i need a good stat textbook for self study.

April 2, 2007 10:52 PM

i've been looking at reviews on amazon for introductory stat textbooks, and i can't seem to find a single one that doesn't have blatantly contradictory reviews.

i'd appreciate any recommendations i can get. i don't need an overly simplified book, but i do need one with good, clear explanations. i'd prefer one that covers theory as well as application, or at least delves a bit both directions. it's okay if it uses calculus in explanations and whatnot. i am a decently proficient programmer, and i don't mind higher level math -- so long as it's well explained.

if it's at all relevant, i'll be using this as a stepping stone towards machine learning/ai in general, and financial time series analysis as well.

thanks for your help.

posted by tehgeekmeister to education (8 comments total)

A class I took last quarter that was all about stats for AI and learning used All of Statistics. I can't find it in my room so I can't recall the author (although now that I think about it, I want to say it was by Wasserman?) but I know there's only one book with both that title and a bright fuchsia color.

The course was pretty bad, but I liked the book. It skips lots of the proofs, so if you're interested in those (which I was a little) it's annoying to have to look other places, but it manages to touch on a lot of things that I would have thought a bit esoteric. Basically, it's good in that it's well-written and very broad, and gives you the tools to look for things in more depth if that's what you want.

Also, haha, nice crosspost to the LJ mathematics community :)
posted by crinklebat at 11:00 PM on April 2, 2007

I saw the crosspost but I'll answer here instead:

If you're headed towards machine learning, the DeGroot/Schervish book is pretty good, it covers probability theory pretty thoroughly. The most recent edition touches on some stuff that's useful for machine learning/AI like Bayesian filtering and non-parametric stats (I think).

It doesn't look too mathy for intro self-study -- granted, I only used it to relearn things I saw in probability theory (which was taught with course notes rather than a text) the year before.
posted by thisjax at 12:58 AM on April 3, 2007

I had Wasserman (of All of Statistics semi-fame) as a professor for statistics. Well, part of it. Until I dropped the class (it was heavily theory-based and Maths-y; also v. fast--not a professor problem but a me problem). But his textbook (or pre-textbook, as it was at the time) was still the best stats book I could find. When I later re-took statistics, I kept wishing I had kept all of the pdfs that he gave us.

To crinklebat--the book skipped the proofs in part, I bet, because we had to write those proofs as homework in class. Stupid excercises for the reader. I still can't write proofs to save my life.

Schervish, on the other hand, was my other stats professor. I hated his book and his teaching, but I had to suffer through it. It doesn't provide enough in terms of useful information for examples and such, in my mind, and while the probability theory stuff may be okay, once it got to statistics I found it largely useless.
posted by that girl at 5:16 AM on April 3, 2007

I took a graduate statistics sequence intended for people who will analyze experimental data and we were taught out of course notes. There were some recommended textbooks but the instructors pointed out the respective strengths and weaknesses of these books and would refer us to specific chapters if we needed additional depth on some topic. That was my experience with a machine learning course as well, giving the impression there was no one book that had all the desired characteristics. This might explain the contradictory reviews on Amazon.

(The recommended textbook in the machine learning course was The Elements of Statistical Learning by Hastie, Tibshirani and Friedman, and the book does expect knowledge of statistics up to regression. Given your interests, you might also want to take a look at "Intro to econometrics" type books. These books are all about statistics. I was referred to Greene's Econometric Analysis but again I've encountered mixed opinions on it, both in person and in Amazon reviews, and that books also presumes some knowledge of statistics. On preview, this has turned into more of a 'books to avoid' answer, but I think I'm going to check out crinklebat's suggestion at my uni library!)
posted by needled at 5:22 AM on April 3, 2007

I've been taking a few stats courses recently, and I found that Wasserman's All of Statistics is a great introduction and overview (especially if you're thinking about machine learning), and Casella and Berger's Statistical Inference gives more thorough coverage of the statistical fundamentals. If you're going to get one book, I'd start with All of Statistics.
posted by thandal at 9:33 AM on April 3, 2007

A friend of mine teaches criminal justice research at a state university. While her texts aren't as hard-core math as maybe what you're looking for, they may be helpful to those who are searching for a stats book with real-world applications.

~ Healey, J.F. 2005. Statistics: A Tool for Social Research. Wadsworth.
~ Caldwell, Sally. Statistics Unplugged. Wadsworth.
~ Fox, J.A. & Levin, J. 2005. Elementary Statistics in Criminal Justice Research: The Essentials. Allyn and Bacon.
posted by parilous at 11:32 AM on April 3, 2007 [1 favorite]

I'm going through All of Statistics currently. It's good, but he surely doesn't waste any words. Sometimes I wish there were more proofs, but if you take your time and do research on the internet when something doesn't make sense, you'll be fine. Plus, Dr. Wasserman is a really nice guy, so you should buy his book.
posted by jewzilla at 7:03 PM on April 3, 2007

Seconding Casella and Berger's Statistical Inference -- but only if you want a 'solid undergrad' book -- one that would be used for an introductory upper-division course. It has lots of calculus, but explains things well.

Trevor and Hastie's Elements of Statistical Learning is also a great book for machine learning, but requires more statistical maturity (as well as some knowledge of linear algebra).
posted by bsdfish at 9:06 PM on April 3, 2007