Ten years after the 1978 ANES Congressional election study set in motion an avalanche of research on incumbency, casework, representation, accountability, and election finance, our understanding of Senate elections and the institution itself remained, by comparison, impoverished. The problem was not that Senate elections were of little theoretical interest; nor was the research community indifferent. The problem, rather, was the absence of appropriate data.

National surveys -- including those conducted by ANES -- that were ideal for the study of voting in presidential and House elections could not support the analysis of voting for the Senate. Because only one-third of Senate seats are up for election in any year, only a subsample of all respondents in any given national survey reside in states holding Senate contests. Moreover, because national samples draw respondents predominantly from the most populous states (as one would hope and expect), the contributions of small state elections are obscured in analyses of standard national samples. This is a serious problem because Senator-constituent relations and campaign styles vary substantially with size of state. A combination of sampling problems, question quality, diverse survey practices, and negligible or superficial coverage of Senate elections also made it impossible to sustain the systematic empirical analysis of elections to the U.S. Senate from the spattering of state surveys that existed. This led ANES, with support from the National Science Foundation, to embark on the 1988-90-92 Senate Election Study, a data collection hand-tailored to the special characteristics and theoretical opportunities presented by elections to the U.S. Senate.

The ANES Senate Election Study came in three installments: in 1988, 1990, and 1992. Each study involved a 35-minute, post- election telephone interview with a fresh random cross-section of citizens of voting-age drawn from each of the fifty states. Because of our interest in the full population of Senate elections, each election in each state was covered with samples of approximately equal size. An average of 62 respondents were interviewed from each of the fifty states in each of the three elections for a total of 9253 interviews.

Several aspects of this study design were crucial to meeting the project's theoretical and substantive objectives. Each year's sample was stratified by state to maximize variance in the types of candidates and campaigns and to ensure that we interviewed citizens from small states as well as large ones. The 50-state design in conjunction with the three waves of the study also provided the analytical leverage essential for understanding the impact of the six-year term on the dynamics of citizen interactions with their U.S. Senators. Interviewing over three elections in states with and without Senate elections permitted the over-time mapping of how constituents' relationship with their Senator changed two, four, and six years into the term.

Beyond the data gathered from the interviews, the Senate Election Study also provided a large number of state-level contextual variables. These data included presidential and Senate election returns; social and economic characteristics of the state; demographic and political qualities of the Senators and Senate candidates; and characteristics of the Senate campaign. By interviewing over the course of three elections, we were also able to ensure that the conclusions reached generalized beyond a particular election and beyond the particular Class of seats that happened to be contested in that year. Moreover, pooling cases across the three elections provided an ample number of cases to investigate the motivation behind split-ticket voting and the reasons for split Senate delegations -- thereby illuminating important aspects of divided control of government. The 1988 and 1992 installments of the study, when used in conjunction with the 1988 and 1992 Pre/Post-Election Studies, also permit comparisons of voting across three offices: House, Senate, and President. Finally, because each of the 50 state samples is self-representing, the Senate Election Study data can be used to generate state-level estimates of partisanship, ideology, opinions on policy questions, and more. (See Table 1 for more information.)

One consequence of the party reforms that were adopted in the early 1970s was a doubling in the number of states holding presidential primary elections. Once merely trial heats in which party leaders could assess how candidates would likely do in November, primaries became the basic mechanism for selecting each party's presidential nominee. Nomination battles were not only often more competitive and more interesting than the general election campaigns, but they provided a unique opportunity to study how Americans learn about politics and make political choices in multi-candidate races in the absence of partisan cues.

