The early studies

A SRC national survey of foreign policy attitudes in autumn 1948 happened to ask about vote intentions in the upcoming presidential election (Miller Reference Miller1994). The commercial polls incorrectly forecast a Dewey victory that year when Dewey’s large early lead caused them to end their polling too quickly. By contrast, the SRC pre-election survey found the race to be too close to call, and a majority of their post-election respondents voted for Truman. Moreover, a subsequent report (Mosteller et al. Reference Mosteller1949) for the Social Science Research Council found SRC’s greater accuracy was also due to its sampling method: its using probability sampling rather than the quota sampling that commercial pollsters were employing. That success gave the quality of SRC’s polling increased credibility.

The only previous academic national election survey was the National Opinion Research Center’s 1944 study, and the lone analysis of that at the time was Sheldon Korchin’s (Reference Korchin1946) path-breaking unpublished psychology dissertation. The sole book based on an election survey was The People’s Choice (Lazarsfeld, Berelson, and Gaudet Reference Lazarsfeld, Berelson and Gaudet1944), analyzing an innovative 1940 seven-wave panel study with control groups, conducted in Erie County, Ohio by Columbia University’s Office of Radio Research (which later became the Bureau of Applied Social Research). The book’s focus was sociological, mainly considering socio-demographic predictors, interpersonal influence, cross-pressures, and the effects of social groups, as well as analyzing voter activation, reinforcement, and conversion across the election year.

Campbell and Kahn (Reference Campbell and Kahn1952) reported the results of the 1948 Survey Research Center study in The People Elect a President, which was the first book-length report on a national election survey. It described how voters resolved their original indecision, and it analyzed the demography of voters, their perception of the candidates, and their attitudes on election issues.

Campbell also mined a 1951 SRC survey for an article in which he and political scientist George Belknap analyzed the relationship between party identification and foreign policy attitudes (Belknap and Campbell 1951–1952). The party identification question was simply how people would vote “if a presidential election were held today,” but that apparently led to Campbell becoming interested in developing a measure of party identification that was not directly tied to voting. Campbell then devised the “generally speaking” question that was used starting in the 1952 CPS election study (Converse Reference Converse2006, 608).

Preparation for a 1952 election survey

After World War II, the Social Science Research Council’s (SSRC) Committee on Political Behavior (CPB) helped channel foundation funds to further the development of the behavioral approach (Hauptmann Reference Hauptmann2011, Reference Hauptmann2016a). The success of the SRC’s polling led to Campbell’s appointment to the Committee. The Committee held an interdisciplinary conference in Ann Arbor in late summer of 1949, co-hosted by ISR director Ren Likert and political science department head James K. Pollock. Reporting on the wide variety of topics considered at the conference, a report in the SSRC newsletter (Heard Reference Heard1949, 2) stated that the ISR’s survey research procedures were seen “as having much potential value in the study of human phenomena,” that the 1952 election “was mentioned as a suitable object of collaborative study by political scientists, psychologists, and others,” and that “a sample of 100 counties might be studied.” Such a study was described as including not only survey techniques but also investigating characteristics of local party machines, preelection campaigns, and historical data including election returns.

In early spring 1950, Campbell shared with Pollock his draft proposal for a massive four-year project including participant observation of politician behavior, mass observation of behavior, intensive interviewing of key individuals in the community, collecting historical and current records, and sample surveys, at $49,600 per year (a 4-year total equivalent to $1,988,000 in 2016 dollars). ^{Footnote 1} Pollock responded, “I see no reason why you should not go ahead to circulate this” to the CPB, but ominously added, “I have some doubts about the organization and control of the survey.” ^{Footnote 2} In any case, in 1951 Pollock, who himself had analyzed voting data, ^{Footnote 3} requested a new faculty position for someone who did the ISR type of research (Jackson and Saxonhouse Reference Jackson and Saxonhouse2014, 16).

Already anticipating a 1952 election study, Campbell needed to put together a staff. Social psychology graduate student Gerald Gurin was to be the 1952 study director, but, with George Belknap leaving, Campbell desired to bring in a political science perspective. A Syracuse University graduate student with political science interests had received top grades in sampling, attitude change, scaling, and organizational behavior courses in Ann Arbor in winter and summer 1950. Campbell apparently heard about the political scientist who had done so well, and at the end of that summer he offered that graduate student, Warren Miller, a job. Miller responded that he wanted to return to Syracuse to work on his dissertation, so Campbell told him to come back the following year (Miller Reference Miller1988).

Miller joined the SRC in 1951 as an assistant study director, working on possible directions for the 1952 election study. ^{Footnote 4} His only involvement with the political science department at that time was attending department picnics because his first wife, Mildred [“Kip”], was Pollock’s secretary (Miller Reference Miller1988). In Miller’s first article (Janowitz and Miller Reference Janowitz and Miller1952) after joining SRC, he used the SRC 1948 survey to show the limitations at the national level of the Columbia School’s explanation of the vote by their Index of Political Predisposition that was based solely on voters’ religion, class, and urban/rural residence (Lazarsfeld, Berelson, and Gaudet Reference Lazarsfeld, Berelson and Gaudet1944).

