A lot of good content but it needs to go to the right audience
Reviewed in the United States 🇺🇸 on September 28, 2021
This review is based on using the text for a semester-long course that made use of Chapters 1-12 and Chapter 16. As such, I am not really commenting on the other Chapters. Overall I like the book but it's not really suited for my needs as an instructor, sadly.
I've taught the second course in regression to grad students in social, behavioral, and health sciences for over a decade. I think over that time I've taught it eleven(!) times. I switched to this book this year, as I was hoping to get a text that had a more contemporary coverage. There's a lot I like about it. It's conversational and easier to read than many statistics books. It also has a lot of important advanced topics that tend not to be in most regression books, such as Bayesian estimation, poststratification, missing data, and causal inference (great fit for Jennifer Hill's expertise!), among others. However, there are some notable issues that instructors should know about that, for my review, dock it two stars, for my use of the text. Many of these are the downsides of the strengths.
-The book tends to gloss over some important details in points. I get (and largely share) the authors' viewpoints, but there's a good bit of "due diligence" that comes with education and those glossed points can be a bit of a problem. For instance, they mostly duck formulas, which I can understand, but it really is necessary for important and common formulas to show up and be explained. Many homework problems are unnecessarily complicated because of it.
-I mostly agree with the authors and their overall orientation, but there are some spots I really part company (e.g., the notion that a measure could be valid but not reliable).
-The exercises are extremely difficult for my students. Stat book problems are notorious for taking too much for granted on the part of the students, so this isn't unique to this text. A number that seemed pretty reasonable to me turn out to really throw some students for a loop. Unevenness among their backgrounds means that some know some things and others don't. For instance, some students found the R programming no big deal and had issues with more theory while others had the vice versa issue. This is very hard to predict. This is compounded by the fact that important details were glossed over so it's not like the problems or I can say "Hint: See the formula on p. 57". That's something I can fix as an instructor to some degree but generally shouldn't have to as much as I did. I also don't like problems that say "Find a dataset that's of interest to you and do X, Y, and Z with it." I get where that's coming from, but the unevenness of grading is likely to be more of a problem than it's worth. As such, I never assign problems like that. I think a lot of this would be ameliorated by having a lab section where lots of the kind of smaller ad hoc questions could be addressed, but my course does not have a lab section.
-There's R code for all the examples in the book, which is great! However, quite a bit of it is rather complicated by details that mostly make the output look good. Again, I appreciate that and the output does indeed look good. Many of the graphs look great! But I often find myself having to strip it back to something quite a bit simpler so the students don't get lost in the R and lose the point of the examples. Over the course of the semester, I found myself more and more just writing my own examples rather than using the text's.
-I like the integration of Bayesian computation and I intend to incorporate more of that going forward. That is simply a responsible representation of the state of the literature now. However, installing Stan is a real pain and despite multiple attempts some students still don't have working copies (and I'm fairly certain they never will, nor do I have a ready way to help them with this---I am NOT going to touch students' computers, way too risky if I break it). This means I'll have to edit code and problems to deal with the fact that a number of the people in the class can't run the Bayesian computations and the students won't be able to follow along. (My adaptation to this turned out to be not so bad: I ran most examples using lm but had some Bayesian parallel analyses.)
I really wish there was a better alternative for my students. Much as I like the book, I probably won't use it again next year. It's just too difficult for my circumstances. This is unfortunate because there really isn't a text I like out there.
All that aside, I do highly recommend the book for someone who's already reasonably solid on the material AND on R. It's worthwhile, but if you're an instructor, be prepared to have to deal with a good bit of these issues, particularly if your students aren't all a bunch of R experts already.
17 people found this helpful