To understand the nature of citizen choices during the nominating process, ANES carried out three coordinated data collections in conjunction with the 1980, 1984, and 1988 presidential year election studies. In 1980, the traditional Pre/Post-Election Study was supplemented by two panel studies that were conducted over the course of the election year. Each was devised to support description and explanation of the dynamics of individual political change during the presidential campaign. Thus each focused extensively on citizens' interest in and attention to the campaign, their assessment of the major contending candidates, their allegiance to the parties, and their preference among the candidates. The Major Panel began in late January (about a month before the New Hampshire primary) as a national cross-section. Respondents were then reinterviewed in June immediately after the last set of primary elections; interviewed again during September, and questioned a fourth and final time (this time over the telephone) immediately after the general election. The Minor Panel began with an independent cross-section interviewed in April (n=965), in the midst of the primary season during the three-week "window" following the early primaries, and ended with a post-election telephone re-interview. The panels were deployed so that they could be analyzed either independently or in combination with the traditional Pre/Post- Election Survey to provide a succession of snapshots of the electorate at key moments across the campaign. (See Table 2.)

In 1984, ANES implemented the Rolling Cross-Section Study (hereafter, RXS). RXS consisted of 46 independent weekly cross- sections. Interviewing began in early January, before the formal campaign got underway, and ended in mid-December, after the campaign had subsided. Each cross-section was made up of an independently drawn sample of Americans of voting-age, questioned over the telephone, and selected through an adaptation of the two-stage random digit dialing procedure (as suggested by Waksberg). On average, slightly more than 75, 40-minute interviews were completed each week, cumulating in a total N of 3496. RXS was designed to provide a general purpose capacity to examine the impact of the campaign on the public's evolving beliefs and preferences. In this respect, the 1984 RXS complements the Panel Studies undertaken in 1980. The Panel Studies are well-suited to the description and analysis of change in individual citizens' opinions between key junctures in the campaign, and how change in one opinion influences changes in others. But because of the large gaps between panel interviews, it was virtually impossible to isolate how specific campaign events and the media's interpretation of those events affect citizens' beliefs and values. For that, continuous monitoring was required, and RXS supplies it. (See Table 2.)

Finally, in 1988, ANES conducted a Study of the Presidential Nomination Process (the Super-Tuesday Study) designed to further our understanding of how learning and vote choice occur in the context of primary elections. A particular focus of the study was how voters learn about the viability of the candidates in the field, how rapidly this learning takes place, and how this information is used in vote decisions. The 1988 study was designed around the Super-Tuesday elections held on March 8. Over the seven week period preceding Super-Tuesday, ANES interviewed by telephone (again using random sampling techniques) 2117 citizens of voting-age residing in the 16 states holding primary elections on Super-Tuesday. To permit analysts to unpack the impact of the campaign as well as political events occurring earlier in the primary season, the sample was released as seven independent weekly samples, with a larger number of interviews taken in the week between the Iowa caucus and the New Hampshire primary and during the three weeks preceding Super-Tuesday itself. Brief re-interviews were conducted with 80 percent (1688) of the pre-election respondents in the two and a half weeks immediately after the day of the primary.

The pre-election survey included questions on: attentiveness to the campaign and the media's coverage of the campaign; perceptions of each candidate's viability, electability, ideology, character, and policy positions; vote intentions; a full array of demographic variables; and one question that asked for the names of newspapers that the respondent read for information about politics. This last item was included to aid those who wished to match the content of newspaper coverage of the campaign with respondents' views on candidates and issues. The post-election questionnaire included questions on turnout, vote, nomination preference, and presidential preference.

Many questions that analysts have about political behavior involve change over time. How do political campaigns affect intention to vote? How has the nation's involvement in a large- scale international crisis affected people's evaluation of the President? Are the attachments of individuals to political parties weakening over time? How does the performance of the economy affect vote choice? There are many ways, using the entire body of NES work, to examine change over time.

First, surveys themselves have often asked for retrospective reports, for recall of past voting behavior, or past partisanship. Second, pooling of the 1952-1992 National Election Studies also permits long-term trend analyses, based on sample means or variances. A third way is to examine the difference between the attitudes and behaviors of different age- cohorts within a cross-section sample (or across successive cross-sections). Still each of these approaches has serious limitations for getting at change over time. Faulty memories and selective recollection, for example, are among the problems that threaten the quality of data that an individual has recalled at the invitation of an interviewer. Trend analyses (comparing aggregate means or variances) leaves the individual, who is the source of the change, entirely out of the equation. And cohort analysis begs the question of what were the individual-level processes that produced the change.

