<< 1 >>
Rating: 
Summary: Very useful
Review: Nearly all of traditional statistical tests assume that data points are distributed along the "normal," Gaussian bell curve. The assumption may be explicit, may be hidden inside a chi-squared phrase, or may stand silently behind discussions of mean and variance. If your data don't match that assumption, stated or not, the tests give wrong answers. Nonparametric stats work without that assumption. In fact, most work without !any! assumptions about the distributions of data. These are very robust techniques, and this book demonstrates a number of simple and effective ones. The author describes everything you need to know about a number of tests: when each applies, how to perform it, and how to interpret results. It's a very useful guide for people who need thigh-quality answers from low-quality data. Best, many of these procedures are simple enough to apply routinely to almost any data set. I find them helpful for quick checks before applying more detailed kinds of parametric analysis. If your data fail the loose bounds of non-parametric testing, you know that fussier, more high-strung tests have no hope of reasonable answers. There are no theorems here, and very little development of the underlying principles. That's good for someone who just wants the answers. On the other hand, it's a real weakness if you need to customize analysis for an unusual problem. You just won't find the fundamentals you need for sound improvisation. (This review describes the first edition.)
<< 1 >>
| 
