Richard D. Anderson, Jr.

Department of Political Science

UCLA

randerso@ucla.edu

Shanto Iyengar

Department of Communication

Department of Political Science

Stanford University

siyengar@leland.Stanford.EDU

June 1999

Prepared for delivery to the Annual Meeting of the American Political Science Association, Atlanta, Georgia, September 2-5, 1999

Copyright American Political Science Association 1999

Studies of election campaigns stipulate either identity reinforcement or information transmission as the processes by which voters choose between candidates. Identification with a political party ("partisanship") is the single most important determinant of whether and how the citizen votes (see Kinder, 1997, Rosenstone and Hanson, 1993). Salient group identities also exert important effects on voting. Voters who identify with candidates on racial grounds, for example, are more attentive to campaign messages (Tate, 1991) and more informed about public affairs in general (Bobo and Gilliam, 1991).

In keeping with the long-standing conclusion of "campaign effects" research (for a recent review of this literature see Iyengar and Simon, 2000), Senate campaigns are known to reinforce the effects of identification on voters' evaluations of the candidates (Franklin 1991). While reinforcing partisan identity, campaigns also increase voters' information about candidates. Franklin (1991) studied voters' knowledge of incumbent Senators' ideological location by comparing Senators who were campaigning for office to those who were not. This natural experiment enabled him to show that campaigns increased both citizens' information about candidates and the precision of their knowledge. Krasno (1994, 99) showed that hard-fought campaigns for the Senate sharply decreased the average evaluation (as measured by the "feeling thermometer") of incumbent Senators relative both to Senators who face lackadaisical challengers and to Senators not running in a given election year. Using experimental methods, Ansolabehere and Iyengar (1995) showed that citizens exposed to campaign advertising were more able to place Senatorial candidates correctly on salient issues than citizens not exposed to campaign advertising.

Reinforcing identity and transmitting information are conceptually distinct effects of campaigns. Yet, to our knowledge no one has disentangled the identity component of campaign effects from the informational component. Indeed, in many studies, information about candidates or parties is taken as a measure of identification with the candidate or party. We agree that identity facilitates the acquisition of information, but use of information as a proxy for identification does not allow the two to be disentangled. One might suppose that any information encountered in campaign messages reminds voters of an identity between their personal political preferences and the preferences of the candidate. This view is implicit in Franklin's decision to treat identity as a control variable in his study of informational effects of Senate campaigns and even more to measure identity by a combination of respondents' own ideological location and the interaction between the respondent's ideological location and evaluation of the candidate (again measured by the "feeling thermometer"). Franklin hypothesized, and his empirical data confirmed, a "projection effect" by which voters initially judge the incumbent by projecting their own identity onto a favorably evaluated incumbent and then use information gained during the campaign to correct this initial assessment.

At the same time, Franklin also considered another possible conception of how campaigns evoke identity. Citing Fenno's (1978) observation that incumbents spend much of their time in the district persuading citizens that the incumbent is like one of them, Franklin raised the question, which he could not answer for lack of fine-grained survey data, whether campaigns build identification with candidates independently of the information that the campaign provides about the candidate's issue positions. To date, however, neither Franklin nor anyone else has followed up on this basic question.

We attempt to disentangle the effects of campaigns on identity from their effects on information using an experimental manipulation that retains the informational component of campaign cues but suppresses identity cues by making the candidate appear different from the voter. The experimental technique takes advantage of the fact (well documented by linguists) that any linguistic text contains two kinds of cues, "informational" and "relational" (Brown and Yule, 1983, 1-3). People use messages to discover information about the world external to the sender and the receiver, including the traits of sender and receiver. Simultaneously, they also use messages as cues to whether they belong to the same social group as the sender. An obvious example of a relational cue is the use of a foreign language. The obstacle presented by a foreign language to interpreting informational cues makes it clear to the receiver that the sender belongs to a different social identity. While this example may seem far removed from politics, there are many historical and contemporary examples in which ruling elites have deliberately confined their political communications within a language foreign to subject populations. This practice served as a means (conscious or unconscious) of emphasizing the separation between elite and popular identities (see Anderson, 1997a). Within any given natural language, people use a variety of more subtle cues to judge similarity of social identity between themselves and those with whom they enter into oral or written communication (Gumperz and Cook-Gumperz 1981). Messages rich in cues to the separateness of the sender's social identity from that of the receiver are known as "distancing" messages. By exposing respondents to campaign messages in either the original version or in a version rewritten to be rich in distancing cues while preserving informational cues, we can test whether political messages build identity with respondents independently of the information they provide.