Repeatedly interviewing the same individuals, on the other hand, reduces problems of recall, and allows the analyst to ascertain the individual level sources of change. The panel method most closely approximates the archetype experiment where subjects are observed in a "before condition," there is a treatment of some sort, and respondents are observed again in an "after condition."

ANES has conducted several panel studies over the years. To begin with, of course, all of the Presidential year Pre-Post studies are two-wave panels: the same individuals, with some attrition, are interviewed before and after the election. It is fair to say, however, that the panel aspects of these studies have not been very fully exploited. The vast majority of questions, with the significant exception of party identification in some years, and vote intention/actual reported vote for President, are not repeated from the Pre- to the Post-Election waves. Also, with only two successive waves of interviewing, it is difficult to distinguish "real" change from measurement unreliability.

The ANES Pilot Studies are three and sometimes four-wave panels. And there is a conscious effort to repeat questions from one wave to the next. But, the focus of these studies has very largely been research and development, with the significant exceptions of the 1991 and 1993 Pilot Studies.

The studies listed in Table 3 are those ANES studies on which the analytic focus has been on individual level change over time. The first is the 1956-58-60 Panel Study. The earliest themes of the election studies of the early 1950's have to do with the stability of partisan identification over time and the role that short term and long term forces play in the structuring of opinion and vote choice. After the early findings of The American Voter, based on the cross-sectional studies of 1952 and 1956, it became a matter of pressing analytic interest to determine whether partisanship was truly stable enough to underlie and organize individual response to candidates and issues. There was a great deal of interest too in who it was that was moving in and out of the electorate in presidential and off-years, in "surge and decline" and the "floating voter." The decision to press for an investigation of these matters at the individual level led to the first panel study.

The years between 1960 and the initiation of the second Panel Study in 1972 saw an explosion of research on voting behavior. A major theme of this research was the meaning of the instability of responses to issue questions, as opposed to the demonstrated stability on partisanship over time. One motivation for empaneling respondents to the 1972 study to form the 1972-74- 76 Panel Study, was to explore the stability of partisanship and issue attitudes over time. Also of interest was understanding the public response to Watergate and tracing the origins of the decline in system support.

In 1980, the traditional pre/post design was supplemented by two Panel Studies that were conducted over the course of the election year, with the objective of understanding the nature of citizen choices during the nominating process. Each of the two panels was designed to support description and explanation of the dynamics of individual political change during the presidential campaign. Thus, each focused extensively on citizens' interest in and attention to the campaign, their assessment of the major contending candidates, their allegiance to the parties, and their preference among the candidates. The Major Panel began in late January (about a month before the New Hampshire primary) as a national cross-section. Respondents were then re-interviewed in June immediately after the last set of primary elections; interviewed again in September, and questioned a fourth and final time (this time over the telephone) immediately after the election. The Minor Panel began with an independent cross- section interviewed in April in the midst of the primary season during the three-week "window" following the early primaries, and ended with a post-election telephone re-interview.

When the 1990 Post-Election Study was fielded, there was no provision for a panel. The 1991 Pilot Study was slated to interview, in the summer of 1991, a subset of 1990 Election Study respondents, for research and development work in anticipation of the 1992 Election Study. But, 1991 was also the year of the Persian Gulf War, and in order to examine the political consequences of this event, the ANES Board of Overseers decided to turn the 1991 Study into the second wave of a panel, reinterviewing by telephone as many of the 1990 respondents as possible. By 1992, interest had turned to explaining the dissipation of the war's effects, and the disintegration of what had seemed an invincible Bush coalition. The Board of Overseers decided that approximately half of the respondents to the 1992 Election Study should be drawn from respondents to the 1990 Election Study. Most of these respondents were also interviewed in 1991, producing a three-wave panel study. (Technically, there are 4 waves, because the 1992 Election Study has both a pre- and a post-election interview).