In February 1952, Campbell submitted to the Committee on Political Behavior a reduced request at half the cost of the original proposal: $90,000 (equivalent to $803,000 in 2016 dollars) for a double-interview national survey on the 1952 election, without the other research elements that were part of his 1950 preliminary proposal. ^{Footnote 5} In March, SSRC President Pendleton Herring (who was also the incoming APSA president) wrote Pollock, acknowledging Pollock’s letter supporting Campbell, indicating that the Carnegie Corporation wanted the grant to be under the CPB, and expressing happiness that “all is well on this at Michigan.” ^{Footnote 6}

However, all was not well at Michigan. As recounted later by Philip Converse (Reference Converse1991), “The powers that were in the political science department were very angry that this kind of work was going on at ISR because they felt it was on their turf and being done by a bunch of psychologists who didn’t know anything about politics.” In April, Pollock wrote APSA executive director Edward Litchfield complaining that the project was to be administered by a social psychologist, without direct participation of political scientists, and saying that he was “deeply concerned” and “not interested in just another survey; that unless active groups of political scientists who are knowledgeable in field studies and state and regional situations cannot [sic] be associated with the survey that we are not interested;” ^{Footnote 7} Pollock also wrote Herring, objecting that “a rather limited survey” was intended. ^{Footnote 8} The SRC saw this as Pollock trying to get the SSRC to reconsider the decision to give ISR the grant and instead to give it to him and his colleagues (Miller Reference Miller1988). Campbell set up an advisory committee (including Pollock) and a working group including other political and social scientists at Michigan to advise the SRC’s core group, ^{Footnote 9} but that did not mollify Pollock.

In May, Herring wrote Pollock that the grant’s terms could not be changed, ^{Footnote 10} and the Carnegie Board adopted a resolution giving the SSRC $90,000 for the “cost of non-political and non-partisan research on political behavior.” ^{Footnote 11} The SSRC then gave that grant to the Michigan SRC for its national survey, ^{Footnote 12} making the SRC rather than the department the home of the election studies.

It can be considered appropriate for Pollock as department head to defend disciplinary turf and to support the research of his junior colleagues who wanted a non-survey component. However, the Committee on Political Behavior was seeking to build a type of political research that was different from the “old-line” political analysis that Pollock represented. As Hauptmann (Reference Hauptmann2016a, 184) concluded, “Making the grant to the SSRC’s CPB put control over the project out of Pollock’s reach—something Gardner [from Carnegie], Campbell, and Herring all agreed was desirable.” ^{Footnote 13} In any case, this contretemps led to considerable distrust of Pollock by the SRC, especially when, after failing to stop them from obtaining the grant, Pollock declared the SRC people “out-of-bounds” for the political science faculty (Miller Reference Miller1988).

After the 1952 election

Miller scored two American Political Science Review articles in 1953: a solo piece on party preference and issues in the 1948 and 1951 surveys, and a collaborative article with the SRC research team on issues and the vote in the 1952 data. He finished his dissertation, Issue Orientation and Political Behavior, analyzing the limited issue questions in the 1952 election survey to test five conditions that he posited were necessary for issue commitment to affect voting.

Campbell, Gurin, and Miller (Reference Campbell, Gurin and Miller1954) reported the results of the 1952 SRC study in the book The Voter Decides, introducing the notion of predispositions as they compared the roles of issue, candidate, and party orientation in individual voting decisions. The data analysis consisted of cross-tabulations of variables, occasionally with a control variable.

Also in 1954, the Columbia group published their book on the 1948 election, Voting (Berelson, Lazarsfeld, and McPhee Reference Berelson, Lazarsfeld and McPhee1954), a sociological analysis of a panel survey of Elmira, New York. Voting continued to emphasize sociological variables but also moved into considering political issues and perceptions. In contrast to earlier political science studies that focused on what happened in elections, the Columbia and Michigan studies both provided analysis of why it happened. However, these studies differed in their choices of independent variables, with the Columbia researchers using primarily sociological explanations and the Michigan researchers emphasizing social-psychological (attitudinal) explanations. ^{Footnote 14}

Meanwhile, Warren Miller received his doctorate of the social sciences degree from Syracuse in 1954, and left the SRC to become an assistant professor in political science at Berkeley. With Miller gone, the SRC conducted a small pre-election survey on the 1954 congressional election. However, they did not conduct a post-election wave, so their only dependent variable was “probable vote” based on their attempt to estimate whether the person would turn out on Election Day. The results were published in an ISR monograph (Campbell and Cooper Reference Campbell and Cooper1956).

During this period, the SRC convened summer workshops on political behavior in 1954 and 1958 and began hosting researchers from other universities who wanted to analyze the election surveys. Those moves helped create a scholarly community that was supportive of continued funding for the election studies.

Preparation for a 1956 election survey

In February 1955 Campbell sent the SSRC Committee on Political Behavior a memo in which he outlined thoughts for a 1956 study, ^{Footnote 15} but the Committee decided to have only an informal relationship to that project. ^{Footnote 16} Instead, Campbell secured funding from the Rockefeller Foundation for the 1956 study, ^{Footnote 17} with the support of social psychologist Leland DeVinney on the Rockefeller staff helping to overcome the reluctance of other staff members (Hauptmann Reference Hauptmann2011). Fortunately, the dramatics at Michigan involved in getting the 1952 grant were apparently not repeated.

Again needing a study director for an election study, Campbell lured Miller back to Michigan. Miller returned in 1956 with an appointment as an assistant professor in political science. That academic appointment may have become feasible because of the credibility that Miller gained from his position at Berkeley.

The third member of the research team was Philip Converse, who had come to Ann Arbor from western Michigan in the fall of 1952 when his wife Jean was working on her teaching credits at the university. She worked on the campus campaign for Adlai Stevenson with some people in Michigan’s social psychology program, so Phil became acquainted with some of the leading social psychology professors in the country.

After hearing Warren Miller give a speech on the 1952 election survey, Phil approached him and insightfully asked if they ever thought of reinterviewing the same respondents over time (Weisberg Reference Weisberg1983). Converse had been in the creative writing program of the State University of Iowa, where he obtained a master’s degree in English literature. Having found someone with writing skills and interest in the area, Miller encouraged him to do graduate work. Given the nature of the Michigan departments at the time, Miller reacted more favorably to Converse going into the social psychology program than the political science program, which was not behavioral (Converse Reference Converse1991).