We have conducted experiments both in Russia and in the United States. Russia presents a case of a polity in which "partisanship," at least in the U.S. sense, is only beginning to emerge (White, Rose and McAllister, 1997, 134-141). Partisanship is weakly developed in Russia because until 1990 there were no competing political parties to which people could develop partisan attachments. There was of course a single Party, but it did not compete, only six percent of the population were permitted to become Party members, and this "Party" functioned as a state personnel agency and territorial administration rather than as a political party in the electoral sense.

One series of experiments in Russia compared responses to authentic texts from the Soviet era with responses to authentic texts from the post-Soviet, electoral era. Soviet texts are known to be more distancing than post-Soviet texts (Anderson, 1996). The results indicated that the authentic Soviet texts elicited more disaffiliative and fewer affiliative comments than either of the two main ideological variants of post-Soviet texts (Anderson, Chervyakov and Parshin 1995, Anderson 1997b). A second experiment presented Russian respondents with authentic electoral texts or with the exact same texts rewritten in distancing form. (The rewriting was done by a native Russian speaker who is a lifetime resident of Moscow and holds an advanced degree and academic post in linguistics.) In this experiment, the partisan differences in responses when respondents read the original texts nearly disappeared when the respondents read the distancing rewrite (Anderson, Chervyakov and Parshin n.d.). Together, these experiments suggest that distancing language has the effect of suppressing partisan identification when partisanship is weak.

Even though partisanship has weakened in the United States since 1960 (see, for instance, Wattenberg, 1984), the United States remains a case of a polity with a mature level of partisanship. In an experiment conducted with campaign speeches by both major candidates in the 1996 presidential election, we found that distancing language had no identifiable overall effects. However, this null result concealed specific effects among a particular subset of partisans. In particular, as Table 1 shows, attentiveness to politics conditioned the effect of distancing language on partisans. In Table 1, we measure partisan identification with the candidates by the absolute value of the difference between "feeling thermometer" scores for the 1996 candidates, Bill Clinton and Bob Dole. If individuals evaluate their party favorably and campaign exposure increases identification with the candidate of their party, then the individual should evaluate the preferred candidate more favorably, relative to the opposing candidate, over the course of the campaign. Thus for any individual who prefers one candidate, the difference between candidates should increase, regardless of which candidate the individual prefers. As in Russia, relatively weak partisans (defined as those who do not pay very much attention to politics) responded to distancing texts by suppressing their sense of partisan identity. In the inattentive column, exposure to a single original campaign text increased the perceived difference between candidates, while exposure to a single distancing version of the same text decreased the perceived difference. By contrast, among strong (attentive) partisans, exposure to an original presidential speech had no effect, but exposure to a distancing version actually reinforced partisanship. This interaction between attentiveness and the distancing manipulation was statistically robust (F=5.47, p<0.004).

These data are consistent with the results observed in Russia. In the Russian experiments the effects of distancing language on voters' evaluations of political messages varied according to the attentiveness of voters. Inattentive voters became more negative when cued with distancing messages, while attentive voters became more positive when cued with distancing messages (F[1,747]=2.73, p<.099 (controlling for a main effect of the difference between the two cities where the Russian experiment was conducted). As in the United States, this pattern was not observed among political independents (in the Russian case, defined as non-voters in the 1996 presidential election).

Campaign effects are often thought to be minimal, but this may be because they are hard to observe during periods of frequent exposure to campaign messages. We have developed an experimental paradigm that is designed to correct for the subject's prior campaign exposure. It may be that campaign messages appear to exert only minimal effects because the ordinary experiment or survey compares people exposed to one or a few additional campaign messages to people who have already seen many. If everyone is exposed to campaigning, the incremental effect of another exposure can in fact be expected to be marginal. Our method bypasses this difficulty by comparing people exposed to an additional campaign message with people who are exposed to a message designed to subtract from the existing level of campaign effects.

Our most recent experiment was conducted during the last two weeks of the 1998 Senate campaign in California, pitting the incumbent Barbara Boxer against Republican challenger Matt Fong. The dependent variable was once again identification with the candidates, measured by the absolute value of the difference in feeling thermometers scores for Boxer and for Fong. The feeling thermometer provides both a reliable gauge of respondents' preferences between candidates and a measure of likelihood of participating, as people who see wider differences between candidates are more likely to vote. The use of the absolute value of the difference allows us to analyze respondents regardless of their partisanship, an important gain in a location where partisans of one party (Republicans) are relatively hard to find.

