Five decades of research make it clear that the principal effect of American political campaigns is the activation and reinforcement of voters' underlying predispositions. No matter what the research methodology -- surveys, experiments, or focus groups -- or the particular stimulus -- news reports, paid advertising, or televised candidate debates -- exposure to campaign communication strengthens voters' support for "their" candidate (for reviews of the campaign effects literature, see Iyengar, 1996; Holbrook, 1996).

In the American political context, of course, party identification is the most fundamental of predispositions. Although most partisans are inclined to vote along party lines regardless, exposure to campaigns makes them even more likely to do so (Ansolabehere and Iyengar, 1995; Finkel, 1993; Gelman and King, 1993; Iyengar and Petrocik, 1998). By reinforcing voters' sense of partisanship, campaigns polarize the electorate into Democratic and Republican loyalists.

The reinforcing effects of campaigns have been detected for predispositions other than partisanship. In their classic study of the 1940 campaign, Lazarsfeld and his collaborators noted that voters came to choose between the candidates in a manner consistent with their group interests. While working class Catholics moved towards Roosevelt, affluent Protestants settled on Willkie (Lazarsfeld, Berelson and Gaudet, 1944). More recently, Bartels has found that voter choice and assessments of the state of the national economy become more tightly bound up with each other over the course of presidential campaigns. Those who view the economy favorably are drawn to the incumbent and vice-versa (Bartels, 1997; also see Markus, 1988). In a similar analysis, Iyengar and Petrocik (1998) found that exposure to campaigns brings voters' candidate preferences into alignment with their evaluations of the incumbent President's job performance; those who approve of the incumbent's performance become increasingly likely to vote for him (or his successor) while voters with more critical views gravitate to the challenger. All told, the "campaign effects" literature demonstrates that exposure to campaigns amplifies the effects of basic political predispositions on vote choice (Gelman and King, 1993).

The structuring effects of campaigns appear to be especially prominent among voters with weak predispositions. In Ansolabehere and Iyengar's experimental studies of campaign advertising, for instance, exposure to advertisements boosted the level of party voting the most among weak partisans and those with relatively low levels of interest and information (Ansolabehere and Iyengar, 1995). Iyengar and Petrocik observed similar "levelling" effects during the 1992 and 1996 presidential campaigns. Younger voters and weak partisans were especially likely to engage in party voting as a result of exposure to the campaign. As Iyengar and Petrocik (1998, pp. 23-24) put it, "The most frequent manifestation of campaign activation is the shepherding of wayward or undecided partisans back into the fold . . . Exposure to the campaign means the most for political 'have-nots'."

The specific mediators of the activating and reinforcing effects of campaigns have yet to be identified. Trying to explain why campaigns seem to exert most influence on less involved partisans, some researchers have suggested a cognitive account by which campaigns "prime" voters to behave in accordance with their partisanship or assessments of the economy. It is well-known that more accessible criteria are weighted more heavily in choice situations and campaigns serve to make basic predispositions more accessible (Iyengar and Kinder, 1987; Simon, 1998; Iyengar, 1991). Because their party identification is already accessible, strong partisans need no additional stimulation to rely on partisanship when voting. Weak and less-engaged partisans, however, need the stimulation (i.e., activation) provided by the campaign if they are to vote in partisan fashion.

An alternative explanation of the reinforcing effects of campaigns is that exposure to the campaign triggers group identification and the resulting "in-group" bias (Tajfel 1982; Tajfel and Turner, 1986). It is well-known that people categorize themselves into groups effortlessly and that group identity, even when based on patently trivial categorization schemes (e.g. "Kappas" and "Gammas") exerts powerful attitudinal consequences including the expression of discriminatory affect for in-group and out-group members (the former are liked, the latter disliked). These effects appear to be robust and pervasive. As Sidanius (1993, p. 189) has noted, "No culture has yet been discovered within which the minimal groups effect has not been found to operate." The fact that Democrats and Republicans, those who approve and disapprove of the incumbent Administration, economic pessimists and optimists all move in opposing directions over the course of the campaign is consistent with this basic precept of social identity theory.

