REVIEWS
Kenneth C. Land and Stephen H. Schneider (eds), Forecasting in the Social and Natural Sciences, 1987 Dordrecht and Boston: D. Reidel, 1987.381 pages. This book by two editors, one a social scientist (Land) and one a natural scientist (Schneider), achieves the unusual feature of bringing together social and natural science forecasters side by side. It emerged from a conference in 1984 sponsored b y the Committee on Social Indicators of the Social Science Research Council. Despite the diversity of background of the participants, Land and Schneider detected a sizable number of shared positions, listed in the editors' Foreword, held in common by all the authors. It has become standard "boiler plate" for reviewers of edited books to comment on the poor coordination between chapters, and so I hesitate, but only momentarily. The problem exists, and is compounded by very little "connective tissue" supplied by the editors to link the sections. In spite of this, many of the contributions contain valuable ideas. In a work produced by specialists in so many different areas, some pages will defy the reader's best efforts at penetration, and potential readers should expect that in this book. But I found such areas to be mercifully short, and found no chapter in which the central theme could not be followed. It is, naturally, impossible to comment on all points of interest in a review, but I shall single out a few for special mention. The book opens with the standard overview chapter by the editors. It seemed to me to be very useful in developing some ideas in subsequent chapters and in providing a degree of coordination. This being said I should also note an over-generalization of conclusions about prediction as a whole: the chapter concentrates on forecasts produced by causal models, leaving out such alternatives as judgmental or leading indicator techniques, yet reaching conclusions said to apply universally to all forecasting. There is a valuable discussion of the difficulties of forecasting. I liked the discussion in the overview concerning the construction of asynchronous couplings between models when the time-scales of two linked models are totally dissimilar (pp. 23--5). This is the subject of the contribution by Clark. Social Indicators Research 22:319--325, 1990.
320
REVIEWS
Clark's chapter on scale relationships in the interactions of climate, ecosystems, and societies contains important insights on the effects of time and distance scale disparities on the interlocking of models, as well as a tantalizing partial demurrer concerning the usual gloom and doom on the greenhouse effect (pp. 364--5). It is also occasionally an object lesson on how little one understands the technical vocabularies of other disciplines. My personal favorite here is "baroclinic instability," which does not mean what I guessed. The chapter by Long and McMillen on Census Bureau projection methods is a clear, concise survey of techniques and models now in use and on the horizon. I found it interesting that the authors adhered to the "party line" on the great gulf between demographic projections and demographic forecasts, despite articles by Hajnal and Keyfitz that decry such face-saving sleight of hand. The chapter, one of the most informative to me, is slightly marred by mismatch between table and textual comments. When I started the Denis chapter on the importance of the decisionmaking context, my first thought was that it was totally out of place in a collection on forecasting. But then on reflection I realized that Denis was dealing with the "real" world, a world of planners who often have no interest in the accuracy of a forecast but only whether its conclusions advance the interest of their agency -- as they see it. The cautionary tale of the same forecast being revised to correspond with revised vested interests is made more grotesque still by the utter indifference toward accuracy. This chapter is a necessary corrective to a technical vision that ignores such political realities. The chapter by Smith on the economics, politics, and sociology of the social forecasting industry was gratifyingly appreciative of my own work, and among its other virtues did a masterful job with its subject. My only qualm is that much more is needed: Smith deliberately restricted himself to the social forecasting trade, narrowly considered, while the economic forecasting business so desperately cries out for the same sort of dissection. I was happily impressed with the clarity of the chapter on forecasting errors and statistical decision theory by Berk and Cooley. It seemed to me that they successfully marched a very large number of ideas past the
REVIEWS
321
reader concerning error cost functions. In particular, a large part of the chapter dealt with unusual (but plausible) error cost functions. The final section of the chapter, involving the effects of intervening on the accuracy of the forecast, was provocative. What of the overall worth of the book? The editors see the scope of the book as encompassing "essays that (a) describe the organizational and political context of applied forecasting, (b) review the state-of-theart for many forecasting models and methods, and (c) discuss issues of predictability, the implications of forecast errors, and model construction, linkage, and verification" (p, 1). I think I have already registered my conclusion that the chapters accomplish these aims with high competence but uneven coverage. Just as readers should know what is to be gained by reading this book, they should also know what is not to be found therein. Two omissions in particular disturb me. First, notwithstanding that the very raison d'&re of the book is to bring together for mutual stimulation forecasting in the social and natural sciences, it is fair to say that absolutely no attention is paid to the treatments in philosophy of science, and philosophy of social science, of the comparability of the two realms. The nondiscussion of significant points of difference, such as self-altering forecasts, is without doubt a natural consequence of this lacuna. Another important subject whose absence is felt is the evaluation of forecast methodologies. How evaluation can be done, what extant evaluations have concluded, and the objections to different evaluation techniques, are highly relevant for such a book. Naturally, several authors do discuss evaluation in their own specific domain, but there is no general discussion of this central topic, and even the scattered discussions that are present prove difficult to locate. In sum, this reviewer's appraisal is favorable, primarily because of the high overall quality of the contributions and the new insights in specific areas. In my opinion the present work compares favorably with the other recent edited work on prediction in science and society, although the two works are not as comparable as they first appear? Nevertheless, if a second volume is contemplated important gaps should be filled.