The experimental manipulation involved presentation to subjects of a printed text, incorporated in the middle of the questionnaire, representing either the original or a distancing version of the opening statements by both candidates in their second debate, held on live television on October 12, 1998, but watched by relatively few Californians and reported only in summary form by Los Angeles newspapers.

The distancing was composed by taking advantage of linguists' research on how English speakers communicate when they emphasize the separateness of their social identity from that of their listeners. In such situations, English speakers make abundant use of nouns in preference to verbs, avoid the first person singular in favor of the first person plural or third person impersonal forms, prefer the possessive over personal pronouns, and avoid contrastive conjunctions and particles (e.g., "but," "not," "however") that add nuance to utterances (for literature, see Anderson 1996). To create the distancing text, we therefore rewrote the opening statements by each candidate to replace all verbs, except the minimally necessary "to be," "to have," and the future auxiliary, with a corresponding noun form, preserving the identical semantic root whenever possible. We also transformed all first person singulars into plurals or the third person, where possible we converted personal pronouns into possessives, and we eliminated contrastives.

The resulting revisions are only subtly different from the originals of these quite short texts:

Distancing Version

Barbara Boxer: At the time of our election to the US Senate, our promise was hard work for a reversal of the direction of the economy, reduction of crime, and a stand for education, the environment, and choice. Our progress has been real; there is still more for future achievement, like quality education, health reform, more community policing, and sensible gun control. On all these issues and more, our opponent is different from us. We are pro-choice; it is Mr. Fong's statement that the legal guarantee to women of a right of choice is wrong. On gun control, he is against any new laws on gun control. It is his statement that there is too much enforcement of the Clean Water Act, the Clean Air Act, even the Food Quality Safety Act. We have one of the highest environmental voting records in the Senate. We are the sponsor of the Patient Protection Act under which HMO's will be accountable; he is not in support of that point of view. So the differences are many: we have the support of the Sacramento Bee, the San Francisco Examiner, he has the support of Newt Gingrich and Oliver North.

Original Version

Barbara Boxer: When you sent me to the US Senate, I promised that I would work hard to turn this economy around, to reduce crime, and to stand up for education, the environment, and choice. We've made real progress, but there's still more to do, like quality education, health reform, more community policing, and sensible gun control. On all these issues and more, my opponent and I differ. I'm pro-choice; Mr. Fong says that the law that guarantees women a right to choose is wrong. On gun control, he's against any new laws on gun control. He says there's too much enforcement of the Clean Water Act, the Clean Air Act, even the Food Quality Safety Act. I have one of the highest environmental voting records in the Senate. I sponsored the Patient Protection Act which will hold HMO's accountable; he doesn't support that point of view. So the differences are many: I have the support of the Sacramento Bee, the San Francisco Examiner, he has the support of Newt Gingrich and Oliver North.

Distancing Version

Matt Fong: Tonight our obligation is a civil and honest conversation about the issues important to most Californians. California's recent experience has been survival through some pretty tough times. In the 21^st century, in our belief, a better performance is possible. Our goal is removal of the barriers that, in our belief, are responsible for prevention of realization by Californians -- and Americans -- of their full potential. The significance of this is a focus on ways for the improvement of public education. The significance is ways for the protection of our families and neighbors and employees and employers from criminals here at home and our country and vital interests abroad from terrorists. The significance is help to hard-working Americans, who are at work in one or two jobs, their discovery that with harder work their paycheck is less, by reduction of taxes and by return of less to Washington, DC. As your State Treasurer, our promise was a more accountable government. Our practice was the retention of the memory that it is your money. And as your United States Senator, our practice will be an address to the key issues in front of us, in ways that will be a source of unity instead of division.

Original Version

Matt Fong: Tonight we should be engaging in a civil and honest conversation about the issues that matter to most Californians. California has survived some pretty tough times. In the 21^st century I believe that we can do better. I want to remove the barriers that I believe are keeping Californians -- and Americans -- from realizing their full potential. This means focusing on ways to improve public education. It means ways to protect our families and neighbors and employees and employers from criminals here at home and our country and vital interests abroad from terrorists. It means helping hard-working Americans, who are working one or two jobs, find that they are working harder and finding less in their paycheck, by lowering taxes and returning less to Washington, DC. As your State Treasurer, I promised to make government more accountable. I have never forgotten it's your money. And as your United States Senator, I will address the key issues that are facing us, but in ways that will unite us, not divide us.

