The Spiral of Silence in the 1948 Presidential Election

COMMUNICATION McDonald et al. The RESEARCH Spiral of Silence April in 19482001 DANIEL G. MCDONALD CARROLL J. GLYNN SEI-HILL KIM RONALD E. OSTMAN The Spiral of Silence in the 1948 Presidential Election Noelle-Neumann s spiral of silence is tested with data from the 1948 Elmira election study. Results indicate that support for the theory and trends in the data are consistent with expectations for the spiral of silence as an additional explanation for the famous election misprediction. Statistical tests affirm Noelle-Neumann s suggestion that social isolation is associated with fear of isolation and provide partial support for the idea that social isolation interacts with fear of isolation in the spiral of silence effect. Bridging individual-level and macrolevel processes, the spiral of silence is one of the most highly developed and one of the most researched theories in the field of public opinion (e.g., Glynn & McLeod, 1984; Glynn & Park, 1997; Lasorsa, 1991; Noelle-Neumann, 1974, 1984, 1985; Salmon & Neuwirth, 1990; Salmon & Oshagan, 1990; Willnat, 1996). The theory focuses on the idea that individuals fear social isolation and monitor their social environments for evidence of the extent to which their opinions match the trends of dominant opinion. Because of the fear of isolation, those who perceive their opinions to be dominant or gaining ground will be more likely to express their opinions in The data used in this article were made available by the Inter-University Consortium for Political and Social Research. The data for the Elmira Community Study, 1948, were originally collected by Paul F. Lazarsfeld, Bernard R. Berelson, and William N. McPhee. Neither the original collectors of the data nor the consortium bear any responsibility for the analyses or interpretations presented here. All correspondence regarding this manuscript should be addressed to Professor McDonald: mcdonald.221@ osu.edu or 3080 Derby Hall, 154 N. Oval Mall, The Ohio State University, Columbus, Ohio, 43210-1339. Telephone: (614) 292-5811; Fax: (614) 292-7055. COMMUNICATION RESEARCH, Vol. 28 No. 2, April 2001 139-155 2001 Sage Publications, Inc. 139

COMMUNICATION RESEARCH April 2001 public than will those who perceive their opinions to be in the minority or losing ground. The dynamic that results from these tendencies is a spiral of silence in which those who perceive themselves to have minority views become increasingly silent as those who perceive themselves in the majority become increasingly vocal (Noelle-Neumann, 1974). The process continues until only a small group of those with minority views (the hardcore ) continues to speak out without regard to the climate of opinion. After a time, in the face of continued hardcore support, the trend of opinion expression becomes static. At this point, two things happen. First, because the downward trend has stopped, those in the perceived majority begin to feel slightly less sure of themselves, whereas those who had remained silent while their opinion was losing support feel more comfortable with their opinions. Second, an avantgarde begins to speak out in favor of the minority position, and a trend of opinion expression in the opposite direction begins. Others who hold the minority position begin to speak out as they perceive their opinion is gaining ground. Those in the majority notice the trend and begin to silence themselves for fear of social isolation. By far the majority of spiral of silence research has examined the relationship between willingness to speak out and the climate of opinion (Glynn & McLeod, 1984; Lasorsa, 1991; Noelle-Neumann, 1984; Oshagan, 1997; Salmon & Neuwirth, 1990; Salmon & Oshagan, 1990; Willnat, 1996). Glynn and McLeod (1984) found that perceived opinion climate was related to willingness to speak out, and found some support for the idea that those who see a candidate as gaining support would be more likely to express a preference for that candidate. Glynn and Park (1997), Oshagan (1997), Salmon and his colleagues (Salmon & Neuwirth, 1990; Salmon & Oshagan, 1990), and others have found additional support for this basic relationship. Although there have thus been a number of empirical tests of support for a relationship between perceptions of who is winning and the expression of preference for a candidate, only a very few authors have investigated the idea of the hardcore (Glynn & McLeod, 1984; Willnat, 1996). The hardcore appear to be those with a high interest in the issue and who remain unchanging in their opinions. Both Glynn and McLeod (1984) and Willnat (1996) found the hardcore less susceptible to the spiral of silence. Although the hardcore concept is difficult to operationalize, the lack of research incorporating the concept into tests of the spiral of silence is surprising. If the theory is correct, inclusion of members of the hardcore in analyses of the spiral of silence will lower the strength of any relationship that might be found between opinion expression and perception of the climate of opinion. 140

