For twenty-five years prior to the creation of ANES, from 1948 until 1976, the Survey Research Center and the Center for Political Studies of the Institute for Social Research carried out an unbroken series of national studies of the American electorate, covering all eight presidential and six midterm elections during these years. In 1977, the U.S. National Science Foundation (NSF) formally established ANES as a national resource to sustain and enhance the diversified data base that supports basic research on voting, public opinion, and political participation in the United States. This document describes the data collections that comprise the ANES Time Series (that is the Pre/Post-Election Studies in presidential election years and the Post-Election Studies in midterm years).

An Overview of the ANES Time Series Data Collections The Early Years: The "Michigan Studies"

The early studies may be characterized as the development and working through of a model of vote choice which placed heavy emphasis on attitudes toward the candidates, political parties, and major issues covered in the campaign. As opposed to the other major investigations of voting behavior which were roughly contemporaneous (the 1940 Erie County study; the 1948 Elmira County study) the Michigan studies a) were national in scope; b) were conducted at a single point in time just before and just after the election, as opposed to a number of waves during the entire campaign year; c) focused on the role of intra-individual social psychological processes, as opposed to the individual's social context -- group membership, community characeristics and so on -- or the role of mass media and interpersonal influence on vote decisions. Other themes explored in these early years were the reliability of measurement over time, and the relationship between citizens' attitudes and the behavior of their elected members of congress: representation.

Emphasis began slowly to shift to a recognition of the growing importance to political analysis of the development of a longitudinal data base. Hence, themes in the analyses of the Election Studies were the over-time measurement of partisanship; candidate evaluations, issue preferences, and political behaviors. With the Watergate years came an outpouring of scholarly work on support, or lack thereof, for the political system. And with a series of Election Studies, it was proving possible to confront the deeper issues of political socialization: what effects were generational, which life-cycle, and which resulted from the context of a particular slice of time? Also in this period, confrontation of the "Michigan model" with the rational choice model was the impetus for a growing amount of work on issues and their dimensionality and on voter participation.

The siting of the Michigan studies in these major lines of inquiry, which demanded the continuation of the Time Series for their analyses, provided the major intellectual justification for the first proposal to the National Science Foundation for long- term support. With the provision of this support, the intellectual agenda of the studies developed along a number of different lines, including a reincarnation, under slightly different conceptual cover, of the early interest in representation.

Studies of Congressional Elections

In 1978 the ANES Board designed and carried out the first Election Study under the auspices of the new long-term agreement with the National Science Foundation. The first study marked a watershed in the study of Congressional elections. By effectively translating the shared interests of specialists in Congressional elections into new instrumentation and a new study design, the ANES Standing Committee on Congressional Elections not only transformed the 1978 National Election Study, but brought about an explosion of research that led to a revolution in our understanding of Congressional elections.

Like its midterm predecessors, the 1978 survey interviewed a random cross-section of citizens following the November election. But in 1978 the sampling scheme also permitted the 2304 respondents to be disaggregated into small self-representing samples of 108 Congressional Districts. This design enabled a more powerful examination of the central themes of the study: the nature of the relationship between Representatives and their constituents, and the impact of (local) context on Congressional races. The 1978 survey also developed many new questions with these themes in mind. At the same time, ANES gathered, coded, and disseminated extensive candidate-level and district-level contextual data and validated measures of voter registration and turnout.

With slight adaptations and additions, features of the 1978 Study were incorporated into later ANES midterm studies, and, in somewhat abbreviated form, into later ANES presidential studies, although the Congressional District sample frame was used only once more, in 1980. Most of the 1978 expanded Congressional level content became core content for subsequent election studies, and the validation of voter registration and turnout was an important feature of a number of ANES studies that followed. Beyond advancing our understanding of Congressional elections, midterm Election Studies have also served other substantive purposes as well, and have regularly included questions on topical matters or other new or special content, Table 1.

Studies of Presidential Elections

In 1980, the ANES Board designed and carried out the first presidential election study under the auspices of the new long- term agreement with NSF. The 1980 Study was an ambitious and complex project, consisting of eight discrete survey data collections, distributed strategically over the course of the campaign year. The core of the study, and the main carrier of the complement of Time Series questions, was the Pre/Post- Election Study.

The pre- and post-election interviews focused on candidate evaluation and vote choice. Instrumentation on Congressional incumbents and candidates introduced in the 1978 National Election Study was also included in the 1980 post-election interview to enable systematic comparisons between voting for Congress in midterm and presidential years.

This traditional design -- pre-election interview plus post- election interview, concentrating on the determinants of the vote (including Congressional vote), and with an eye on sustaining the Time Series of core questions -- was also central to subsequent ANES studies of presidential elections. Each included new instrumentation, too, a reflection, in broad terms, of the intellectual vitality of the elections and public opinion fields, and, in narrow terms, of the success of the ANES Pilot Studies. (See the ANES Research and Development Studies.)

