Norman M. Bradburn & Seymour Sudman: Polls and Surveys. Understanding What They Tell Us. San Francisco: Jossey-Bass, 1988, 249 pp.

Survey research has in the recent decades emerged as a major approach in the social sciences as well as in applied fields. The extensive use of surveys and polls has created a demand for studies that assess the various methodological and political problems of surveys and their use. The book by Bradburn and Sudman, both wellknown scholars and authors of books on similar topics, gives an overview of some of the most important controversies of survey research. The book covers four major topics: the history of survey research, how surveys are done, errors in surveys and the social and political impact of polls.

The antecedents of modern public-opinion polling are the straw polls of newspapers,market and the early social surveys. The history of the straw polls goes back to the early nineteenth century when they were used to predict presidential elections in the USA. The use of straw polls by newspapers proliferated in this century with the Literary Digest poll as the most prominent case. The major weakness in this and other polls was the introduction of sample bias. This derived both from the use of mailing lists that did not reflect the social and political composition of the electorate and from self-selection bias among those who actually returned the questionnaires. The failure of the Literary Digest poll to predict Roosevelt's victory in 1936 is a landmark event in modern survey research in that it demonstrated the failure of the straw polls and at the same time provided an opportunity for George Gallup to promote his newly founded Gallup poll. Gallup, as well as Roper and Crossly, all correctly predicted Roosevelt by a substantial margin. Moreover, what made the successful predictions more impressive was the

fact that Gallup had warned his subscribers well in advance that the Digest's methods would lead to the prediction of the wrong man. While this episode gave opinion polls a boost, the deeper growth stimulus of survey research in the 1930s and 1940s was to a great extent rooted in the two major societal crises of the period, the Great Depression and the Second World War. The Roosevelt administration used surveys to measure the effect of farms programs as well as to assess the support in the general public for the reforms and policy positions of the President. The agriculturalsurveysunit the direction of Rensis Likert became the Survey Research Center of the University of Michigan in 1946 when the demand for government research was reduced. Similarly, the National Opinion Research Center of the University of Chicago and what was later to be the Bureau of Applied Social Research at Columbia were able to expand their operations as a consequence of the war. The success of public-opinion research was temporarily halted when the major opinion polls failed to predict the election of Truman in 1948. This was mostly caused by the fact that the polls stopped field work well in advance of elections and thus were unable to pick up the pro-Truman shift in the final days of the campaign. This of course led the opinion polls in the post-Truman period to monitor closely trends up to election day.

The authors give a brief overview of how the uses of surveys have expanded both in areas of political polling, for administrative and planning purposes (the extensive use of surveys by the US government may be unknown to many), in the work of pressure groups, market research and for purely scientific reasons. While the coverage is not as broad and deep as the authoritative Survey Research in the United States: Roots and Emergence, 1890-1960 by Jean Converse, the book gives a useful overview of the main lines of the development of polls and of the institutional structure of survey research in the United States.

How surveys are actually carried out is described in chapters dealing with the relative merits of face-to-face interviews as compared to telephone interviewing or mail surveys, and the pros and cons of various types of samples. Telephone interviews generally produce data equally as valid as face-to-face interviews. This, in conjunction with the fact that telephone interviewing can be done faster and with lower costs, has made interviews by telephone the predominant mode of data collection in modern survey research. Mail questionnaires tend to be dismissed as a useful tool for the study of general populations, but constitute a cost-efficient and valid alternative for surveys of a special population like members of organizations. However, even mass mail surveys may produce high response rates, the most obvious example is the SOM study by the Department of Political Science at the University of Goteborg which has obtained a response rate of 70 per cent with a mail questionnaire to a representative sample 15-74 years of age in Sweden.

Errors in surveys are divided into sampling and non-sampling errors. Sampling error is the least problematic since it can be estimated for most types of samples and reduced by increasing the sample size. Non-sampling errors are less well understood, and cannot be estimated within a probability model. This category includes such errors as non-response for individuals and items, effects of question wording and form and of interviewer behavior. The discussion of these problems is fairly comprehensive and gives a good introduction to the more advanced treatments of the subject in other books like Questions and Answers in Attitude Surveys by Howard Schuman and Stanley Presser and the recently published Survey Errors and Survey Costs by Robert Groves.

The widespread use of surveys and polls has raised popular concern that they
may have too much impact. This discussion has been most heated in the case of

political polls, which have been seen as unduly influencing the outcome of elections. Before one discusses whether polls should be banned one should know how much influence polls have in various areas, and if the effects are good or bad. Overall the authors tend to emphasize that if the impact is large then the effects of polls generally are good. If the effects are bad they are, luckily, marginal or weak. An example of the first is market research, which is seen as a mechanism for providing information which enables business to serve the consumer more efficiently rather than as an instrument for manipulating consumer needs and promoting harmful products. Similarly, political polling is mostly evaluated as an improvement of democracy by continuously informing elected officials and bureaucratic elites about the preferences and priorities of the citizens. A possible exception is made for the specific American phenomenon of exit polls whereby the voters on the West Coast are informed about the outcome of the election before they cast their vote. But even here the authors oppose restrictions. While the discussion of the impact of polls is reasonably balanced, the backgrounds of Bradburn and Sudman as representatives of the survey-research establishment might have influenced their view in the direction of being less critical to the use of polls than might be the case with a similar treatise by disinterested scholars.