As is evident, the distancing versions are marginally longer (one line in Boxer's case, three half lines for Fong). In our experiment on the presidential campaign, we used about three pages of single-spaced text from each candidate as a stimulus, with quite heavy revision; in this experiment we have relied on much shorter texts as stimuli and introduced many fewer revisions. Of course the revised texts read like gobbledygook, but that is the point: "Gobbledygook is an icon, pure and simple, of non-linguistic formal behavior" (Haiman 1985, 154).

Respondents were randomly assigned to five experimental conditions. A control group, consisting of approximately one third of the respondents, filled out the questionnaire without seeing any textual stimulus. Approximately one-sixth of the respondents in an "original" condition saw the originals of the Fong and Boxer statements. Two groups, each approximately one-sixth of the respondents, in "mixed" conditions, saw either Boxer's distancing and Fong's original statement or Fong's distancing and Boxer's original statement. A final group of approximately one-sixth of the respondents in the "distancing" condition saw both Fong's and Boxer's statements in the distancing version. In all four conditions with a textual stimulus, the statement by Fong, who went first in the debate, appeared in a column on the left-side of the page, while Boxer's statement appeared in a column on the right side of the same page.

The sections of the questionnaire preceding exposure to the stimulus texts included questions about a variety of forms of political identity, including race, political party, and ideology. We measured race by a question with five possible answers: Asian, Black, Hispanic, White, Other. We measured partisanship by a single question with a seven point scale: Strong Democrat, Moderate Democrat, Leaning toward the Democrats, Independent, Leaning toward the Republicans, Moderate Republican, and Strong Republican. We also separately measured ideology two ways, by a single 5-point question (Liberal, Moderate Liberal, Moderate, Moderate Conservative, and Conservative) and by a battery of questions about affirmative action, the death penalty, and abortion.

Respondents were recruited in shopping malls (and in some cases, a pedestrian shopping area not part of a mall) in the western part of the Los Angeles metropolitan area. The interviewers were twelve undergraduate students enrolled in a seminar taught by Anderson at the University of California, Los Angeles. The students received partial course credit for handing out and entering the data from the questionnaires. Until the collection of respondents was complete, students were instructed not to read the questionnaire or debriefing document, and they were not told the purpose or design of the experiment. They offered each potential respondent ten dollars in return for participating in "a research study concerning the November election." Interviews were conducted over the last two weeks of the campaign, with the earliest on 17 October and the last on 2 November, the day before the election.

We anticipated that race of the student interviewer would strongly affect willingness to participate, and we felt ethically obliged to alert the students to this possibility. Of course, the warning to the students may have exacerbated the tendency that some of them observed for persons not of their own race to reject the offer of participation. The one African-American student in the seminar reported that almost exclusively African-Americans were willing to accept his invitations to participate, and a Chicano student and a male Asian-American reported similar difficulties. Female Asian-American students and white students reported that they were generally able to obtain favorable responses.

We failed to anticipate the degree to which young interviewers recruiting in shopping malls would collect an overwhelmingly youthful sample. Table 2 shows the distribution of respondents by age and race. The youthfulness of the sample is disadvantageous only because the number of independents is increased, who for theoretical reasons are not very interesting participants. The sample is particularly useful because its considerable racial diversity allows us to test for the impact of racial identity on processing of political messages. Whites make up only 53 percent of the sample, while African-Americans compose nearly a quarter, and both Asian and Hispanic respondents are present in considerable numbers.

(7 missing data)

Although this is a convenience sample, we have good reason to believe that it provides a reasonable approximation of relationships that we expect to hold in the local population. In particular, we found that participants’ voting preferences were significantly influenced by the "standard" antecedents of senatorial vote choice including party identification, liberal-conservative ideology, approval of presidential performance, and positions on major policy issues (see Squire and Fastnow, 1994, Atkeson and Partin, 1995, Delager 1996).

A fundamental problem for experimental research on campaigns is the variability of the researcher's expectations by electoral context. Any experimental methodology should seek replication as one of its principal aims, but if findings depend on context and context differs from one election to another, what counts as replication? Senate elections are known to be different from Presidential elections, and we should not necessarily expect to find the identical pattern of results in an experiment on Senate elections that we find in experiments on Presidential elections.