322
REVIEWS NOTE
J. Mason and J. Westcott (eds.) Predictability in Science and Society. A Joint Symposiumof the Royal Societyand the British Academy.London:Royal Societyand the BritishAcademy,1986.
Sociology Department, University of Western Ontario, London, Canada, N6A 5C2.
RICHARD L. HENSHEL
Angela Dale, Sara Arber, and Michael Procter, Doing Secondary Analysis. (Contemporary Social Research Series, Volume 17) Unwin Hyman Ltd., London, 1988, 241 pages, ISBN 0-04-312041-5 (hardback s 0-04-312042-3 (paperback s Rapid Review General appraisal: Accuracy of information: Scope: Clarity of writing: Quality of illustrations: Comparative value of book:
Recommended. Good. Limited. Good. Good. Excellent for secondary users of survey data.
Doing Secondary Analysis, the seventeenth volume of a series edited by Martin Bulmer, is a practical guide to the use of survey data for secondary analysis. It is a well-written monograph expanding upon the related themes of two earlier volumes in the series, Secondary Analysis in Social Research by Catherine Hakim (Volume 5, 1982) and The Survey Method by Catherine Marsh (Volume 6, 1982). This volume is sociologicai in nature although the techniques described are not restricted to sociological analyses. The terms of reference of the book are limited in two regards. First, although acknowledging that secondary analyses can be applied to a variety of data, the focus is on the use of survey data, and in particular, "large-scale government survey data" (p. xi). The other limitation is that the examples and some of the issues are specific to Britain. I think that a more accurate title would have been 'Doing Secondary Analysis of Survey Data in the UK'. This is not
SocialIndicatorsResearch22: 322, 1990.
322
REVIEWS NOTE
J. Mason and J. Westcott (eds.) Predictability in Science and Society. A Joint Symposiumof the Royal Societyand the British Academy.London:Royal Societyand the BritishAcademy,1986.
Sociology Department, University of Western Ontario, London, Canada, N6A 5C2.
RICHARD L. HENSHEL
Angela Dale, Sara Arber, and Michael Procter, Doing Secondary Analysis. (Contemporary Social Research Series, Volume 17) Unwin Hyman Ltd., London, 1988, 241 pages, ISBN 0-04-312041-5 (hardback s 0-04-312042-3 (paperback s Rapid Review General appraisal: Accuracy of information: Scope: Clarity of writing: Quality of illustrations: Comparative value of book:
Recommended. Good. Limited. Good. Good. Excellent for secondary users of survey data.
Doing Secondary Analysis, the seventeenth volume of a series edited by Martin Bulmer, is a practical guide to the use of survey data for secondary analysis. It is a well-written monograph expanding upon the related themes of two earlier volumes in the series, Secondary Analysis in Social Research by Catherine Hakim (Volume 5, 1982) and The Survey Method by Catherine Marsh (Volume 6, 1982). This volume is sociologicai in nature although the techniques described are not restricted to sociological analyses. The terms of reference of the book are limited in two regards. First, although acknowledging that secondary analyses can be applied to a variety of data, the focus is on the use of survey data, and in particular, "large-scale government survey data" (p. xi). The other limitation is that the examples and some of the issues are specific to Britain. I think that a more accurate title would have been 'Doing Secondary Analysis of Survey Data in the UK'. This is not
SocialIndicatorsResearch22: 322, 1990.