McDonald et al. The Spiral of Silence in 1948 Reference Groups, Social Norms, and the Spiral of Silence Reference groups and social norms have been an issue since the first empirical tests of the theory were published (Glynn & McLeod, 1984), and they continue to be an issue (Glynn & Park, 1997; Salmon & Oshagan, 1990). The key issue is whether individuals will weigh their friends and family members opinions more heavily than they weigh those of the general public (Glynn & Park, 1997; Krassa, 1988; Oshagan, 1997). Noelle-Neumann (1984) has elaborated on the idea of social groups and the spiral of silence. According to Noelle-Neumann, certain societal roles preclude full participation by certain members of the population, and so the members of those groups are more fearful of social isolation than those in more powerful positions in society. Noelle-Neumann proposes that those who are more socially isolated will be more susceptible to the climate of opinion. She suggests that because some people are already somewhat isolated from others, with few friends to talk to, they will not want to risk further isolation (Noelle-Neumann, 1973, 1974, 1984). We investigate these ideas with three hypotheses: Hypothesis 1: Non-hardcore individuals (i.e., those who are changeable) who believe there is a trend in support of their position are more likely to express their opinions publicly than are those who do not. Hypothesis 2: The degree of social isolation will interact with perception of the climate of opinion in predicting opinion expression. Hypothesis 3: There will be a significant improvement in prediction when hardcore respondents are removed from the analysis. The present study investigates these hypotheses by conducting a secondary analysis of the 1948 Elmira data set (Berelson, Lazarsfeld, & McPhee, 1954). In 1948, Harry S. Truman was the incumbent Democratic president, Thomas E. Dewey was the Republican challenger. The Democratic party had splintered into Democrats and Progressives, led by Henry A. Wallace. During the convention in August, as it became clear that Truman supported equal civil rights for all people, Strom Thurmond stormed out of the convention, taking much of the South with him. The 1948 campaign and election are most remembered for two interdependent events. First, the scientific public opinion polls, thought to be a great advance over the straw ballot techniques that had disgraced the Literary Digest in 1936, predicted an easy win for Governor Dewey. Virtually no one 141

COMMUNICATION RESEARCH April 2001 but Truman thought Truman could win (Clifford, 1991). These predictions were widely reported by newspapers and radio, which picked up the information from press releases distributed by the polling companies (Albig, 1956; Mosteller, Hyman, McCarthy, Marks, & Truman, 1949). The second event resulted in a famous photograph, showing a beaming and jubilant Harry S. Truman holding aloft a Chicago Daily Tribune front page with the mistaken headline Dewey Defeats Truman. Truman had in fact defeated Dewey, with 49.6% of the popular vote to Dewey s 45.1%. Some 51.2% of the electorate had cast a ballot (Ross, 1968). Thurmond, with 2.4% of the vote, had taken 39 (7.3%) of the electoral votes; Wallace, with nearly an identical proportion of the popular vote, had received no electoral votes. Truman had received 303 electoral votes, Dewey 189. Although this was a large electoral victory, in some respects the election was quite close. A change of less than 1% of the vote in three states Ohio, Illinois, and California would have meant a change of 78 electoral votes and a very different election result (in Ohio, Truman won by.3% of the vote; in California by.5%; and in Illinois by.9%). The misprediction generated considerable discussion among academic researchers, pollsters, and the media. Colloquia were held about what had gone wrong, and a number of ideas were advanced suggesting causes for the error (Brennan, 1949; Campbell, 1950; England, 1949; Gallup, 1949; Gosnell, 1952; Guest, 1949; Hyman, 1953; Jacquemyns, 1950; Katz, 1949). The Social Science Research Council (SSRC) convened a special committee headed by Frederick Mosteller to conduct an investigation of the 1948 election polls under the direction of the SSRC (Mosteller et al., 1949). The SSRC committee reported that the failure of public opinion polls to predict the outcome created widespread confusion and misgivings about the reliability of polls (Mosteller et al., 1949). The SSRC report included seven major conclusions about why the prediction error occurred. These can be grouped in terms of two types of errors: errors of procedure (e.g., the election was very close and pollsters failed to take account of potential sampling/interviewing and estimation error; the media failed to report margins of error or provide information on accuracy) and errors in understanding the phenomenon of mass political behavior (e.g., last-minute shifts; nonproportional voting by undecided respondents, etc.). The SSRC report noted that the former group of errors was easily fixed; techniques were already developed to take account of sampling error and other accuracy issues. However, the committee suggested that the latter group was only correctable through basic research on political behavior and the psychology of elections because the second group of errors in this forecast was primarily applicable to situations involving expression of opinion or intention to act. 142