The Converses spent their next year in Europe, with Phil studying at the University of Paris. On their return, he wanted to enter Michigan’s social psychology doctoral program, but it had a prerequisite of a master’s degree in either sociology or psychology. He chose sociology, and he gained experience in survey research because the sociology master’s program required students to take the Detroit Area Study (DAS) course in which students prepared the survey and conducted interviews themselves in Detroit.

Converse obtained his master’s degree in sociology in 1956, and started the social psychology doctoral program. He was studying under Ted Newcomb, who told Converse it was time for him to get a research position and suggested checking with Angus Campbell at the ISR (Converse Reference Converse1991). Miller was in frequent mail contact from Berkeley with ideas for the 1956 study that he gained from conversations with scholars in the Bay Area (Burns Reference Burns2006, 8), but Campbell needed an assistant study director who could handle preliminaries on the 1956 survey before Miller’s return to Ann Arbor, so Campbell hired Converse for that role.

By the time that Miller came back to Ann Arbor in summer 1956, Converse was already using his DAS experience to work on the 1956 study (figure 1), which surprised Miller who had expected Converse to be hired mainly for his writing abilities (Miller Reference Miller1988). Converse was assigned responsibility for developing questions on the influence of social groups on the vote to replace Jerry Gurin’s group questions that were used in the 1952 study. Converse’s new questions became the core of his dissertation, Group Influence in Voting Behavior, in which he analyzed the voting distinctiveness of labor union members and households, Catholics, blacks, and Jews. He completed the dissertation in 1958, along with an analysis of social class that led to his first publication (Converse Reference Converse, Macoby, Newcomb and Hartley1958).

Figure 1 From left-to-right: Philip E. Converse, Warren E. Miller, and Angus Campbell, circa 1956

"Survey Research Center National Elections Study; BL005604.” http://quod.lib.umich.edu/b/bhl/x-bl005604/bl005604. University of Michigan Library Digital Collections. Accessed: July 10, 2016.

The research team became complete when Donald Stokes showed up unannounced in Ann Arbor in 1955, having come there on a two-year SSRC pre-doctoral fellowship to study social science statistics while working on his doctoral degree in political science at Yale. Stokes paid a courtesy call on political science department head James K. Pollock, who firmly told him that he would never be allowed to teach in the department (Weisberg, Macdonald, and Rabinowitz Reference Weisberg, Macdonald and Rabinowitz1998). Fortunately, the SRC provided Stokes a more welcoming home. By mid-April, Campbell was meeting weekly with Converse, Stokes, and Norwegian visitor Henry Valen for preliminary discussion of the 1956 survey. ^{Footnote 18}

Stokes (figure 2) became an assistant study director at SRC in 1957, working on the 1958 election study. Soon he was analyzing the 1952 and 1956 election studies for his dissertation, Partisan Attitudes and Electoral Decision, which he completed in 1958. He combined answers to the open-ended questions into six summary measures relating to issue and candidate responses, and then analyzed the relative impact of these six “components” on the vote decision. His first article was from that work, appearing in the APSR in 1958 (Stokes, Campbell, and Miller Reference Stokes, Campbell and Miller1958; see also Campbell and Stokes Reference Campbell, Stokes, Burdick and Brodbeck1959).

Figure 2 Donald E. Stokes, circa 1971

“Portrait of Donald E. Stokes; HS6287.” http://quod.lib.umich.edu/b/bhl/x-hs6287/hs6287. University of Michigan Library Digital Collections. Accessed: July 10, 2016.

Criticisms of the early SRC work

The reviews of the early SRC books were at best mixed, with many written by sociologists who were more positive to the Columbia team’s sociological approach. For example, the SRC book on the 1948 election book drew a very negative review by sociologist Shirley Star (Reference Star1953) of the University of Chicago’s National Opinion Research Center (NORC) and a collaborator on the pioneering study of The American Soldier (Stouffer et al. Reference Stouffer, Suchman, DeVinney, Star and Williams1949). Star called the SRC book unoriginal, unexciting, pedestrian, unimaginative, and “ending with a highly proficient research sterility” (Star Reference Star1953, 285). ^{Footnote 19} Additionally, she bemoaned the lack of partial-correlation analysis to control some of the demographic factors. Fortunately for the Michigan group, their 1952 study was funded and executed before the reviews of The People Elect a President were published.

The SRC published The Voter Decides in 1954, but Miller (Reference Miller1988) recalled that Campbell saw it as a failure because his fellow psychologists at the American Psychology Association meetings did not think it had made any contributions to psychology. To rub it in further, that same year the Columbia group published its favorably received book on Voting (Berelson, Lazarsfeld, and McPhee Reference Berelson, Lazarsfeld and McPhee1954).

While less harshly worded than Star’s review of the SRC 1948 book, most reviews of The Voter Decides disputed its ultimate significance. Sociologist Robert Bower seemed to dismiss the book’s importance when he wrote that “the large contribution of the Michigan work lies in their detailed elaboration of subjective variables” (Bower Reference Bower1954, 787). Richard Wahl’s (Reference Wahl1956) review in the American Journal of Sociology and Joseph Harris’s (Reference Harris1955) review in the American Political Science Review both started off positively, but ended by challenging the usefulness of the enterprise. Wahl finished that “neither the method nor the indexes have resolved the problem of separating voters from non-voters and of properly allocating the persons who are undecided about candidates” (Wahl Reference Wahl1956, 495–496). Harris also concluded by casting doubt on the potential of the approach: “As the authors point out, these factors [partisanship, issue orientation, and the appeals of the candidates] are closely related to one another and are not neatly separated in the mind of the voter. For this reason, it is questionable whether this technique can ever provide fully satisfactory explanations of voter motivation” (Harris Reference Harris1955, 228).

