The National Annenberg Election Survey (NAES) of the 2000 presidential election was a landmark in the history of American political opinion polling. Never before had so intense a study been conducted of the public's day-to-day reactions to the political events of a U.S. presidential election and its extended contestation in the courts. The rolling cross-sectional (RCS) design of the survey allows one to study campaign dynamics in a way that has seldom been possible in U. S. elections. Richard Johnston and his colleagues perfected this design in their studies of Canadian national elections, and he collaborated with Kathleen Hall Jamieson in designing the 2000 NAES. The 2004 version of the survey followed the same strategy, and we provide the data from both of these studies in this second edition of Capturing Campaign Dynamics (CCD).
The first edition of CCD explored ways to summarize and analyze the data from this massive survey. In this edition, we again provide most of the same material with a few additions. First, we provide more background regarding exploratory and descriptive statistics that can be used in initial analyses of the survey (Chapter 6). We hope this chapter serves as a bridge to the more complex material regarding multivariate methods covered in later chapters. In addition, we provide more examples of ways to analyze trends in voter knowledge during the 2004 election campaign (Chapter 10). This chapter is intended to highlight some of the innovative analyses that one can undertake to understand campaign dynamics even in small segments of the population.
The majority of the book is based on a course taught by Dan Romer and Kate Kenski on the analysis of the RCS design at the Annenberg Summer Institute on Methods and Statistics at the University of Pennsylvania. The course was designed as a practical guide to analyzing the survey rather than a rigorous examination of the statistical underpinnings of data analysis. The presentation assumes that the reader is familiar with basic statistical concepts such as significance testing, but in this edition we review some of this material in the new chapter described above. In addition, important concepts are reviewed when necessary, and an appendix of technical terms is provided as a reference for readers. Because chapters were written by different members of the study team including graduate students from the Annenberg School for Communication, we credit them for the chapters to which they contributed. A brief overview of the plan of the book will help readers to find the material that is most useful to them.
In Chaper 1, Kathleen Hall Jamieson and Kate Kenski again discuss the motivating premise of the NAES that a more complete record of the public's reaction to U.S. presidential elections will help to answer important questions about the influence of campaign events on the electorate. Despite the skepticism often expressed about the importance of election campaigns, Jamieson and Kenski provide compelling evidence for the usefulness (and limits) of the NAES in understanding events over the course of a presidential campaign. They also provide some intriguing examples of similarities and differences in media use and political discussion across the last two presidential campaigns.
In Chapter 2, Ken Winneg, Kate Kenski, and Christopher Adasiewicz provide an overview of the content of the surveys for both 2000 and 2004. Not only do they describe the main RCS datafile, but they also review the many smaller panel and cross-sectional studies that were conducted during each election. In addition, they describe other files that provide helpful methodological information for the use of the dataset, such as demographic weights and response rates for the survey. They also discusses practical considerations surrounding the use of these measures.
In Chapter 3, Kate Kenski reviews important concepts in survey and research design. She then shows how they apply to the design of different survey methodologies for the study of elections. In particular, the strengths and weaknesses of the RCS design are discussed in relation to other survey methods, such as simple cross-sectional and panel studies.
In Chapter 4, Kenski discusses the underlying strategy of the RCS design and the specific sampling and interviewing protocols that ensure comparable samples and questioning on each day of the survey. These procedures enable researchers to study change attributable to events during the election campaign rather than to changing features of the survey methodology.
In Chapter 5, Kenski reviews the importance of graphical analysis of survey data. In particular, graphic displays permit the researcher to identify trends in the data and potential violations of assumptions that might be made in subsequent data analysis. She also discusses some strategies for smoothing data to identify trends over time.
In Chapter 6, Natalie Jomini Stroud and Dan Romer review univariate and exploratory data analytic strategies that may be useful to readers before progressing to more complex analyses. In addition, several fundamental techniques such as crosstabulation and correlation are discussed with specific examples that apply these methods to the NAES.
In Chapter 7, Romer reviews the use of linear and logistic regression for the analysis of cross-sectional data. He reviews the assumptions one makes in using these analyses to identify causal relations in survey data. He also discusses strategies for studying aggregate changes that unfold over time and the effects of voter experiences that might moderate these effects.
Chapter 8 by Kenski and Romer reviews strategies for the analysis of panel data. The NAES contains several panels that were designed to permit the study of important election events such as presidential debates. Strategies to permit stronger causal inferences from this design are discussed using examples from the NAES.
Chapter 9 by Romer provides an overview of the use of time series analysis for the NAES. This analysis uses data aggregated at the daily level to study effects of events during the election campaign or relations between variables measured on a daily basis. Many of the analytic techniques are borrowed from the literature on economic forecasting and may be more novel for readers than the approaches discussed in earlier chapters.
Chapter 10 by Dannagal Goldthwaite Young, Russell Tisinger, Kenski, and Romer examines the use of the NAES to study trends in segments of the population that might be too small to analyze in most crossectional surveys even with large samples. The authors provide examples of trend analyses in knowledge acquisition in late and early primary states as well as in late-night comedy show audiences (e.g., The Daily Show) during the 2004 election.
As noted earlier, a brief appendix of technical terms is provided so that readers can easily find definitions of concepts. The appendix is organized by chapter and by topic to help readers see the connections between concepts.
The codebook for the datasets found on the CD-ROM was written by Christopher Adasiewicz who also helped to coordinate the implementation of the survey.
NAES data are provided in the format used by the Statistical Package for the Social Sciences (SPSS), and most of the analyses we present use this program. However, we do not discuss specific details for using SPSS or other statistical programs for data analysis. Readers who are unfamiliar with SPSS or other statistical packages should consult the manuals of those programs for information about the procedures needed to run those programs.
We do not recommend the use of any particular statistical package for analysis of the NAES. For the analysis covered in Chapter 9, readers will want to use packages that have specific procedures for time series analysis. SPSS can accommodate all of the analyses if the package includes the Trends module. Other packages that can also be used for time series data include the Statistical Analysis System (SAS), STATA, Eviews, and Matlab.