McDonald et al. The Spiral of Silence in 1948 The present study offers a somewhat different perspective by examining one possible explanation of psychological functioning during an election. Rather than addressing how Truman won by comparing various social groupings with their behavior in prior elections or last-minute shifts, we test Noelle-Neumann s (1974) theory of the spiral of silence as one potential social-psychological explanation for the late shift in the electorate. If the spiral of silence was operating in 1948, Truman supporters would have become less likely to express their preferences during the conduct of the campaign, until the point at which only the hardcore Truman supporters were providing their opinions. Eventually, the avant-garde would begin to speak out, and there should be an increase in the perception that Truman was gaining ground, resulting in a rebound of Truman support, prompting more of those who had previously been quiet to express their opinions. If something such as the spiral of silence did occur in 1948, it might offer an additional explanation for the failure of pollsters to account for the trend toward Truman. Study Background and Method To assess how well the spiral of silence might explain some of the problems in the 1948 election, the present study takes advantage of data collected in 1948 in Elmira, New York (Berelson et al., 1954). The Berelson et al. (1954) study, although designed long before Noelle-Neumann s (1974) theory was published, included most of the components for testing the theory in a longitudinal design that has seldom been replicated in the social sciences. The method of Berelson et al. s 1948 study is nearly as famous as its findings. Using a technique pioneered in their study of voters in Erie County, Ohio (Lazarsfeld, Berelson, & Gaudet, 1948), the researchers constructed an area probability sample and interviewed respondents face-to-face on four separate occasions during the campaign season. The first interviews were conducted in June. Respondents were reinterviewed in August, October, and briefly after the election in November. Of the original 1,029 respondents in the June panel, 746 respondents were interviewed successfully in all four waves. Each wave was unique and elicited data on new topics, but every wave asked respondents for certain questions, including perception of who would win the election and the respondent s vote choice. If there was any change in response from one wave to the next, interviewers asked follow-up questions. They did this whether the respondent was stating his or her intention for the first time, changing from one candidate to another, or rejecting a candidate to become uncertain or stay home from the polls. Further methodological details are available in Berelson et al. (1954). 143

COMMUNICATION RESEARCH April 2001 Measures Because this is a secondary analysis of the data, a number of conceptualization issues arise in measuring the concepts of the theory. For our analyses, we return to the earliest methods Noelle-Neumann used for measuring the climate of opinion and opinion expression (Noelle-Neumann, 1973, 1974, 1984). Full questionnaires are reproduced in Berelson et al. (1954). We employ these measures for three distinct reasons: (a) They tie in more closely with the historical event of the 1948 election; (b) they are more clearly linked to the theory than are scenario measures that tap perception of a behavior rather than actual behavior; and (c) they offer a test of the theory in a U.S. election setting, using measures very similar to Noelle-Neumann s own operationalizations. The exact measures are as follows: OPINION EXPRESSION We also follow Noelle-Neumann s lead in using candidate preference for our dependent variable of opinion expression. Noelle-Neumann s (1974, 1984) original analyses were based on the expression of candidate preference during an election. Further tests of the theory, often her own, have usually relied on a scenario asking people for their perception of whether they would speak up in a public situation essentially moving one step further from the process described in the theory. The operationalization employed here departs from many recent studies of the spiral of silence, which tend to use scenario measures rather than actual perceptions or opinion expressions. We were aware of only one other recent study (Glynn & Park, 1997) that has used actual opinion expression, even though it is the measure employed in Noelle-Neumann s earliest studies. However, the present study seeks to take the analysis closer to the original theory and the historic election poll results by using actual opinion expression. There were four points at which a person expressed candidate preference: Before the conventions, they asked if the candidates were Truman, Dewey, and Wallace, which candidate would the respondent favor (Wave 1); after the conventions, they asked which candidates the respondent favored (Waves 2 and 3); and the report of actual vote (Wave 4). We use the first wave as a baseline preference measure (see the climate of opinion measure below), and use preference expressed during the past three waves as dependent variables. In addition, as a separate measure, we sum the number of times a person expressed candidate preference during the final three waves (0 to 3), and use this as a final summary measure of opinion expression. 144