In several studies since 1980, experimentation with mixed-mode interviewing was introduced into the sample design, beginning with the 1984 National Election Study, in which a random half of the respondents were re-interviewed in person after the election and a random half were re- interviewed by telephone. This design furnished additional evidence on the relative merits of personal and telephone interviewing that went beyond the methodological experiment conducted in 1982, and the experiment was repeated in the 1996 Study post-election wave. The 1998 Study implemented experimentation to investigate possibilities for maximizing use of telephone interviewing, and most interviews in 1998 were conducted by phone, although traditional probability area sampling was used and every effort was made to obtain initial face-to-face contact. In 2000, design innovation with mixed mode was carried even further, so that the 2000 data actually represent two presidential studies: the core study preserved past commitment to probability area sampling and face to face interviewing, while, supporting the core study, the efficiencies of RDD sampling and telephone interviewing were used for a second set of respondents.

Beginning in 1992, there have also been studies integrating a panel component into the sample; in each study, 'fresh' cross-section cases combined with respondents interviewed once or more in previous studies together constituted a total representational cross-section. The content of such studies has reflected double duty. The data can be used in traditional fashion to support cross-sectional analysis of the electorate and, in addition, can also be analyzed in panel format because there is a very extensive set of "lagged" measures for a sizeable portion of respondents. ANES prepared and released data files to support both kinds of analysis.

Maintaining Core

ANES is, in part, an institutional device for protecting and maintaining the Time Series of core questions relevant to national elections, public opinion, and civic participation. The core Time Series data serve two fundamental purposes.

First, the Time Series enables the National Election Studies to register the effects of exogenous shocks to the political system. Second, the existence of the full series of Election Studies has opened up several important lines of inquiry into the nature and causes of political change. Although the meaning of any small, year-to-year fluctuation may be obscure, by accumulating over-time measures in the public's responses to political phenomena, social scientists can uncover and understand electorally-relevant secular trends in the public's perceptions of and participation in politics. Scholarly work that has exploited the ANES Time Series include studies of electoral change, support for third party candidacies, change in partisan attachments, alteration in the importance the American public assigns to national problems, change and continuity in the public's views on race, the ebb and flow of conservativism, and fluctuations in the American public's participation in political life. These works share an interest in the dynamics of change and the presumption that vote, opinion, and participation are shaped by ongoing alterations in the political, economic, and social environments that citizens face. Research on political and social change will continue to prosper; indeed, we believe that the most dramatic progress in our understanding of American mass politics over the next decade may well come here, led by the availability of ever-lengthening Time Series data provided by ANES.

Study Design and Sampling

For all of the ANES studies described in Table 1 (and for each new study that is planned) there are three main stages of implementation: drawing the sample; conducting the interviews; and preparing the data and documentation for dissemination to the user community. Each of these steps was, and continues to be, a joint effort between the ANES Project Staff and the various units that comprise the Survey Operations Unit of the Survey Research Center of the Institute of Social Research. The overall study design, discussed briefly below, has been consistent over the entire time period, with occasional modifications, and is the responsibility of the Board of Overseers.

Design

With the exception of several studies which experimented with telephone interviewing, the Time Series data collections have traditionally been face-to-face interviews. In presidential election years there have been two waves: a pre-election wave usually beginning the day after Labor Day and ending the day before the November general election, and a post-election wave of re-interviewing beginning the day after the election and generally continuing until late December to mid-January, with 50-60% of the interviews typically completed by Thanksgiving, 90% or more by Christmas. In mid-term elections, the interviewing takes place in a single wave, beginning immediately after the November election.

The following describes the normal sampling process, excluding the notable exception of the 2000 Study, for which half of the sample was obtained by RDD sampling. Generally, however, the sampling process begins when, following the specifications provided by the ANES Project Staff, the Survey Research Center Sampling Section draws the samples from their national sample frame. The sample universe is U.S. households in the continental United States. (The most recent SRC National Sample Frame extends the sample to Alaska and Hawaii).

SRC National Samples are multi-stage area probability designs. That is, in a series of hierarchical steps, geographically defined sampling units of decreasing size (size is measured by the most recent U.S. census data) are selected with probability proportionate to their total number of occupied housing units. Prior to the first stage selection, the primary areas (SMSA's, counties, and county groups) are stratified within each of the four Census regions by geography and size. One primary sampling unit is selected from each stratum using a controlled selection procedure. Sixteen of the 84 selections comprise the entire SMSA and are designated as "self-representing."

The second stage sampling units are comprised of Census Blocks (in metropolitan primary areas) or Enumeration Districts (in some rural areas). Sampling of second stage units is performed with probability proportionate to number of occupied housing units in the area. Before second stage selection takes place, the second stage sampling units are stratified by geography, size, and in 36 of the 84 primary areas (where the data are available) by median per capita income. Within the 6 largest primary areas, between 6 to 25 segments are selected; in all other areas, 6 secondary selections are made.

At the third, and final stage, the SRC field staff lists all the housing units within each segment and selects a random subset. Once a household is selected, the interviewer lists all eligible adults residing there. (For ANES studies, eligibility is defined as being able to vote at the November general election: i.e., citizens of voting age.) To insure that all people in the household have an equal probability of being chosen, in the final stage the interviewer uses a selection table to choose randomly the respondent to be interviewed.

In addition to the 2000 Study's introduction of an RDD half-sample, study designs in some years have introduced other occasional modifications into this process. These are noted in Table 1.