Skip to Main Content
Have library access? Log in through your library
Pre-Election Polling

Pre-Election Polling: Sources of Accuracy and Error

Irving Crespi
Copyright Date: 1988
Published by: Russell Sage Foundation
Pages: 220
  • Cite this Item
  • Book Info
    Pre-Election Polling
    Book Description:

    Since 1948, when pollsters unanimously forecast a Dewey victory over Truman, media-sponsored polls have proliferated, accompanied by a growing unease about their accuracy.Pre-Election Pollingprobes the results of over 430 recent polls and taps the professional "lore" of experienced pollsters to offer a major new assessment of polling practices in the 1980s.

    In a study of unusual scope and depth, Crespi examines the accuracy of polls conducted before a range of elections, from presidential to local. He incorporates the previously unpublished observations and reflections of pollsters representing national organizations (including Gallup, Roper, and the CBS/New York TimesPoll) as well as pollsters from state, academic, and private organizations. Crespi finds potential sources of polling error in such areas as sampling, question wording, anticipating turnout, and accounting for last-minute changes in preference. To these methodological correlates of accuracy he adds important political considerations-is it a primary or general election; what office is being contested; how well known are the candidates; how crystallized are voter attitudes?

    Polls have become a vital feature of our political process; by exploring their strengths and weaknesses,Pre-Election Pollingenhances our ability to predict and understand the complexities of voting behavior.

    "Combines intelligent empirical analysis with an informed insider's interpretation of the dynamics of the survey research process....Should be studied not only by all practitioners and students of opinion research but by anyone who makes use of polls." -Leo Bogart, Newspaper Advertising Bureau, Inc.

    eISBN: 978-1-61044-144-5
    Subjects: Political Science, Sociology

Table of Contents

Export Selected Citations Export to NoodleTools Export to RefWorks Export to EasyBib Export a RIS file (For EndNote, ProCite, Reference Manager, Zotero, Mendeley...) Export a Text file (For BibTex)
  1. Front Matter
    (pp. i-vi)
  2. Table of Contents
    (pp. vii-viii)
  3. List of Tables
    (pp. ix-xii)
  4. 1 Introduction
    (pp. 1-10)

    So wrote S. S. Wilks in the opening chapter of the report of the Social Science Research Council (SSRC) on the performance of the 1948 pre-election polls (quoted in Mosteller et al. 1949, 1). The intervening years have seen many changes in poll methodology, for example, in sample design, treatment of turnout, interviewing methods, and timing. To varying degrees, these changes have resulted in improved accuracy. For example, the average deviation from election results of the Gallup Poll has been reduced from 3.6 percentage points in eight national elections from 1936 to 1950 to 1.2 percentage points in seven national...

  5. 2 Design and Conduct of the Research
    (pp. 11-25)

    This study investigates and evaluates the correlates of accuracy of media-sponsored pre-election polls in state and local as well as national elections. The study focuses on media-sponsored pre-election polls for two reasons: (1) the results of media polls are in the public domain, unlike most “private polls” conducted for candidates and parties, and therefore are relatively accessible; and (2) media polls are designed to measure voting preferences and, unlike private polls, are not primarily intended to provide guidance for campaigning that is meant to change preferences. That is, many private pollsters are more interested in guiding campaign directors than in...

  6. 3 Sampling
    (pp. 26-67)

    Controversy between adherents of probability and nonprobability sampling was central in the Social Science Research Council’s evaluation of the 1948 pre-election polls. Although the sampling procedures used by pollsters have changed in many ways since then, this issue persists. Some pollsters are now committed to probability designs, while many others use designs that include probability procedures at some but not at all stages of sample selection. Thus, most pollsters who use quotas also incorporate probability procedures in such stages as the selection of primary sampling units, block clusters, and households (e.g., in random digit dialing). As a result, many of...