McDonald et al. The Spiral of Silence in 1948 CLIMATE OF OPINION Noelle-Neumann (1984) notes that perception of who will win the election has been employed as a measure of the climate of opinion for decades. On each interviewing occasion, Berelson et al. (1954) asked perception of who would win the election in November. In the earliest wave (before the conventions), they asked several questions about preference for various potential candidates and which party would win; in later waves, they asked which candidate would win. We assessed perception of a climate trend at two time points. We first used Wave 1 candidate preference ( If the candidates were Dewey, Truman, and Wallace, who would you vote for? ) as a baseline. For the first three waves, we calculated climate trend in a simple dummy coding in which 1 indicated that the candidate preferred in Wave 1 was perceived as the one who would win the election. This simply reports whether, in Wave 2 or 3, a respondent who expressed a preference for a candidate in Wave 1 saw that candidate as winning the election in later waves. In assessing the report of actual vote as an opinion expression in Wave 4, we compared Wave 1 preference to reported vote, with a 1 coded for those who preferred Truman in Wave 1, and 0 for those who preferred Dewey (because Truman actually won the election, only those who preferred him can perceive a positive climate of opinion after the election). HARDCORE Although Noelle-Neumann s theory is not very explicit about how we might identify the hardcore, our operationalization was based on two aspects of the hardcore that she has written about: They do not change their opinion to flow with the perceived trend, and they are not afraid to speak out even if they perceive themselves to be on the losing side. We take both of these into account in our measure. The 1948 study asked people to compare their political discussion with others, using a 4-point scale. Our operationalization of the hardcore consists of those who were consistent in their party/candidate preference during the first three waves of data and also reported that they either participated about equally or tried to convince others of their positions when they were engaged in political discussions. By default, all those who did not fit these two criteria were not hardcore. Although Noelle-Neumann does not often refer to the remaining respondents (the non-hardcore majority) by any particular name, she has occasionally referred to them as changeable (Noelle-Neumann, 1984, 1991), and we will use that term here. 145

COMMUNICATION RESEARCH April 2001 SOCIAL ISOLATION Noelle-Neumann (1974, 1977, 1984) suggests that the greatest spiral of silence effects are found among those who are already isolated from society. For this analysis, we measured social isolation through the combination of two questions. The first asked if the respondent belonged to any club or organization. The second asked respondents who they would ask if they had a question about politics. This last could be any family member, coworker, friend, or anyone else. We considered people to be socially isolated if they did not belong to any club or organization and could not name any person they would be able to ask if they had a question about politics. In this way, the measure taps the extent to which people were connected to others, formally (through groups or organizations) and informally (in simply being able to ask someone a question about politics). As measured here, socially isolated individuals are thus isolated in both formal and informal relationships. DEMOGRAPHICS For demographic variables, we included self-report measures of age, educational level, and perceived social class when growing up; all of which have been used by Noelle-Neumann (1984) in her analyses. THE INTERACTION OF SOCIAL ISOLATION AND OPINION CLIMATE Because the Noelle-Neumann (1974) hypothesis dealing with social isolation is essentially an interaction, we calculated a multiplicative interaction term combining the climate of opinion with social isolation. The Data Set Data were obtained by the first author from the Inter-University Consortium for Political and Social Research (ICPSR) (ICPSR data set #7203) in electronic multipunch column binary format, read and analyzed using Windows 98 and the Statistical Package for the Social Sciences, versions 7.5 and 8.0, using SPSS-X (mainframe) input protocol for column binary data. Results Because much of the analysis requires the omission of the hardcore, we provide Table 1 as an overview of other characteristics of the hardcore in comparison to the changeable. Interest in the election was measured three times 146