NORC Sociologist Peter Rossi’s (Reference Rossi, Burdick and Brodbeck1959) important review essay on the early survey-based work on voting behavior repeated some of these criticisms but treated them as matters that could be remedied. He criticized The Voter Decides because the variables it used were all at the same level, arguing, for example, that “it helps us little to know that voters tend to select candidates of whom they have high opinions” (Rossi Reference Rossi, Burdick and Brodbeck1959, 41). He also criticized the lack of sufficient tables that held demographic variables constant in assessing the importance of the book’s motivational variables. Another matter he raised was the “static character” of the study’s design that did not examine how candidate preference and issue attitudes evolved over the campaign. Rossi compared The Voter Decides with Columbia’s Voting book, saying of the SRC book that “the analysis is less thorough; its conceptualization is less rich, but the authors’ conclusions are more firmly supported by their data” (53–54). ^{Footnote 20} In the same book as Rossi’s essay, two psychologists pointed to further matters that were not considered in The Voter Decides: its not taking familial influences on the vote into account (C. Wahl Reference Wahl, Burdick and Brodbeck1959, 270) and the lack of a “baseline” by which to assess changes caused by an election (Brodbeck Reference Brodbeck, Burdick and Brodbeck1959, 416).

Additionally, sociologist David Gold (Reference Gold1958) eviscerated SRC’s monograph on the 1954 election. He criticized the lack of a formal model of analysis and its “neglect of the crucial question of the time order of the variables” (Gold Reference Gold1958, 216). As to party identification, Gold said, “just what the knowledge that a self-designated Republican will tend to vote Republican and a self-designated Democrat will tend to vote Democratic contributes to the understanding of political behavior escapes me!” (215). He emphasized that party identification is probably an intervening variable between social-demographic predictors and the vote, so the relationships between partisanship and those sociological variance needed to be interpreted.

The challenges for the Michigan authors

These reviews pointed to important limitations of the early SRC work. The authors had some interesting findings, but they had not conquered the puzzle of explaining voter motivation. Lacking a logic of data analysis, they had not dealt with the time order of their predictors. They had not shown that party identification was more than tautological in its relationship to the vote. Their sampling was sophisticated, but their data analysis was limited to cross-tabulations. The SRC election surveys were also being criticized for omitting sociological factors, not capturing the dynamics of change, ignoring the legal-institutional setting of elections, and not discussing the consequences of elections for the political system. If the authors were to prove the value of their approach, they would have to move to the next level in these many regards.

The reviews were correct in describing The Voter Decides as deficient in theory. The book had presented attitudes on party, candidates, and issues as three “orientations,” simply showing how they were related to each other and to the vote, both separately and together. This was a departure from the Columbia School’s reliance on sociological predictors, but The Voter Decides went too far in downplaying socio-demographics. It was necessary to find a way to bring the sociological predictors back in, as well as dealing better with how party identification, attitudes toward the candidates, and issue attitudes interrelate. Campbell had studied with Kurt Lewin at Stanford University, and Lewin’s field theory became part of how the Michigan authors would deal with these matters. Additionally, for one of Converse’s social psychology classes, he wrote a seminar paper in which he combined the social-demographic variables and The Voter Decides’ orientations into a comprehensive explanation. For that paper, Converse devised a “funnel of causality” metaphor, with the social-demographics at the mouth of the funnel, leading to the long-term predisposition of party identification, which in turn affected short-term issue and candidate predispositions, all of which affect the vote (Converse Reference Converse1991; Converse Reference Converse2006, 606).

A crucial element was their use of party identification. Converse (Reference Converse1991) emphasized that the party identification concept and measure were the “conceptual center” in their early work. However, “various cat calls, from the political science world, including the local department, that said this party identification stuff is real rubbish, that people are telling you how they are going to vote. So it was very important to know whether Democrats were continuing to consider themselves Democrats while voting for Eisenhower a second time.” The crucial concern was whether the Democrats would maintain their marked lead in party identification in 1956 regardless of the Eisenhower presidency, since that would provide a key test of the concept’s differentiation from vote. Proof of the pudding was provided when the 1956 study “basically proved [that] to be true” (Converse Reference Converse1991).

Furthermore, the authors recognized the need to ratchet up the statistical analysis. Converse used correlation coefficients in his dissertation when he looked at social group voting. Stokes contributed to the enterprise his knowledge of social science statistics. He performed principal component factor analysis on the vote components (Campbell, Converse, Miller, and Stokes Reference Campbell, Converse, Miller and Stokes1960, 102 fn. 12, 132 fn. 6) and regression analysis for his first articles, both of which were very difficult to calculate in the pre-computer era.

One more intellectual issue was introduced by the publication of Anthony Downs’ (Reference Downs1957) An Economic Theory of Democracy. The significance of its spatial model of voting was recognized early, so it was important that it be taken into account. Stokes’ mathematical background permitted him to address the Downs model.

Outlines

Early outlines of the book, dated April and May 1957, have been preserved ^{Footnote 22} and are posted online with the supplementary material for this article. The first was an April 22 single page “outline for discussion” for “The Book,” which listed 10 chapters with only barebones titles, such as “the parties” and “group memberships.” There is no way to tell whether this was the proposal that Campbell rejected, but it was so brief (less than 30 words total) as not to look like a serious effort.

That skeletal version was quickly superseded by an April two-page “tentative chapter outline” by Miller for The American Voter in Mid-Century. Its major sections were on “political motives and voting behavior,” “origins and contexts of motivated behavior: institutional and individual factors in voting,” and “the election.” As an example of how it fleshed out the topics listed in the original outline: “parties” became separate chapters on party identification and party images, both under the political motives section of the book. Similarly the “groups” topic was divided into group memberships, reference groups, social class, non-economic demographic categories, geographic mobility, and primary groups.

Campbell responded on May 22 with a four-page outline. The title for this outline reflects Campbell’s psychology background: The Psychology of Political Man. Its two major sections were on “forces on the voter” and “the political act.” “Parties” became one topic in the forces section, with subcategories for party images and party identification, and with specific points under each, such as “the party image differs among voters in cognitive and affective structure,” and “most people have some degree of attachment to a party and its symbols.”