REVIEWS
323
meant as a criticism of the book, in fact I feel that it will prove to be invaluable to a specialized audience. The use of existing data for secondary purposes is a widespread method in the social sciences. Many studies by economists and demographers involve the use of existing data. Sociologists are more apt to use surveys and field work to gather data but they also use existing data. Singleton et al. base their research text (Approaches to Social Research, Oxford University Press, New York, 1988) on four major methods: (1) experimentation, (2) survey research, (3) field research, and (4) the use of available data. They list five sources of available data: (1) public documents and official records; (2) private documents; (3) mass media; (4) physical, nonverbal materials; and (5) social science data archives. Thus the secondary analyst can take advantage of a large range of sources of data. The authors of the book under review note that "many other data forms may become the subject of secondary analysis" (p. 3) and refer the reader to Hakim (1982) for more details. They then proceed to restrict their focus to the secondary analysis of microdata derived from surveys. The growing popularity of the secondary analysis of survey data can be traced to at least five factors: (1) the high and increasing cost of conducting surveys, (2) the low and decreasing amount of monies available to social scientists to do research, (3) the requirement by some funding agencies that data collected be made available for secondary research, (4) the establishment of major data archives such as the ICPSR at the University of Michigan and the ESRC at the University of Essex, and (5) an acceptance of secondary analysis of surveys as a legitimate form of research. This book is evidence of the advanced nature of this method. Doing Secondary Analysis contains 11 chapters. The first three are the most general dealing with an introduction, a sociological perspective, and the benefits and costs of secondary analysis. The fourth chapter presents the potential of hierarchical surveys for secondary analysis. The next four chapters are very specific covering how to choose and order a data set, elements of computing technology, software, and preliminary analysis once the data set is obtained. Chapters 9 and 10 follow-up on Chapter 4 presenting techniques that can be used to generate complex variables from hierarchical surveys. The final
324
REVIEWS
chapter is an example based on a study by the senior author on inequality of income and assets. The material on the use of hierarchical data is a major strength of this publication. Public use census data as well as some large scale survey data are distributed in a hierarchical format, i.e., with several levels and types of records per case. The major example used in Doing Secondary Analysis is the General Household Survey sponsored by the Office of Population Censuses and Surveys, UK and deposited at the ESRC Data Archive. This file consists of four types of records: households, family units, individuals, and doctor consultations. Each household contains one or more families, each family contains one or more individuals, and each individual may not have had any consultations or as many as 12 in a two week reference period. This type of data file is considerably more complicated to process than the typical rectangular file but it can yield many variables by combining levels. For example, if the household is considered the case then the total income for the household can be computed as the sum of the individuals' income within the household. Another example would be to use housing tenure from the household level for each individual if the individual were considered the case. Thus the secondary analyst, through some creative computing, can generate variables that were not gathered directly by the primary researchers. A potential source of confusion relating to hierarchical files is worth noting: the use of the terms 'unit of analysis, 'level of analysis', and 'case'. I realize that the methodological literature is imprecise on these terms but I would have liked the authors to have made it clearer how they intended to use them. On page 63 the authors give the ability to choose the level and unit of analysis as the first advantage of hierarchical files. They note that "in some cases the researcher might choose to analyse only certain units within a particular level of analysis". The example given is within the level of establishments, a researcher might choose to study only retail establishments, The variables being analysed are based on the units of analysis and they can be selected from a particular level or derived from more than one level. The term case appears to be synonymous with unit of analysis (diagram on page 162 refers to cases while the accompanying text on page 163 refers to units of analysis). I feel that the use of the three terms is quite logical but, due
REVIEWS
325
to their centrality to the hierarchical data discussion, should have been clarified in the beginning of the book. Given the number of computing systems and the constant change associated with hardware and software I wonder about the appropriateness of Chapters 6 and 7, Chapter 6 covers hardware and operating systems, 7 overviews a few statistical packages and SIR, a database management system. This information is available in other sources and is not specific to secondary analysis. In sum, I think that this book is a good resource for secondary analysts but its specialized nature somewhat restricts the potential audience. The material on hierarchical data files is innovative and very useful to those investigators who wish to take full advantage of these resources. I would like to close with a cautionary note. From 1975 until 1983 I worked in the Social Science Computing Laboratory, University of Western Ontario. During that time I witnessed an explosion in the secondary use of survey data. I certainly feel that this method is legitimate and desirable in that it promotes the full use of data, often gathered at great cost. My caution is that we should be careful not to rely so heavily on existing data that we stop doing primary analysis. Large general purpose surveys that can be used by many researchers are an excellent source of information and may be cost effective. I find, however, that students and researchers, under pressure to get theses and publications completed, are letting the variables in the existing data dictate the type of analysis that they will conduct. The secondary use of survey data plays an important role in our array of methods but it should only be used when it is the best method to answer the substantive questions under investigation or when it can help to triangulate findings from other methods.
Department of Sociology & Anthropology, Universityof Guelph, Guelph, Ontario, Canada.
F R E D E R I C K T. E V E R S