McDonald et al. The Spiral of Silence in 1948 Table 1 Differences Between the Hardcore and Other Respondents Variable n M SD T a p Political interest T1 Others 913 2.83 0.93 5.588.000 Hardcore 114 3.29 0.81 Political interest T2 Others 763 2.93 0.89 8.976.000 Hardcore 113 3.54 0.64 Political interest T3 Others 693 2.89 0.90 7.449.000 Hardcore 114 3.46 0.72 Asked opinions Others 842 0.44 0.76 4.31.00 Hardcore 110 0.82 0.87 Talked politics recently Others 911 0.25 0.43 8.70.00 Hardcore 114 0.66 0.48 Discussion frequency Others 225 0.55 0.50 1.37.173 Hardcore 75 0.64 0.48 Opinion strength T1 Others 915 1.25 1.05 0.51.608 Hardcore 114 1.30 0.85 Opinion strength T2 Others 904 3.43 0.63 4.31.00 Hardcore 113 3.68 0.52 Opinion strength T3 Others 915 3.48 0.66 4.15.00 Hardcore 114 3.72 0.52 Political efficacy Others 915 0.52 0.71 0.53.60 Hardcore 114 0.56 0.70 Age Others 915 3.09 1.24 0.68.50 Hardcore 114 3.01 Education Others 907 4.18 1.19 4.34.00 Hardcore 113 4.69 1.13 Social class Others 876 2.90 0.71 0.06.95 Hardcore 110 2.90 0.60 Political efficacy Others 915 0.52 0.71 0.53.60 Hardcore 114 0.56 0.70 Note. T1 = Time 1; T2 = Time 2; T3 = Time 3. a. Equal variances not assumed. in the Elmira data set. As evident in Table 1, in each of the three waves, the hardcore were more interested in the election than were the changeable respondents (the remainder of the sample). We also include interpersonal political discussion measures in Table 1. As evident from the table, the hardcore are more likely to be asked opinions and to have talked about politics with someone recently, but the difference between the groups was not significantly different in terms of the frequency of political discussion. All three means indicate the difference is in the expected direction. The remaining variables included in the table indicate significant differences in educational level (with the hardcore more educated) and strength of opinions (with the hardcore respondents opinions stronger than the general public for two of the three time points). 147

COMMUNICATION RESEARCH April 2001 Our first hypothesis was that the changeable would show a positive relationship between the perceived trend of support for their position (the climate of opinion) and their public expression of opinion (candidate preference). We use a hierarchical logistic regression because the dependent variable is a simple dichotomous variable (coded 0 for those who did not express their opinion and 1 for those who did express their opinion). Cohen and Cohen (1983) and a number of other texts provide an explanation of the use of logistic regression for dichotomous variables. Table 2 provides the results of a hierarchical logistic regression of opinion expression on the demographic and spiral of silence variables. As evident from the table, after controlling for age, social class, education, and social isolation, the perception of the opinion climate was a strong predictor of opinion expression at all three time points, supporting the hypothesis. Our second and third hypotheses focus on social isolation. The second hypothesis was that social isolation would be negatively related to opinion expression. As provided by Table 2, Wave 2 and Wave 3 data offer support for the hypothesis; the Wave 4 coefficient was in the appropriate direction, but not statistically significant. The test of the third hypothesis, that social isolation would interact with perception of the climate of opinion in predicting opinion expression, also offers a reasonable level of support. Wave 2 and Wave 4 interactions were statistically significant, but Wave 3 was not. These results provide a fairly clear indicator that social isolation does play a part in opinion expression, although the relationship is not consistently evident or particularly strong. Table 3 provides a test of the hypotheses using the total of opinion expression (Wave 2 + Wave 3 + Wave 4) as the dependent variable. Under these conditions, the climate of opinion and educational level of the respondent are the only significant predictors of opinion expression. Given that these data do support the hypothesized relationship between the perceived climate of opinion and opinion expression, we felt it appropriate to provide illustrations of the trend of relationships evident in the data. These illustrations are provided in Figures 1 and 2. In Figure 1, we track the change in Truman preference across the four waves of data for three groups of people: the hardcore, those changeable who had no preference at Wave 1, and the changeable with a preference at Wave 1. As is evident in the table, there is a clear trend for Truman preference among those with no preference at the beginning, climbing to about 18% at Wave 2, 21% at Wave 3, and slightly more than 50% reporting having voted for Truman following the election. The changeable with a preference listed show a clear move toward Truman as well, moving from about 25% in Wave 1 to slightly more than 35% at final vote. The hardcore, of course, remain static by definition. 148