Our research examines the candidates' discourse as a potential mediator of the reinforcing effects of campaigns. The working hypothesis, derived from research in the discipline of linguistics, is that speech cues are fundamental to processes of identification. If language is a bond that connects candidates to voters, manipulating the candidates' discourse should affect voters' ability to identify with them, and thereby the candidates' ability to capitalize on the in-group bias. Specifically, we demonstrate that "distancing language" significantly diminishes differences between in-group and out-group partisans in their evaluations of the candidate, when those exposed to distancing language are the low-involvement partisans whose partisanship real campaigns are most likely to reinforce. By cuing citizens with distancing language instead of ordinary language, we can reverse the typical result of campaign effects research, eroding partisan support among less involved partisans while reinforcing it among more involved partisans. Because distancing language holds constant the ideological symbolism of a campaign, our result demonstrates that the standard effect of campaigning cannot be due to priming the voter's ideological or partisan loyalties, if by priming we mean reminding the voter of what he already thinks. Instead campaigning works on an affective level through the cuing of shared social identity.

The Linguistics of Political Campaigning

At least for the Democratic or the Republican candidate, the goal of a presidential campaign is, of course, winning the election, if possible with a margin sufficient to constitute a mandate for governing. Candidates win by "holding the base," building support among weak partisans and independents, and, sometimes, attracting a significant number of "crossover" voters. As we have noted, most campaigns accomplish the first of these objectives: over the course of the campaign, partisans increasingly back their party's nominee.

There are two possible explanations why campaigns reinforce partisanship. We call the first explanation "semantic priming": candidates attract partisans by providing them with information about relevant attributes -- their issue positions, their prior experience, the state of the country, and so on. In effect, candidates rely on message-based appeals. The alternative explanation we call "identity bonding": candidates generate an affective response from their partisans. Reinforcement occurs not because the campaign reminds partisans of the candidate's relevant attributes but because the campaign cues recognition of shared political identity. The information contained in the candidate's messages matters less than the candidate's ability to make himself or herself "one of us" to the partisans. While semantic priming triggers cognitive processing (acquiring information about candidate attributes and weighting the attributes according to their accessibility), identity bonding cues a more primal, affective process requiring only recognition of the candidate as a fellow partisan.

How can we distinguish semantic priming from identity bonding as mediators of the classic effect of campaigning? The discipline of linguistics offers a method for separating the information communicated by a verbal message from the relationship (between speaker and listener) cued by the same message -- the concept of "interlocutor distance." For linguists, "distance" serves as a metaphor for features of an utterance that emphasize the difference in social identity between speaker or writer and hearer or reader. Any language offers the possibility of formulating alternative messages that communicate the same semantic information but that either affirm or deny the sharing of social identity between the participants in the communication. Distancing messages are constructed by manipulating (1) the length of the utterance, (2) the choice of pronouns, and (3) negativity and conjunction (Anderson 1996).

Across cultures, distancing messages are longer than their semantic equivalents that communicate shared identity. When speakers want to communicate interpersonal distance, they use longer words and longer clauses (Haiman 1985, 151-154). The length of an utterance "iconically" diagrams social distance. For example, in languages where a mother-in-law's relatives are taboo, a speaker must not only shun them physically (by directing the gaze away from them), but must also use a special vocabulary, each word of which is longer than the semantic equivalent used when addressing non-taboo persons (Anttila and Embleton 1995, 106). In Russian, distance is signified by formulating clauses containing all required parts of speech, while interpersonal closeness is signified by "syntactic deletion," omission of otherwise necessary words, that results in shorter utterances (Yokoyama 1994). When respectful, Korean "thank you" takes a suffix of three syllables with vowel lengthening; when directed to an equal, two syllables without vowel lengthening divided by a complex consonant cluster; when directed to an intimate, two syllables with vowel reduction separated by a simple glide (English "w").

Besides lengthening the utterance, speakers communicating social distance also vary their choice of pronouns and manipulate their use of negatives. Plural pronouns or even third-person forms replace singulars, as in French vous for tu or German Sie for du (Diez 1986, Brown and Levinson 1979). Where languages have this feature, negation by the negative prefix ("in-," "un-," "dis-") displaces the negative particles ("not," "neither," etc.) (Anderson 1996).