The most fundamental difference between Senate and presidential elections concerns "salience" – the amount of attention the media and the electorate are paying to the contest. No matter how one measures attention or interest, presidential campaigns outdraw Senate campaigns (e.g., Campbell 1962, Rosenstone and Hansen, 1993, 178-79). Because people typically pay much less attention to Senate campaigns, one cannot necessarily expect that distancing language will impede partisanship among inattentive voters as it did in the presidential election study and in the Russian experiment. Indeed, an attentive voter may be no more attentive to a Senate contest than an inattentive voter is to a presidential election.

These differences in electoral context have obvious consequences for campaign strategy. Presidential candidates win not just by retaining their strong partisans, but also by mobilizing weak partisans and converting some who lean to the other party. The speeches by President Clinton and Senator Dole were presumably composed with this strategic principle in mind and aimed at general audiences. In contrast, anticipating low turnout, Senate candidates in off-year elections must concern themselves primarily with mobilizing their partisan base. Senator Boxer and candidate Fong presumably knew that few people other than strong partisans would watch the debate or follow media coverage of their remarks. Their strategic objective was to mobilize their staunch partisans in the knowledge that voting among weak partisans, leaners and independents would tail off sharply in an election with low turnout. Thus while the original texts in the presidential campaign should have sought to identify the candidate with weak partisans and leaners, the originals in the senate campaign should have been designed to identify the candidate with strong partisans.

In addition to their lower visibility, Senate elections differ from presidential contests in the relative importance of party identification as a perceptual cue. Partisan identity is manifestly a self-categorization. Our question asks the respondents to categorize themselves as Democrats, Independents, or Republicans of varying intensity. Any human categorization combines items that are more prototypical of the category with items that are less prototypical. Armchairs, for example, are more prototypical of the category "chair" than chaise lounges. In general, items become prototypical of a category when they evoke more perceptual cues that are shared with other members of the same category and fewer cues that are shared with members of other categories (Rosch 1978, 28-37). But as Rosch (1978, 41-43) takes pains to point out, perceptual cues are not features of the object, but assignments to the object by the perceiver. Being more familiar to partisans, presidential candidates will be assigned more perceptual cues in general, and are consequently more likely to be seen as prototypical of their party than Senatorial candidates. Consequently the relationship between partisanship and evaluations of candidates is likely to be weakened in a Senate campaign, with incumbency and other attributes competing for influence.

In addition to Senator Boxer's incumbent and liberal standing, she was further removed from her party by the circumstance that the election was held at a time when the most salient public issue was the impeachment of President Clinton in connection with his affair with Monica Lewinsky. Boxer chose silence as the wisest policy when cross-pressured between her partisan loyalties to the President and her reputation as an advocate of women's rights, particularly known for her championing of Anita Hill's accusations of sexual harassment against Clarence Thomas. Her challenger Fong was also an atypical Republican. Not only is he Chinese-American, but he is also the adoptive son of the well known state politician March Fong Eu, a long time Democrat who appeared in television commercials on her son's behalf. Fong is also a relatively moderate Republican, favoring the legality of abortions during the first trimester of a pregnancy. By pitting a moderate racial-minority male against a liberal white female (who is also Jewish, although few voters may know that), the California contest cut across some of the usual cleavages separating Democratic from Republican voters.

Because attentive voters are less attentive to Senate elections, especially in off-years, than they are to Presidential elections, because campaign strategy differs between Senate and presidential races, and because partisanship in general may not shape evaluations of Senate candidates in quite the same way as it does evaluations of more prototypical Presidential candidates, we cannot expect the experimental manipulation to interact with political identity in precisely the same manner as it did in our experiment on the Presidential election. In particular, we cannot expect to replicate the interaction between attentiveness and exposure to distancing language, in part because few if any voters in a Senatorial election are as attentive as attentive voters in a Presidential election, because partisans are more partisan toward presidential candidates, and because the presidential texts were aimed at weak partisans while the Senate texts targeted strong partisans. For all these reasons, we expect any effects of the manipulation to be concentrated among strong partisans, to be absent or weaker among weak partisans, and to be reversed among independents. Whereas the originals of the Presidential speeches activated partisanship among inattentive partisans and failed to affect attentive partisans, in the Senate study we expect the original versions of the candidates’ opening statements to activate strong partisans and to exert weaker effects on moderate partisans and leaners, and to demobilize independents. Conversely, the distancing versions should attenuate the mobilizing effect of the originals among strong partisans and the demobilizing effect among independents, while mixed stimuli (the coupling of an original by one candidate with a distancing version of the other candidate's text) should produce an effect intermediate between either pure version.