That version was soon followed on May 29 by an extensive 19-page outline with a title that tried to combine the two previous titles: The Psychology of Political Man: The American Voter and Presidential Politics. Its three major sections went back to Miller’s April outline, but with considerably more detail. For example, the party images category included points on the major features of those images (the party as agent of political leadership, as manager of government, and as agent espousing particular policies), that Democratic and Republican voters have similar cognitive images of the two parties but differ in the affective or evaluative dimension (with a party’s adherents minimizing unfavorable aspects of its agents, but each party being perceived as having weaknesses in one area of government policy), and that major features of each party’s image are stable across elections but with changes that affect the party’s electoral success (such as hostile evaluations of the Republican Party’s domestic issue positions being replaced by more balanced in 1956 due to the Eisenhower administration).

While it is not clear whether any of these early versions was their working outline for writing the book, it is still interesting to see how quickly their ideas developed over those five weeks. The 19-page outline makes evident that they had accumulated a large amount of specific findings by late May 1957, certainly enough to convince Campbell that they could accomplish the task of writing a major book.

Title and schedule

In June 1957, they sent book proposals to several publishers. ^{Footnote 23} While these proposals have not been located, the responses show that the book’s title was still in flux with publishers referring to the book as The American Voter at Mid-Century, The Psychology of the Vote, and Psychology of the Elective Process. ^{Footnote 24} The authors chose to go with John Wiley and Sons, which was a leading publisher of academic research, and in September 1958 signed a contract with them for “a work now entitled The American Voter.” ^{Footnote 25}

The authors were apparently writing vigorously throughout the 1957–58 academic year. They were all in Ann Arbor that year, so there is no correspondence documenting their progress until Campbell went to Norway for the 1958–59 academic year. There were also letters in summer 1959 when Miller did summer teaching at the University of Illinois, and then when Converse went to France on a Fulbright for 1959–60.

In December 1958, Wiley specified the schedule for the book. The three authors in Ann Arbor were to complete their part of the project by the end of the year. The manuscript was then to be sent to Campbell in Norway where he was to devote the spring to a final revision, to be delivered to the publisher by June 1959. ^{Footnote 26} That timetable proved to be unrealistic when the authors in Ann Arbor were not able to finish all of their chapters in 1958 and when Campbell apparently made little progress on the manuscript during his year in Europe. The new plan was for the authors to have the material ready for Campbell on his return to Ann Arbor in summer 1959, with the manuscript to be submitted to Wiley that August. ^{Footnote 27}

The book’s title was still at issue through the second half of 1959, with the main contenders evidently being The American Electorate and The American Voter. A letter from Converse in August stated that Stokes’ favored title was The American Electorate: A Behavioral Study of Presidential Voting (Politics?), while Miller in November referred to the book as The American Electorate. An early December letter from Miller to Converse said that Stokes and Miller were “strongly switching back in favor of The American Voter,” though Campbell seems to be “still lean[ing] toward The American Electorate.” The last mention was in a mid-December letter from Converse to Miller, “I too think it [The American Voter] is otherwise preferable to The American Electorate, although that always seemed next best.” ^{Footnote 28} Apparently the title was settled about then. Fittingly, Angus Campbell’s son Bruce later titled his book on voting behavior The American Electorate (B. Campbell, Reference Campbell1979).

Authorship

There is some information about chapter authorship. Phil Converse gave Bill Jacoby a list of who had responsibility for each chapter (see online supplementary materials), though it is unclear whether this was their original assignments, what they actually wrote, or a later reconstruction.

The available information is that Angus Campbell wrote chapter 1 on the “setting” of the election. Chapter 2 on “theoretical orientations” was based on Converse’s seminar paper. Stokes was listed as the author of chapters 3–5 on the “perception of the parties and candidates,” “partisan choice,” and “voting turnout,” respectively. Campbell is shown as the author of chapter 6 on “the impact of party identification,” which fits with his interest in the party identification question though the chapter also relies heavily on the vote components that Stokes is associated with. The list that Converse gave Jacoby has Converse as the author of chapter 7 on “the development of party identification,” but the version that Converse (Reference Converse1991) relied on in his oral history interview showed that Campbell wrote that chapter. Miller is listed as the author of chapter 8 on “public policy and political preference,” which was based on his interest in the conditions for issue voting dating back to his dissertation. Converse (Reference Converse1991) indicated that his chapter 9 on “attitude structure and the problem of ideology” and his chapter 10 on levels of conceptualization (“the formation of issue concepts and partisan change”) were not present in the original book proposal but were added based on his thinking after he read through the answers to the 1956 open-ended questions.

Miller wrote chapter 11 on “election laws and political environment.” Converse authored chapter 12 on “membership in social groupings,” which included the analysis in his dissertation of the distinctiveness of the voting of social groups controlling for their other social characteristics. Based on his early book chapter, Converse also authored chapter 13 on “the role of social class,” which required his writing a paragraph explaining the correlation coefficients that he was using. Miller may have written chapter 14 on “economic antecedents of political behavior,” though there is a question mark after his initials on the chapter list. Converse is listed as the author of chapter 15 on “agrarian political behavior,” Miller for chapter 16 on “population movement,” and Converse for chapters 17 on “the electoral effects of other social characteristics” and 18 on “personality factors in voting behavior.” Campbell is listed as writing chapter 19 on “the electoral decision” and the concluding chapter 20 on “electoral behavior and American politics,” though the multivariate analysis that underlies chapter 19 is from Stokes’ dissertation and the discussion of Downs’ model in chapter 20 may also be due to Stokes.