Because these devices are so general across languages, it can be assumed for any speaker that he or she knows how to be distancing or to avoid distancing, even if the speaker could not explain how he or she does it. English is no exception to these general findings. Research on English speech in formal situations (such as academic lectures and scientific prose) has found that nouns displace verbs (Brown and Fraser 1979, 41-49). Nouns take over from verbs because nouns represent concepts and things as objects, while verbs represent them as happening (Halliday 1966, 23-24). Because a clause is always defined by one event occurring (Chafe 1994, 69), and in English this event is signified by a single verb phrase, speakers who want to communicate distance by elongating clauses can always accomplish this objective by piling up noun phrases around the verb (for the same process in German, see Slotty and Seidel-Slotty, 1961; in Russian, see Anderson, 1996). The displacement of verbs by nouns need not alter the information content of a message. In English, almost any verb can readily be converted into a noun by the process known as "nominalization." For example, the verb "convert" can readily be nominalized into the noun "conversion." If nominalization elongates the clause, an English-speaking audience applies the cross-linguistically standard rule of iconic diagrams of social distance, interpreting the elongated clauses as an indication of intent to express social distance.

Techniques for expressing distance are also visible in conflicts, such as negotiations. During conflictual exchanges, besides nouns displacing verbs, plural pronouns and third-person formulas also displace the first-person singular (Diez 1986, 229-231). For example, "I think" gives way to "it is our position that...." Negative particles decrease relative to negative prefixes because negation by particles is ambiguous (Horn 1985, 140). For example, consider the choice between "not honest" and "dishonest." "Not honest" is more ambiguous, because it leaves open the possibility of "not dishonest," which "dishonest" forecloses. Iconically, negation diagrams its own ambiguity by forcing the audience to choose arbitrarily among two standards comparison. The audience can either compare (a) the shorter "honest" against the longer "dishonest" or (b) the pair of shorter words "not honest" against the single longer word "dishonest." In case (a), "honest" is clearly shorter than "dishonest," and social distance is less. In case (b), it is unclear whether a pair of shorter words is longer than a single longer word. Leaving open a middle ground that the prefix eliminates by forcing the audience to process an utterance at least as long as the alternative, the negative particle is iconically less distancing (Brown and Levinson 1978, 263-264). Finally, the English conjunction "and" is a discourse marker shaping the interaction between participants in a communication by expressing the speaker's intention to continue speaking, while English "but" is a semantic marker expressing contrast between a preceding and a following clause (Schiffrin 1987, 184-185). As "and" proliferates relative to "but," English utterances lengthen, and audiences hear distance.

According to the theory of interlocutor distance, English speaking candidates trying to express social identity with voters will restrict the number of nouns relative to verbs, prefer the first person to the third or the impersonal, express negativity with particles rather than prefixes, and avoid an excess of "and." If so, by reversing these tendencies, an experimenter can produce distancing messages that communicate the same information while disrupting the audience's experience of identification with the speaker. In short, linguistic theory enables us to test whether the reinforcement of partisanship so typically observed in the study of campaigning is attributable to semantic priming or identity bonding.

Hypotheses

Semantic priming can be tested against identity bonding by producing campaign messages that hold constant the semantic information while varying the audience's experience of identity with the speaker. None of the textual changes associated with distancing in English need alter the semantic information communicated by a campaign speech. Conversion of a verb into a noun need not affect semantic meaning.. Substitution of a third-person impersonal form for a first-person pronoun does not change the reference to the speaker. Substitution of negative prefixes for negative particles merely reduces ambiguity, while elimination of "but" merely removes one signal to a contrast that continues to be communicated by the meaning of the clauses joined by this conjunction. A distancing message can be semantically identical to a message in ordinary language.

In conformity with the classic finding that campaigns reinforce partisanship, we hypothesize about the effect of ordinary and distancing campaign messages on partisan loyalties. We assume that when campaigns reinforce partisanship, exposure to campaigns causes partisans to intensify their preference for their own party's candidate relative to the opposing candidate. We measured support for 1996 presidential candidates Clinton and Dole using a standard 100-point "feeling thermometer," with respondents asked to score the candidates, among twelve other personalities used as fillers, from 0 to 49 depending on how unfavorably the respondent evaluated the candidates, 50 if the evaluation was neutral, or from 51 to 100 depending on how favorably the respondent evaluated the candidate. We then constructed a scale from the absolute value of the difference between the respondent's evaluations of Clinton and Dole. This scale takes the value of 0 if the respondent is indifferent and 100 if the respondent maximally favors one candidate and maximally disfavors the other. We assume that this difference increases if exposure to a textual cue reinforces the respondent's preference for a favored candidate and decreases if exposure to a textual cue erodes the respondent's preference. Thus the scale measures intensity of support for the voter's preferred candidate.

