong emphasis on concepts, great examples (though I wish more were from biology!), and counter-examples (showing how 'the obvious' can sometimes be wrong), and exercises.  It is easy reading, but this can be deceptive: you have to think.  "Statistical Models" (Freedman) is a second or third course book.  It's also very conceptual.  You'd probably want a more standard book for learning the basics of least squares methods (regression, anova, etc.).  Freedman is more concerned about when the models are justified (usually as good approximations to "truth") and when not.  Very important now, when you can run very complex models at little more than the push of a button, but not have much idea of what you assumed or what the results mean.
Davison's book is also excellent, but more technical and practical: it describes the most important standard models (and some less standard) in a variety of areas and shows ways to analyze them.
A: They're quite different. 
(A) is explicitly introductory (but in many ways not elementary). That may seem contradictory: perhaps it's fair to say that (A) assumes intelligent readers willing to think hard, but not previous knowledge of statistics. There are no gimmicks such as colour photographs of happy people, boxes of various kinds with extra materials, or rude stories based on the author's wilder experiences or over-fertile imagination. (I allude without references to some of the more appalling alternatives in the market.)  A smart high school student or anyone who remembered most of their high school mathematics would find it rewarding, as well as the more obvious undergraduate market. 
(B) is more a second text and would be tough going for anybody who didn't find the content of (A) familiar. I'd say (B) depends on readers having encountered most of the material at least once before, because many of the explanations are cleverly concise but equally rather condensed. I'd say it's really for researchers, minimally final-year undergraduates preparing a dissertation or research paper. It's also more opinionated, which you'll love or loathe according to whether you agree with Freedman, whose high standards often excluded almost anybody else's work. 
I re-read (A) with profit and pleasure every few years and have done so since the first edition (with skimming and skipping). 
Disclosure: I am not a statistician either; nor I have ever taken courses taught by statisticians. 
Gossip: A biography of John Tukey (see here for details and a review) twice includes an undocumented story that David Freedman as a graduate student at Princeton really couldn't get on with Tukey's sometimes elliptical and elusive teaching style. It is tempting to speculate that this may have been an underlying reason why (A) avoids box plots and Tukeyish exploratory methods generally. 