McDonald et al. The Spiral of Silence in 1948 Table 2 Unstandardized Coefficients for Logistic Regression of Speaking Out About Candidate Preferences on Spiral of Silence and Background Variables (changeable respondents only) Wave 2 Wave 3 Wave 4 Demographic block Age.042.221.288 Social class.032.036.127 Education.029.381***.377*** Nagelkerke R 2 (incremental).001.068.081 Social isolation.398.387*.196 Nagelkerke R 2 (incremental).000.011.005 Perception of opinion climate.387**.304**.524*** Nagelkerke R 2 (incremental).030.013.011 Isolation Climate Interaction 1.316**.282.801* Nagelkerke R 2 (incremental).019.001.009 2 for equation (df = 6) 16.257** 45.415*** 62.735*** n 605 670 782 *p <.10. **p <.05. ***p <.01. Table 3 Standardized Regression Coefficients From Analysis of Total Opinion Expression Regressed on Background and Spiral of Silence Variables Changeables Age.051 Social class.014 Education.110** Incremental R 2.020** Social isolation.009 Incremental R 2.000 Perception of opinion climate.164*** Incremental R 2.033*** Isolation Climate Interaction.043 Incremental R 2.001 **p <.05. ***p <.01. Although Figure 1 provides evidence of preference trends, evidence for the theory requires trends in the climate of opinion if it is to be acceptable as a possible explanation for the polling misprediction. Figure 2 provides such evidence in a graph depicting the trend in the perception that Truman might win the election. From Wave 1 to Wave 2, there is a marked, sharp decline among all three groups in the perception of a Truman victory. Between Wave 2 and Wave 3, however, the trend changes dramatically and in 149

COMMUNICATION RESEARCH April 2001 Figure 1. Percentage of Truman Preference of Hardcore, Changeable, and Those With No Preference at Wave 1 Note. T4 is the percentage reporting that they voted for Truman. exactly the manner we might expect. Those most sensitive to the trend (the changeables with no base preference) show a near doubling in the percentage believing that Truman will win the election (from 12% to 21%); those changeable respondents who indicated a preference at Wave 1 show the next highest slope (from 14% to 16%). The analysis of the hardcore does not show an increase in the percentage of those foreseeing Truman as the winner, but instead, reveals what might be described as a leveling off of the perception that Truman would lose the election (that is, the declining negative trend levels off). Discussion We began with the idea that the spiral of silence might offer an explanation for some of the problems that resulted in the 1948 misprediction of Thomas Dewey s election as president of the United States, and so sought to 150

McDonald et al. The Spiral of Silence in 1948 Figure 2. Perception That Truman Would Win the Election: Hardcore, Changeable, and Those With No Baseline Candidate Preference test the spiral of silence as one potential explanation for the prediction problem of 1948. Overall, these data show consistent support for the spiral of silence. The question then remains: Could it have been operating in 1948, and thus be a partial explanation for the misprediction? The spiral of silence offers a social-psychological explanation that was hinted at but not investigated in the major analyses after the election. Mosteller et al. (1949) noted that some people would not provide their preferences to pollsters because they preferred Truman, and they felt he was going to lose, and others did not provide their preference because they preferred Thurmond, and they did not want to appear to be prejudiced or bigoted. Similar and consistent observations were noted by Dinerman (1948), who noted in preliminary analyses of the Elmira data set that there was a consistent trend toward Truman among those people who were leaning toward one candidate or the other. Lazarsfeld s own Erie County study of the 1940 election (Lazarsfeld et al., 1948) relied on the bandwagon effect for part of its explanation of elections. 151