When the book was nearing completion, Campbell was SRC director and was spending the 1958–59 year in Norway, and Miller had been promoted to associate professor in 1958 in response to an offer from Berkeley. Converse and Stokes were both getting feelers from other schools by early 1959. ^{Footnote 29} Converse, about to go to France on a Fulbright in 1959–60, was appointed as an assistant professor in Michigan’s sociology department as of 1960. Pollock seems to have gained a very favorable impression of Stokes, ^{Footnote 30} so Stokes then received an assistant professor appointment in political science at Michigan starting that fall regardless of Pollock’s comments when they first met. To give an idea of their relative seniority, Campbell was approaching age 50, Miller about 35, and Converse and Stokes in their early 30s.

The book is known as “Campbell, Converse, Miller, and Stokes,” but author orders always entail an explicit decision. While it might have seemed most natural to follow the institutional order with Campbell as lead author, Miller could have been first author given his major role in shaping the vision of the book project or, more likely, Converse based on how many chapters he authored. Correspondence with Miller during the summer he taught at Illinois makes clear that this was a matter of some agitation among the junior authors, who eventually suggested an alphabetical ordering due to recognition that any other ordering would have erroneously implied that the last named authors had made lesser contributions to the effort. ^{Footnote 31} In an aside in a long letter to Converse in December 1959, Miller announced the final decision on author ordering in the shortest sentence in all their correspondence: “It will be alphabetical.” ^{Footnote 32} In his oral interview with Miller (Reference Miller1988), Heinz Eulau suggested that Campbell expected to be first author because he was the principal investigator; the other authors were, in Miller’s words, “junior staff members.” However, the junior authors could consider the alphabetical ordering as a “great leveller,” ^{Footnote 33} while still allowing Campbell to view the book as “capping” off his contributions to the field. By the time of his oral history interview, Miller (Reference Miller1988) described the book as a “corporate product” since each chapter was gone over collectively 6-8 times.

Finalizing the book

In any case, the chapters dealt with most of the issues that were raised by reviewers for the previous SRC election reports. Significance tests are not shown, but they may have guided the choice of which results to display. Control table logic was improved, ^{Footnote 34} though the only regression analyses are for the group distinctiveness analysis and the vote components model. Stories are that Stokes did a very large number of additional regression runs, but the authors apparently decided that the book would be more readable if they downplayed their statistical presentations.

As the Ann Arbor authors were racing to finish their chapters, the data from the SRC 1958 election study became available. Converse’s (Reference Converse1991) oral history indicates that Stokes felt it would be “scientifically irresponsible” to go ahead with the book without testing it with the 1958 data since replication data was in hand. In the end, they agreed to “peek” at one table for verification purposes, looking at the party identification turnover from 1956 to 1958 in order to check their claims about the stability of people’s partisanship even though they did not include that table in the book. Its stability led to the decision that they could go ahead, without incorporating the 1958 study into the book. Letters between Miller and Campbell make clear that Campbell’s “surge and decline” analysis was originally to be part of the book, but it was dropped as they pulled 1958 material out of the book. ^{Footnote 35}

Miller in March 1959 wrote that they were dropping chapters 24 and 25, so chapter 26 “has a more terminal quality than I expected.” ^{Footnote 36} This is the only available indication that the book was originally planned to have as many as 26 chapters rather than the 20 that appeared in the published book. The topics of those dropped chapters are not evident, except for a political communications chapter that Stokes was to have written. ^{Footnote 37}

Converse was in charge of editing the final manuscript. He (Converse Reference Converse1999) remembered his rewriting Miller’s chapters for style (but not content), but Miller described that editing more broadly in a May 1959 letter. Miller wrote of “the tremendous job which Phil is now doing in his role as editor. His sense of the total scope of the volume is leading to a number of major and I think exceedingly useful changes within particular chapters. He is also giving sufficient attention to detail to ensure congruence of treatments for specifics throughout the book.” ^{Footnote 38}

However another problem emerged after they sent the manuscript to Wiley: the book was too long. Campbell returned from an early August 1959 meeting with the publisher in New York with the news that they had to cut the manuscript back. ^{Footnote 39} Various letters indicate cuts of 100 to 150 pages, presumably 100 book pages and 150 manuscript pages. ^{Footnote 40} This may have been the stage when the book was reduced to 20 chapters. Harvard University political science professor V.O. Key, Jr., who was spending the year at ISR analyzing the election studies for his own public opinion book, aided in cutting The American Voter draft from its original 828 manuscript pages, and “Mrs. Key” [known as Luella] did some copyediting. ^{Footnote 41} The cuts included removing an explanation of data analysis from the preface and deleting appendix material such as the questionnaires. Miller then read the manuscript and made further cuts, Campbell dropped a few more paragraphs while reinserting some that had been removed, Stokes examined the changes made in his chapters, and Miller made a final read-through before sending the final manuscript back to Wiley.

Possibly because of the mixed reactions to the earlier SRC books, there was some uncertainty about the final manuscript. Converse told the story of walking with Stokes after the book’s final draft was sent to Wiley (Weisberg, Macdonald, and Rabinowitz Reference Weisberg, Macdonald and Rabinowitz1998, 276), and Stokes out-of-the-blue asking “Phil, we finished the book, but is it the right book?” Converse (Reference Converse1991) asserted that Stokes was so unsure that he was even considering pulling his name from the book. Fortunately, it was the right book.

Reviews

The reviews were enthusiastic. V. O. Key, Jr. (1960) in a special review essay for Public Opinion Quarterly termed it a “monumental performance.” Brewster Smith (Reference Smith1961) called it “an interdisciplinary classic.” The Science review (Odegard Reference Odegard1960) praised its “restoring politics to the study of voting behavior.” The APSR review (Eulau Reference Eulau1960) emphasized how the authors responded to criticisms of the earlier SRC work. Murray Edelman (Reference Edelman1961) said that it “is likely to become notable chiefly for its imaginative theory, its stimulating speculations, and its suggestions for fitting the voting act into a model of the total political process.”