The race of the candidates introduces a second expectation about the interaction between the experimental manipulation and race. Among African-Americans, candidates of the same race draw citizens' attention to politics (Bobo and Gilliam, 1991;Tate 1991). We postulate that this effect is not confined to one racial group, so that the generally higher political activity of whites is attributable in part to the prevalence of whites among political candidates and the generally lower political activity of Asian-Americans and Hispanics is attributable in part to the dearth of candidates from these racial groups. Accordingly, in this election the presence of an Asian-American and a white candidate should stimulate identification among respondents of those racial groups and depress it, in the absence of any other campaign cue, among African-Americans, Hispanics and "others." Table 3 displays the absolute value of the feeling thermometer scores for Boxer and Fong among members of the control group whose racial identity is shared or not shared with the two candidates and who vary by intensity of their partisanship, while Table 4 shows the analysis of variance. People who belong to the same race as one of the candidates begin with a higher identification in all categories of partisanship, although among moderate partisans the difference is small. As indicated in the analysis of variance, racial identification is a much stronger predictor of difference in the feeling thermometer scores than political partisanship, another indication of the degree to which partisanship was relatively non-salient in the Boxer-Fong contest. We therefore expect that exposure to campaign texts will produce different effects on whites and Asian-Americans than on African-Americans, Hispanics and "others." Specifically, in the group that already identifies with the candidates on racial grounds, distancing texts should disrupt the degree to which information about the candidates' partisan issue stands reinforces identification. Original texts, in which relational cues reinforce informational cues, should reinforce identification among the strong partisans at whom the texts are aimed. In the group that does not identify on racial grounds, prior expectations of identification with either candidate are much weaker. Therefore exposure to information about partisan stands should be sufficient to remind strong partisans in this group of their partisan identification with the candidate; originals containing relational as well as informational cues should make this effect stronger. Mixed conditions should produce effects of intermediate strength among both those who do and those do not identify on racial grounds.

(N=172)

We begin by comparing the effects of exposure to any campaign message for participants who identify with the candidates on partisan and racial grounds. We compare strong partisans to the combination of moderate partisans, leaners and independents, and Asians and whites (racial identifiers) with Hispanics and African-Americans (non-identifiers). The mean thermometer differences are shown in Table 5, with the resulting analyses of variance in Table 6. As expected, the strong partisans display the greatest sensitivity to the campaign. Strong partisans who encounter any version of the candidates’ opening statements become more polarized in their views of the candidates. Exposure to the campaign, if anything makes weak partisans and independents evaluate the candidates more similarly. This diverging pattern yields a significant interaction between exposure to any campaign message and strength of partisanship (p < .014). In Senate campaigns, strong partisans appear to be the only ones listening.

Precisely the opposite pattern holds for racial identity. Among participants given no campaign stimulus, racial identifiers are far more polarized than non-identifiers (by a factor of more than 35 percent). Exposure to any message, however, has the effect of erasing this difference. While racial identifiers move slightly in the direction of less polarization when given a campaign message, non-identifiers move towards greater polarization. Thus, the mean difference for African-Americans and Hispanics (non-identifiers) increases from 22 to 35. This differential effect results in a robust interaction between campaign exposure and racial identity (p < .006).

The starkly divergent results for party and race suggest that exposure to the campaign leads voters to their partisan rather than racial ties with the candidates. In the absence of any message, racial identity is a much stronger predictor of candidate evaluations than partisan identity. (The racial gap in evaluations is twice as large as the partisan gap among participants in the control group.) As soon as participants encounter campaign messages, however, they fall into alignment along party lines, in many cases making Asian-American Democrats and white Republicans cross racial boundaries, while making African-Americans, Hispanics and "others" identify with a white or Asian candidate. In this sense, Table 5 reveals both the familiar reinforcing effect of campaigns with respect to party identification and a striking ability of campaign exposure to transcend race. By strengthening the effects of partisanship on candidate preference, the manipulation necessarily weakened the effects of racial identity.