COMMUNICATION RESEARCH April 2001 Mosteller et al. (1949) note that the weight of the evidence was that there was a last-minute swing to Truman. Although pollsters were beginning to pick up on that swing, they conclude that there was no way for pollsters to analyze the data in such a way as to be able to infer the implications of that swing for the election. According to both a Gallup and a Roper poll (they both obtained the same results), about one in seven voters decided who to vote for within the last 2 weeks of the election, and three fourths of those decided for Truman (Mosteller et al., 1949). As Mosteller et al. (1949) noted, there was a total change of only about 7 percentage points during the final weeks of the election. What cost Dewey the election was that the overwhelming majority of those who changed voted for Truman rather than Dewey. In our Elmira analyses, the perception of opinion climate accounts for between 1% and 3% of the variance in opinion expression, reminding us of Tukey s comment (reported in Cohen & Cohen, 1983) that the.04 difference between.01 and.05 is much more important than the.04 difference between.48 and.52 for almost all purposes except winning elections (p. 266). Although the Elmira sample might be atypical in that there are fewer undecideds than in the national average, the same trend is evident in these data. The questions then center on whether the spiral of silence could be a factor in the last-minute trend. The answer from these data would appear to be yes, in the sense that the trend is evident toward Truman as a preference. However, that is only evidence of a last-minute swing. Combine that swing with an increasing trend in the perception that Truman would actually win the election and statistical analyses that correlate the perception of climate with opinion expression, and the evidence is a bit stronger and much more compelling. It would appear that one reason for the large number of undecided respondents was that there was, early on, a perception that Truman was going to lose. As that perception began to swing the other way, albeit very slowly, more respondents began to provide their preference, and this was the slight trend toward Truman reported by Gallup and Roper. If we extrapolate from these data, it could easily be that during the final weeks of the campaign, when the polls were not being taken, the trend toward Truman as victor probably continued, and more and more undecideds began to feel confident enough to express their preference. The result was a shift that was just large enough to help change the course of the election. A criticism of the spiral of silence as an additional explanation in the present work is that the percentage of variance accounted for is slight. The first explanation for the small percentage of explained variance rests in the dependent variable, which is a simple dichotomous coding of whether someone 152

McDonald et al. The Spiral of Silence in 1948 answered a question about their preferred candidate. The situation is a much stricter test than those commonly found in communication and other social sciences in which responses to several scales are provided within a few minutes. In those cases, there are still relatively weak correlations, but the time lag between sets of questions is almost nonexistent. In the present study, the dependent variable is of a completely different type (a behavior) than is the independent (a perception), and the time lag between measures is sometimes several months. A situation such as the 1948 election provides a context for understanding when small, statistically significant effects are also of practical importance. One wonders about the extent to which Berelson and his colleagues might have gleaned something such as the spiral of silence from these data. Clearly, they were familiar with the experiments of Asch (1951), as well as many of the classic writers whose work serves as a background to Noelle-Neumann s (1974) work. They also had access to Dinerman s (1948) analyses before they wrote the book. Indeed, some of the principles related to the spiral of silence are listed in the end of the second chapter. Particularly noticeable is the concept of cross-pressures, which is somewhat similar to the spiral of silence in the sense that in both cases, a pressure for change develops just through observing others with an opinion different from your own. However, with cross-pressure, as with a number of other explanations and hypotheses in Berelson et al. (1954), much emphasis is placed on the social group as an explanation in itself. Attitudes and values were seen to be a part of group membership, and the conflict seen by Berelson and his colleagues would occur when an individual belonged to two reference groups with conflicting pressure. These ideas would appear to be very central to the 1950s social-psychological viewpoint and may have blinded them to other background processes. A symposium on opinion formation published in Public Opinion Quarterly in 1950, and clearly related to the 1948 election prediction problem, cites Doob s explanation of public opinion as a starting point: Public opinion refers to people s attitude on an issue when they are members of the same social group (Hartley, 1950, p. 669). Later in the chapter, mention is made of Lazarsfeld et al. s (1948) research that indicates that individuals tend to vote in conformity with the preferences dominant in social groups with which they are most closely identified (Hartley, 1950, p. 670). Although cross-pressure comes up as what occurs when the values of two reference groups collide within one individual, no attention is paid to the idea that individuals might have opinions that are not consistent with their reference groups, which is the situation that occurs in the spiral of silence. 153