Criticisms of the book by reviewers were minimal. While lauding the book, Brewster Smith (Reference Smith1961) appropriately called attention to how the personality chapter took too narrow a conception of personality. Oscar Glantz (Reference Glantz1961) essentially complained that the book was elitist in its analysis of levels of conceptualization because it separated “near-ideologues” from “ideologues,” which meant that Glantz was not recognizing the difficulty that these near-ideologues would have in understanding media discussions of liberal versus conservative politics.

The most controversial element of the book was the “funnel of causality.” Most of the reviews touched on it, but with widely varying assessments. V. O. Key, Jr. (Reference Key1960) said that “the imposing mass of the analysis is also held together by a broad theoretical conception—‘the funnel of causality.’” Brewster Smith (Reference Smith1961) referred to the funnel as “useful scaffolding,” though Glantz (Reference Glantz1961) called it “a substitute for useful theory.” Eulau (Reference Eulau1960) recognized it is “a convenient device to establish a hierarchy of influences, some close to and others more remote from the voting act,” but he complained that the funnel became “an obstacle to understanding the ways in which conceptually different levels of analysis and empirically different phenomena can be related to account for the complexity of human behavior.”

The authors

The book proved to be much more important than anyone imagined, and all four authors were recognized as major scholars. Campbell was surprised by the great success of The American Voter; he expressed admiration for his bright young colleagues. ^{Footnote 42} The author order did not really matter in the end. By following an alphabetical order, there was no implication that the junior staff members contributed any less than Campbell. The post-American Voter publications about voting behavior by Converse, Miller, and Stokes verified the excellence of their research abilities, and the institutions created around the voting studies enabled colleagues from around the world to meet all of them and recognize their eminence.

The junior authors benefited from the improvement in relations between the SRC and the Michigan political science department after Pollock stepped down as department chair in 1961. Arthur Bromage served as a transitional chair from 1961–64, which facilitated the promotions of Miller, Converse, and Stokes in 1963. Miller became a full professor; Converse received a joint appointment as associate professor in political science and sociology; Stokes moved up to associate professor rank. The ascension of Samuel Eldersveld to be department chair in 1964 led to an era of active cooperation between the two units. Converse and Stokes became full professors as of autumn 1965, just seven years after receiving their PhDs. Eldersveld recognized the importance of cooperating with Miller to modernize the department, working together to recruit top young political science behavioral researchers to join the faculty with joint appointments at ISR and reestablishing the department as one of the premier political science departments in the nation.

Later collaborations

While there were numerous subsequent articles by the four authors, both individually and in various combinations, the foursome joined together on only two pieces after The American Voter. ^{Footnote 43} A 1961 APSR article on voting in the 1960 election had Converse as the lead author. Their Elections and the Political Order book (Campbell, Converse, Miller, and Stokes Reference Campbell, Converse, Miller and Stokes1966) went back to an alphabetical order, but with the authorship of each chapter separately identified.

Elections and the Political Order can be seen as filling in the unfinished business of The American Voter. Whereas their first book provided important general perspectives, the second developed precise original theories and applications of those theories. This theory emphasis was exemplified by the only two previously unpublished chapters: Converse’s normal vote chapter, which used the size and past voting of each party identification category to estimate how an election would turn out if short-term forces were balanced between the parties, and his application of the normal vote to analyzing the role of religion in the 1960 election. Two other Converse theory/application chapters were his influential “information flow” theory, and his chapter on party realignment in the South. Campbell’s important “surge and decline” theory was included. Stokes contributed three theoretical/methodological articles: on deviating elections, on forces restoring party competition, and on valence voting in his critique of spatial models of party competition. The two Miller and Stokes representation study articles were included, along with two articles that Converse wrote with Georges Dupeux on French politics, and an article Campbell wrote with Henry Valen on party identification in Norway versus the United States.

Institution building

The election studies led to the development of a series of important new institutions. First, Michigan’s Political Behavior graduate program was established in 1960 under the joint directorship of Miller and Eldersveld (Miller Reference Miller1960). Next, Miller led the founding of the Inter-University Consortium for Political and Social Research (ICPSR), so that professional staff could send the survey data to member schools so scholars elsewhere could perform their own analysis. Miller (Reference Miller and Eulau1989, 160) stated that he was most proud of the ICPSR’s role “in creating new standards for professionalism in our research endeavors,” as it created the norm for scholars to share their data (cf. Hauptmann Reference Hauptmann2016b). The desire of junior faculty and graduate students across the country to learn how to analyze these data led to the creation of the Consortium summer program in 1963, with courses in research design and data analysis (Weisberg Reference Weisberg1987). Later, when the political methodology group was forming, Miller invited the methodologists to Ann Arbor where he hosted their first conference in the summer of 1984 (Achen Reference Achen2000). Additionally, the Michigan efforts helped foster the creation of social science data archives and training programs internationally.

With the Michigan researchers conducting collaborative research with scholars around the world and with the Ford Foundation’s interest in bringing training in modern research techniques and behavioral approaches to Latin America, the decision was made to expand the crossnational efforts, spinning off the political behavior program into a new wider-ranging Center for Political Studies (CPS) within ISR. As Michigan grew as a research institution, CPS became a leading center for the quantitative study of politics, broadly interdisciplinary, international, and theoretical. While the SRC winning the 1952 election study grant may have seemed at the time as the political science department losing a turf war, its subsequent cooperation with a vibrant CPS helped the department develop strength across the discipline, becoming one of the very top-ranked departments nationally.

As private foundation funding became more difficult to obtain, Miller, who had taken over responsibility for seeking foundation funding for the election studies, turned to the National Science Foundation. After the NSF helped fund three studies, he worked with David Leege, NSF’s political science program officer at the time, to move to more permanent funding. The American National Election Studies (ANES) was created as a shared “national resource” between universities with NSF funding starting in 1977. Instead of the Michigan principal investigators choosing the questions to ask and having first access to the data, scholars could suggest questions to a Board of Overseers composed of scholars from around the country that would choose the questions, with the data released simultaneously to all scholars. The most recent innovation is the proposal of questions through an Online Commons, which allows for feedback to improve question proposals.

