REVIEWS
P.H. Rossi, J.D. Wright and A. B. Anderson (eds.), Handbook of Survey
Research, Academic Press, New York, 1983, 755 pages. Rapid Review General appraisal Accuracy of information Scope Clarity of writing Comparative value of book
Highly Recommended Excellent Excellent Good to Excellent Excellent
This is the most helpful collection of essays on survey research that 1 have ever read. The editors assert that it is their "highest hope that the Handbook of Survey Research will prove valuable both to advanced students and to practicing survey researchers who seek a detailed guide to the major issues in the design and analysis of sample surveys and to current state of the art practices in sample surveys" (p. xvi). I don't see how they can fail to have their hopes fulfilled. The 16 chapters of the book are written by some of the most productive survey researchers practicing today. While the essays are generally not intended to be introductory, most of them begin pretty close to the beginning and take the reader fairly rapidly into more advanced topics. Much of the book could be appreciated by advanced undergraduates in statistics and social science. After a brief history and overview of sample surveys (by the editors), the book has two essays on sampling. M. Frankel has a theoretical paper covering mainly stratified and cluster sampling, while S. Sudman has an essay examining a variety of concrete examples of sampling reported in well-known journals. His checklist of things to look for and his detailed criticisms are very helpful. There are 3 papers related to questionnaire design and item writing. P. B. Sheatsley's paper has the broadest scope, dealing with mode of administration, types of sample, question writing, question order and pretesting. N. M. Bradburn's focus is primarily on question writing and response effects for Social Indicators Research 17 (1985) 97.
98
REVIEWS
various items. D.A. Dillman's paper gives a thorough review of mail survey techniques, indicating standards that many people wilt regard as phenomenal but which are certainly possible. Two papers deal with management problems related to survey organizations (K. Prewitt) and data collection (E. Weinberg). Two other papers cover theoretical aspects of measurement, including reliability and validity (G. W. Bohrnstedt), and scaling theory (A. B. Anderson, A. Basilevsky, and D. P. Ji Hum). There is one paper on the use and abuse of computers and canned programs (N. Karweit, E. D. Meyers, Jr.), and a related paper on managing missing data (A. B. Anderson, A. Basilevsky, D. P. J. Hum). Finally, there are 3 papers on data analysis, and a paper dealing with special issues related to criminal justice and crime victimization surveys (E. Martin). The first paper on data analysis deals with the general linear model mainly involving 2 variables, but with some discussion of multivariate problems too (R. A. Berk). D. G. Taylor's paper covers the use of contingency tables and latent structure analysis of quantitative data. R. M. Stolzenberg and K.C. Land review issues in causal modeling for recursive and nonrecursire cases. Readers of this journal who are familiar with the Sage series on Quantitative Applications in the Social Sciences may be interested to know that this Handbook is generally aimed at a more advanced level than the Sage series. So the two 'collections' of readings are not quite competitors. The virtue of the Sage series is that one can buy a single paper on a single topic. The virture of the Handbook is that one buys good papers on most topics at a single shot. I think we are fortunate to have both kinds of publications available.
University of Guelph and Institute of Public Affairs, Dalhousie University, Halifax, Nova Scotia, Canada B3H 3J5.
ALEX
C. MICHALOS
REVIEWS
99
G. Nigel Gilbert; Modelling Society: An Introduction to Loglinear Analysis for Social Researchers, Allen and Unwin, Winchester, Mass., 1981, 131 pages, $12.50 (paperback), $ 28.50 (cloth).
Rapid Review General appraisal Accuracy of information Scope Clarity of writing Comparative value of Book
* Excellent Good Excellent Good
Gilbert provides a lucid, nontechnical introduction to loglinear analysis. It should be useful to advanced undergraduate and graduate students in sociology, as well as to those social researchers who have not been exposed to this relatively recent methodological development. The author is indeed correct when he claims that the discussion should be "understandable to anyone with a knowledge of basic statistics and some acquaintance with simple regression" (p. xi). Chapter 1, 'Real and Imaginary Worlds', presents a simple yet important discussion of the relationship between observed and expected data. The emphasis on developing a model (Imaginary World) and assessing its "fit" with the observed data (Real World) is particularly appropriate when using loglinear techniques. The discussion of classification and measurement in Chapter 2 is less satisfying. The topic is a controversial one and intrinsically more complex than its treatment here suggests. I would have been happier with a single paragraph which stated that loglinear techniques can be fruitfully utilized with discrete measures. The remainder of the book presents an orderly exposition of loglinear analysis, starting with a simple two-dimensional table and ending with a discussion of causal analysis and partitioned tables. These chapters demonstrate that the author has effectively taught this technique to his students. The use of analogies, simple yet uncontrived data sets as examples, and the progression from simple to more demanding considerations, make this an ideal text. How could this textbook be improved? In the author's preface, the patently true assertion is made that application of methods of analysis such as regression and analysis of variance, with their more stringent assumptions, Social Indicators Research 17 (1985) 99.
100
REVIEWS
risks "spurious or meaningless results" (p. xi). Since the book provides numerous empirical analyses, it would have been quite illuminating had examples been chosen on the basis of providing cogent alternative interpretations to those presented in the original analysis. Not one example meets this criterion. This is unfortunate since the assertion mentioned above is a frequently made one. Yet seldom does one find examples of re-analysis using different techniques which support plausible alternative interpretations. Since loglinear techniques contain some severe limitations and disadvantages relative to other techniques such as those exemplified by the program LrSREL, converts to loglinear techniques would be more likely had the author provided some controversial examples of spurious or misleading results. A second criticism is that the original emphasis on assessing the fit between a theoretically informed model and the observed data degenerates somewhat in the latter chapters. Gilbert seems to have been seduced by the capabilities of loglinear programs to gr~d out all possible combinations of models effortlessly, mechanically and mindlessly. By the time we reach the third-last chapter, tables are partioned in alternate ways in single-minded pursuit of a smaller chi-square value. The theoretically informed model suffers from comparative neglect. I would sooner have seen that chapter deleted. I would recommend this book both for instructional purposes and for selfteaching for social researchers who have not been exposed to these techniques.
Department of Sociology and Social Anthropology, Dalhousie University, Halifax, Nova Scotia, Canada B3H 1 TS.
VICTOR THIESSEN