  7. 4 Turnout
    (pp. 68-95)

    Obtaining a valid sample of the voting-age population is only the first step in obtaining a sample of the voting population. Each of the next two steps—(1) identifying who is registered to vote and therefore eligible to do so, and (2) identifying who among the registered voters will vote—presents major measurement problems. The importance of solving these problems is indicated by the relation between turnout rate (percentage of total voting-age population that votes) and poll accuracy.

    In the quantitative survey, 39% of the polls conducted in elections with turnout rates of 56% or more are in the high-accuracy...

  8. 5 Determining Candidate Preference
    (pp. 96-116)

    Polls vary in the specific wording of the question they ask measure candidate preference, though they share many features in common. Illustrative of the question wordings commonly used are the following:

    Gallup Poll (“Secret Ballot”): “Suppose you were voting TODAY for President and Vice President. Here is a Gallup Poll Secret Ballot listing the candidates for these offices. [Full names for each ticket’s candidates listed under party heading.] Will you please MARK secret ballot for the candidates you favor today—and then drop folded ballot into the box. (If don’t know or refuse): Well, would you please mark the ballot...

  9. 6 Saliency, Cognition, and Commitment
    (pp. 117-133)

    Influences on accuracy that were discussed in the personal interviews with pollsters include the office at stake, the “type” of election, how well-known the candidates are, whether an incumbent is running for office, whether the election is a primary, whether the election is being held in an on-year or off-year, and at what stage of the campaign voters make up their minds.

    Data from the quantitative survey do not demonstrate any consistent relation between type of office at stake and accuracy. Of the polls related to the presidency, 34% are in the high-accuracy tercile, compared with 30% of the polls...

  10. 7 Stability and Lability of Voting Preferences
    (pp. 134-148)

    Two central issues in any evaluation of poll methodology are (1) whether it is possible to identify a time frame within which pre-election polls must be conducted in order to achieve an acceptable level of predictive power, and (2) whether special research designs are needed if polls conducted within an appropriate time frame are to have predictive value.

    The pollsters who were personally interviewed agreed that since polls measure candidates’ standing as it exists at the time of interviewing, pre-election polls can be very misleading if voting preferences change between the end of interviewing and election day. (It is for...

  11. 8 Characteristics of Polling Organizations
    (pp. 149-162)

    Pre-election polls are conducted under sharply different operating conditions that can significantly affect their accuracy. The conditions differ first with respect to the importance given to achieving accurate election predictions. They differ second with respect to organizational structure, namely, the type of professional staff that designs and analyzes the polls, the type of interviewing staff that is employed, and whether interviewing is conducted at a central telephone location.

    Accurately “predicting” election outcome is not an important goal for many pollsters (see Table 8.1). Some have as their explicit policy the avoidance of any basis for treating their polls “as a...

  12. 9 A Multivariate Analysis
    (pp. 163-169)

    We have seen that the accuracy of pre-election polls is related to a number of methodological and contextual characteristics. Moreover, some of these characteristics are themselves correlated with each other. The question arises, therefore, as to how the combination of associated characteristics acts together to affect accuracy. This question has been answered in part by testing for the significance of differences between pairs of means for cells created by multivariate cross-tabulations. To further answer this question, a regression analysis was performed using data from the quantitative survey, with the accuracy of pre-election polls the dependent variable. The specific purpose of...

  13. 10 Conclusions
    (pp. 170-184)

    The Social Science Research Council’s investigation of the performance of pre-election polls in the 1948 presidential election identified four major sources or error: (1) the use of flawed sample designs, (2) failure to screen nonvoters out of the sample, (3) inadequate methods for treating “undecided” responses, and (4) failure to measure late changes in voting preference. This study has examined these and other aspects of the research designs now used in pre-election polls. It has also examined additional issues related to the political context of election campaigns as well as to selected characteristics of polling organizations.

    It is apparent that...

    (pp. 185-192)
  15. References
    (pp. 193-196)
  16. Index
    (pp. 197-205)