Logistic Regression: a Self-learning Text, 2nd edn

Ideal preparation for this book is claimed to be a course on quantitative methods in epidemiology and a course in applied multiple regression. So, imagine that you are sitting attentively in a lecture, and the formula goes onto the board. Then, just in case you were not quite on the ball (notwithstanding the courses in regression and epidemiology that you have attended), the lecturer helpfully writes underneath: ‘… the logistic model P(X) equals 1 over 1 plus e to minus the quantity α plus β1 times E…’. Well, that was helpful was it not? The next formula swiftly follows: and then more helpful text, ‘… by substituting the value E equals 1 into the model formula … we then obtain 1 over 1 plus e to minus the quantity α plus β1 times 1, or simply 1 over 1 plus e to minus the quantity α plus β1’. Is this not clear yet? We continue: and the text ‘… For E equal to zero … substitute E equal to zero into the model formula and we obtain 1 over 1 plus e to minus α’. But, by now, you are asleep and so am I. Insomniacs need look no further. This text is repetitious beyond belief, tedious beyond hope, unimaginatively detailed beyond caring. As your eyes glaze over reading this review, just keep in mind how lucky you are to have been spared the full narcotic verbiage of the original. The material is mostly of very basic logistic regression, very thoroughly covered. Readers staying awake through the first seven chapters (to page 226) should be able to grasp the mechanics of logistic regression modelling. However, the authors manage to conceal the joy and excitement of statistical exploration: the single running example is used only to illustrate the nuts and bolts and not to motivate one model or another. Only three or four data sets are used. These are described again and again unnecessarily in the same or similar words wherever they appear. The data sets are available on an accompanying disc but are not listed in the book: a pity. About half-way through the book, I had a sudden shocking thought, a thought which should have occurred to me sooner, but then absence is so less stimulating than presence: I could not remember seeing a single graph. Well, I checked and there are fewer than a handful, including these: a drawing on page 5 to show us what the logistic function looks like and two scatterplots on page 339 of 10 (presumably faked) pairs of data to show positive and negative correlation. Need I say more? I shall. We teach our students to graph data, first and last, as probably the most important thing to do with a data set. Many of us will have made some stupid blunder that could have been avoided by plotting the data (I certainly have), and this is all the more important when the models start to become complicated and our natural intuitions desert us. It seems to me ludicrous to offer a text-book on logistic regression which offers not a single genuine data plot, either to help interpretation or to diagnose modelling problems. Really, those interested in learning about logistic regression would acquire a much better feel for the subject by reading chapter 6 of Friendly (2001), which contributes far more to motivation and understanding than this text does. On the positive side, the authors clearly know what they are talking about and, if you do not mind ploughing through the text, you will obtain a fairly good idea of the basic recipes for fitting logistic regression models. This includes a treatment of generalized estimating equation models in rather more detail than in Hosmer and Lemeshow (2000), brief coverage of generalized linear mixed models and a useful appendix on software: SAS appears to be the authors’ choice, but there are examples for SPSS and STATA. The bibliography is short: 21 references. McCullagh and Nelder (1989) is included, but there is no room for the three texts that I consult most for such models: Collett (1991), Agresti (1990) and Hosmer and Lemeshow (2000) (which has 191 references, to save you the trouble). However, just in case you had missed your ideal preparation in epidemiology and applied multiple regression, you will find references to … oh!, well … two texts by D. G. Kleinbaum: happy reading.