Top positive review
39 of 41 people found this helpful
Very good text for MPH students, but not perfect
on February 7, 2011
This is the second year that I've taught introductory public health statistics out of Rosner's book.
* Good sections on probability and fundamentals of inference that show enough of the derivations to allow the better students to appreciate the mathematical basis, without making that the emphasis of the text.
* Includes a brief discussion of Bayesian inference and simple examples. This constitutes a very small fraction of the book, but is typically absent from introductory biostatistics texts.
* As much emphasis on poisson and binomial models as normal models. This is critical for public health students!
* Great section on epidemiologic methods in the final chapters, including logistic regression.
* Good coverage of exact methods and a short chapter on classical non-parametrics.
* Includes some discussion of missing data analysis.
* Hundreds of homework problems to choose from at the end of each chapter, with solution sets available online to registered course instructors.
* The two-tailed p-value is always described as 2 times the one-tailed p-value, which is fine for symmetric sampling distributions but not so good for binomial and poisson distributions. This is a big source of confusion in my class, as it's inconsistent with R.
* An antiquated emphasis on calculating critical values for hypothesis tests, rather than using p-values or confidence intervals.
* The chapter on two-sample t-tests instructs students to assume equal variances if a formal hypothesis test for equal variances does not result in rejection of the null. This is ill advised and makes the entire chapter needlessly complicated.
* More emphasis than I'd like on normal approximations for binomial and poisson distributions and corresponding one-sample tests. Again, this seems to cause needless confusion considering that it is much simpler for the students to obtain exact results.
* Some of the example homework problems are not worded very clearly, so be careful about selection. This also leads to a few surprising interpretations/answers in the online solutions.
* Minor quibble: two-sample binomial tests are covered in two different sections of the book, depending on which version of the effect estimate is used (p2-p1, p2/p1, odds ratio, etc.). I wish they were all presented together.