Numerous national election studies around the world were modeled after the ANES. The Comparative Study of Electoral Systems is another outgrowth, directly coming from the ideas proposed at a conference by Steve Rosenstone, who was then director of ANES; election surveys in more than sixty countries now include its common module of questions in their national post-election questionnaires. The ANES also served as a model for both the General Social Survey and the Panel Study of Income Dynamics, the other two long-running large-scale longitudinal national social science surveys funded by the National Science Foundation.

Later work

While there was not another four-authored piece after the mid-1960s, the authors continued their research on voting. After The Changing American Voter (Nie, Verba, and Petrocik Reference Nie, Verba and Petrocik1979) challenged the applicability of The American Voter results to the elections of the tumultuous 1960s, Converse, writing with Greg Markus, rebutted, using the 1972–74–76 panel study, with an article entitled La plus ça change that found that the relative importance of vote determinants had not changed. Miller coauthored several books, articles on the elections through 1988, and, importantly, The New American Voter’s (Miller and Shanks Reference Miller and Merrill Shanks1996) multi-stage analysis of voting in the 1992 election. Miller’s 1990s articles included a vigorous argument for the continued importance of party identification and a reexamination of the decline in voting turnout; he also coauthored articles on state political polls and campaign finance with his wife, Ruth Jones (Jones and Miller Reference Jones and Miller1984; Reference Jones and Miller1985). Stokes returned to his interest in electoral politics with important work in the 1990s (e.g., Stokes and Dilulio Reference Stokes, Dilulio and Nelson1993), arguing that electoral politics in America had become more volatile as politics shifted from Downsian position issues to valence issues, which everyone favors (such as peace) or everyone opposes (such as recessions). His work on valence politics has proved to be one of his most important contributions to the subsequent voting behavior literature. Their studies of voting were not limited to the United States, with the Butler and Stokes (Reference Butler and Stokes1969) book on Political Change in Britain and the Converse and Pierce (Reference Converse and Pierce1986) book on Political Representation in France both receiving the American Political Science Association’s prestigious Woodrow Wilson Prize.

Additionally, Converse’s writings on attitude formation and change have had a major impact on the study of public opinion. His paper on “Information Flow and the Stability of Partisan Attitudes” discovered a curvilinear relationship between media intake and party defection, with the lowest defection rates among those with moderate levels of media information about politics. His masterpiece, a book chapter on “The Nature of Belief Systems in Mass Publics” (Converse Reference Converse and Apter1964), emphasized the limitations that citizens have in forming consistent and constrained attitudes, showing that most citizens do not think about politics through ideological structures. Instead, there are “issue publics,” composed of citizens who have deep concern about particular issues. Many opinions obtained in public opinion polling are really “non-attitudes,” changing randomly when asked the same question over time; respondents without opinions would just say something to satisfy interviewers so they would go on to the next question. Converse’s appropriate recognition of the limitations of survey research was due to his experience as an interviewer as well as from the work of his wife, Jean Converse, on the survey experience as seen by interviewers (J. Converse and Schuman Reference Converse and Schuman1974). ^{Footnote 44}

The Michigan authors were also innovative in statistics. Converse’s “The Problem of Party Distances in Models of Voting Change” was not only the first study to apply a multidimensional model to partisan competition, but it anticipated individual differences scaling when he showed how different sets of voters could vary in the emphases they gave to the different dimensions. Stokes applied his more advanced mathematics and statistics knowledge to the modeling and analysis of voting. His 1960s publications illustrate his enchantment with statistical and formal models, including his use of a random walk model to test the equilibrium properties of congressional elections, his argument against oversimplification in spatial models of voting, his depiction of cross-level inference as a “game against nature,” and his developing a variance components model to disentangle the extent of nationalization of American elections. Miller was always a skilled analyst of survey data, so it is fitting that his final article on “Temporal Order and Causal Inference” was published in Political Analysis (Miller Reference Miller2000).

Michigan became known for innovative survey designs. Their multi-year panel studies allowed for exploring the stability of attitudes over time. Their 1984 continuous monitoring study was the first academic study interviewing across the election year. Several of their studies involved split-form experiments and they conducted multi-mode surveys in 2000 (with a half-sample contacted by telephone instead of in-person) and 2012 (with a separate Internet sample). The most innovative design was the 1958 Miller and Stokes representation study, which included interviews with members of Congress from the districts in their sample along with the 1958 candidates in those districts. Their 1963 APSR article on “Constituency Influence in Congress,” based on that study, was particularly influential. Their planned representation study book never appeared; the authors were overcommitted (Miller Reference Miller, Baer, Jewell and Sigelman1991, 240) and could not settle on the analysis chapters, perhaps because Stokes recognized there were too many technical issues involved in proper estimation of the models. ^{Footnote 45} Their continued interest in this project was exemplified by a book outline that Stokes prepared in 1969 and Miller’s revision of six draft chapters in 1995. ^{Footnote 46}

The authors’ later work also made important contributions to social science more generally. Campbell joined with Howard Schuman on a study of racial attitudes in American cities and then wrote a research monograph on white attitudes toward blacks. Campbell and Converse’s book on the Human Meaning of Social Change and their book with Rodgers on well-being demonstrate their breadth of interest. Converse wrote books on attitudes on the Vietnam War with Howard Schuman and with Milton Rosenberg and Sidney Verba. Stokes’ final book, published posthumously, was on science policy; Pasteur’s Quadrant went beyond the usual distinction between pure and applied research to argue for redirecting government science funding towards use-inspired basic research.