Having found that exposure to the campaign polarizes strong party identifiers, but homogenizes racial identifiers and non-identifiers, we turn to the effects of relational vs. informational cues. Are the reinforcing effects of the campaign on partisanship weakened when we suppress relational cues in the message? Conversely, are the de-aligning effects of the campaign with respect to race muted when the message is distancing in content? In this analysis we combine the mixed and distancing conditions because both were found to differ significantly for strong partisans from the original condition but do not differ significantly from each other. Thus, the comparison is between identifiers (both partisan and racial) and non-identifiers who encounter the actual opening statements and those who encounter versions in which some of the relational cues were suppressed. Table 7 presents these results, with the corresponding analysis of variance in Table 8.

As expected, the effects of the distancing texts were limited to the strong partisans. Their mean thermometer difference fell by twenty percent -- eleven points -- when they encountered a text with fewer relational cues. Weak partisans and independents, who should not be expected to pay attention to relational cues, were entirely unaffected by the manipulation. The disproportionate effects of the manipulation among strong partisans is shown by the significant interaction between exposure to distancing language and strength of partisanship (p< .05). The same pattern was apparent when considering race as the basis for identifying with the candidates. Distancing language did not affect the evaluations of racial non-identifiers, but did reduce the evaluative gap between the candidates among identifiers. This interaction, however, is robust only if viewed as a one-tailed test (two-tailed p < .17). Original texts mobilize strong partisans more powerfully than do the distancing version, and the difference is robust. Original texts also mobilize racial identifiers more fully than distancing versions, but the difference is only marginally distinguishable from random variation. Given the manifest partisan content of the opening statements, this pattern is not unexpected. The candidates were appealing to party, not racial identifiers. Overall Table 7 leads to the conclusion that exposure to relational cues adds a stimulus to partisan reinforcement that is separate from exposure to partisan information.

In sum, in the particular context of the 1998 California Senate election, which was contested by candidates identifiably atypical of their respective parties, exposure to campaign information had the effect of eradicating differences between races in their evaluations of the competing candidates. Partisan identification was strongest when the exposure provided both informational and relational cues. Messages containing relational cues had the effect of doubling, from ten to twenty-points of difference in feeling thermometers, the partisan identification produced by messages containing informational cues alone.

Social identity theory tells us that people bond with similar others, even when the perception of similarity is based on the most trivial of attributes (see Tajfel and Turner, 1979, 1986). For political campaigns, the need to identify has both positive and negative implications. In the 1998 California Senate election, voters who knew little else about the candidates identified with them on the basis of their race. It should not come as a surprise that people recognize racial differences between candidates. The fact that they fail to correct for racial identification when they evaluate the candidates is normatively unwelcome. On the other hand, the fact that exposure to any campaign message produces a different similarity judgment, one based on political party affiliation rather than race, is normatively quite welcome.

Because strong partisans are most likely to attend to campaign messages in Senate races, the campaign especially affected their evaluations of the candidates, while leaving others’ perceptions of the contestants unaffected. The mobilization of strong partisans was significantly impeded when campaign messages were stated in distancing language. This result indicates that relational cues are necessary to intensify existing feelings of shared social identity. Messages that convey information, but which are stripped of relational cues, attenuate the classic reinforcement effect, failing entirely to elicit it among citizens with racial grounds for identifying with the candidates. For candidates, therefore, the dissemination of information about specific policy positions or programs is less important than the dissemination of information which makes them appear as "one of us."

The greater importance of relational over informational cues in campaign messages we take as normatively unambiguous. Much ink has been spilled about the "rational ignorance" of voters and about the inadequacy of media-based campaigns as arenas for "deliberation" and informed choice (see Fishkin, 1991, 1995). We would argue, with Schumpeter, that the value of democracy lies not in the sophistication or quality of voters’ choices but in the mere fact that voters do choose. Strategic politicians will define and pursue the interests of those most likely to engage in political choice; if that is some narrow segment of the population, politicians will define and pursue narrow interests, but if the public as a whole is inclined to choose, politicians will define and pursue the interest of the broader public (which is not the same, of course, as achieving it). In the campaigns we have studied, voters most likely to be prompted by the campaign (strong partisans in Senate races, weak partisans in presidential races) respond to relational cues in campaign messages by strengthening their identification with the candidates. Strengthened identification, in turn, is well known to increase the likelihood that voters join in choosing those who will govern them.