COMMUNICATION RESEARCH April 2001 The spiral of silence assumes that attitudes can be more individualistic; people have opinions that are derived through a variety of means, and reference groups play a part but are not determinate. The theory goes well beyond the normative explanations current in the 1950s and derives its strength by suggesting that something that is a very natural aspect of social creatures our need to belong is a strong driver of our social behavior. References Albig, W. (1956). Modern public opinion. New York: McGraw-Hill. Asch, S. E. (1951). Effects of group pressures upon the modification and distortion of judgments. In J. H. Campbell & H. Hepler (Eds.), Dimensions in communication: Readings (pp. 200-210). Pittsburgh, PA: Carnegie. Berelson, B. R., Lazarsfeld, P. F., & McPhee, W. N. (1954). Voting. Chicago: The University of Chicago Press. Brennan, E. E. (1949). Last minute swing in New York City presidential vote. Public Opinion Quarterly, 13, 285-298. Campbell, A. (1950). The pre-election polls of 1948. International Journal of Opinion and Attitude Research, 4, 27-36. Clifford, C. (1991). Counsel to the president. New York: Random House. Cohen, J., & Cohen, P. (1983). Applied multiple regression/correlation analysis for the behavioral sciences (2nd ed.). Hillsdale, NJ: Lawrence Erlbaum. Dinerman, H. (1948). 1948 votes in the making a preview. Public Opinion Quarterly, 12, 585-598. England, L. R. (1949). Comments on mass observation of election forecasts. International Journal of Opinion and Attitude Research, 3, 355-358. Gallup, G. (1949). The Gallup poll and the 1950 election. Public Opinion Quarterly, 13, 16-22. Glynn, C. J., & McLeod, J. M. (1984). Public opinion du jour: An examination of the spiral of silence. Public Opinion Quarterly, 48, 731-740. Glynn, C. J., & Park, E. (1997). Reference groups, opinion intensity, and public opinion expression. International Journal of Public Opinion Research, 9(3), 213-232. Gosnell, H. F. (1952). Participation in the forthcoming election. Annals of the American Academy of Political and Social Science, 283, 156-160. Guest, L. (1949). Have these sources of error been fully explored? International Journal of Opinion and Attitude Research, 2, 507-509. Hartley, E. L. (1950). The social psychology of opinion formation. Public Opinion Quarterly, 14, 668-674. Hyman, H. (1953). Trends in public opinion polling since 1948 and their probable effect on the 1952 election predictions. In Proceedings of the conference on testing problems (pp. 64-69). Princeton: Educational Testing Service. Jacquemyns, G. (1950). The polls and the American elections of 1948. International Journal of Opinion and Attitude Research, 3, 13-22. 154

McDonald et al. The Spiral of Silence in 1948 Katz, D. (1949). An analysis of the 1948 polling predictions. Journal of Applied Psychology, 33, 15-28. Krassa, M. A. (1988). Social groups, selective perception, and behavioral contagion in public opinion. Social Networks, 10, 109-136. Lasorsa, D. L. (1991). Political outspokenness: Factors working against the spiral of silence. Journalism Quarterly, 68(1-2), 131-139. Lazarsfeld, P. M., Berelson, B. R., & Gaudet, H. (1948). The people s choice: How the voter makes up his mind in a presidential campaign. New York: Duell, Sloan & Pearce. Mosteller, F., Hyman, H., McCarthy, P. J., Marks, E. S., and Truman, D. B. (1949). The pre-election polls of 1948. New York: Social Science Research Council. Noelle-Neumann, E. (1973). Return to the concept of powerful mass media. Studies of Broadcasting, 9, 68-105. Noelle-Neumann, E. (1974). The spiral of silence: A theory of public opinion. Journal of Communication, 24(2), 43-51. Noelle-Neumann, E. (1984). The spiral of silence: Public opinion our social skin. Chicago: The University of Chicago Press. Noelle-Neumann, E. (1985). The spiral of silence: A response. In Sanders, K. R., Kaid, L. L., & Nimmo, D. (Ed.), Political communication yearbook 1984 (pp. 66-94). Carbondale: Southern Illinois University Press. Noelle-Neumann, E. (1991). The theory of public opinion: The concept of the spiral of silence. In J. A. Anderson (Ed.) Communication yearbook 14 (pp. 256-287). Newbury Park, CA: Sage. Oshagan, H. (1997). Reference group influence on opinion expression. International Journal of Public Opinion Research, 8(4), 335-354. Ross, I. (1968). The loneliest campaign: The Truman victory of 1948. New York: New American Library. Salmon, C. T., & Neuwirth, K. (1990). Perceptions of opinion climates and willingness to discuss the issue of abortion. Journalism Quarterly, 67(3), 567-577. Salmon, C. T., & Oshagan, H. (1990). Communication size, perception of majority opinion and opinion expression. Public Relations Research Annual, 2, 157-171. Willnat, L. (1996). Mass media and political outspokenness in Hong Kong: Linking the third-person effect and the spiral of silence. International Journal of Public Opinion Research, 8(2), 187-211. 155