Because the shift from ordinary to distancing language does not change the information communicated by a text, semantic priming and identity bonding imply contrasting predictions about how exposure to texts will affect the intensity of support for rival candidates among partisans and among independents. Both semantic priming and identity bonding predict strong effects only among inattentive partisans, in conformity with the standard finding of campaign research that campaigns fail to influence attentive partisans, whose minds are already made up in favor of their preferred candidate. We measured attentiveness by a scale composed of the answers to four questions: how often does the respondent read the newspaper, watch the national news on television, discuss politics with others, and follow political events. We divided the sample at the score closest to the median, with 45 percent of the respondents falling below and 55 percent falling above the division point. Persons below the division point were classified as inattentive, while those above were classified as attentive.

Semantic priming implies that an inattentive partisan exposed to information about either candidate's issue positions will react by intensifying support for his or her preferred candidate. Thus semantic priming implies that both ordinary and distancing language will reinforce partisanship relative to a control group exposed to no campaign message. Reinforcement by semantic priming may be attenuated if the distancing text is more difficult to construe, but it should be present in both cases, but in no case can the semantic priming hypothesis imply that a distancing text could diminish support for the preferred candidate below the level expressed by those respondents exposed to no textual cue at all. Even if information is hard to understand, providing more of it cannot be less reinforcing than providing none, if information is what causes reinforcement among inattentive partisans.

Identity bonding implies that exposure to ordinary language, which communicates shared social identity as speakers of common English, will intensify an inattentive partisan's support for the candidate of the same party, but exposure to distancing speech will reduce an inattentive partisans support by disrupting the experience of shared identity. While ordinary language will intensify partisan support relative to distancing language, in an experiment conducted at the end of a long and salient presidential campaign, ordinary language may fail to produce noticeable effects when the comparison set is a control group exposed to no message in the experiment but previously exposed to many campaign messages outside the experiment. Like those of semantic processing, the predictions of identity bonding apply only to inattentive partisans.

For partisans, either hypothesis predicts an interaction between attentiveness and exposure, with the effects found mainly among the inattentive partisans, but the hypotheses predict that the interaction will be found in different cells of the 3X2 design. The semantic priming hypothesis predicts that the interaction will take place between the two exposure groups (or, possibly, only the ordinary language group and the control group), while the identity hypothesis predicts an interaction between attentiveness and distance, as inattentive respondents in the distancing condition move in an opposite direction from those in the ordinary language condition, with possibly neither differing noticeably from the control group.

. For independents, the predictions of the two hypotheses may or may not differ. Semantic priming may imply that exposure will convert independents into supporters of one or the other candidate, or it may imply that exposure will leave them indifferent. Identity bonding implies that independents will recognize the distinctiveness of their own political identity from that of either candidate, with the result that both distancing and ordinary speech will reinforce their indifference. Because, moreover, independents include both indifferent centrists and committed partisans of minor parties located either to the right or to the left of the Democrats and Republicans, we cannot form strong expectations about the effects of ordinary or distancing language on their evaluations of the candidates, except that the identity bonding hypothesis predicts that exposure to ordinary language will reinforce the independents' identity as moderates: i.e., their score should not change. Thus the identity bonding hypothesis would be rejected if the independents sharply increased their support for either candidate in response to exposure to ordinary language.

In sum the hypotheses are: (1) if semantic priming mediates partisan reinforcement, exposure to either ordinary or distancing language should increase support for the party's candidate relative to no exposure among inattentive partisans but not among attentive partisans; (2) if identity bonding mediates partisan reinforcement, exposure to ordinary language should increase support for the party's candidate relative to exposure to distancing language among inattentive but not among attentive partisans; (3) if identity bonding mediates partisan reinforcement, independents should not respond to the difference between ordinary and distancing language.

Experimental Procedure

In order to test the hypothesis of semantic priming versus the hypothesis of cuing social identification, one author constructed artificial texts by applying the principles of distancing speech to authentic speeches actually uttered during the 1996 campaign by Bill Clinton and Bob Dole. An experiment using textual cues that nobody actually sees during a campaign has a rationale. The normal experiment on campaign effects contrasts subjects exposed to an actual or simulated campaign message (such an advertisement) with a control group that sees no cue. Such an experiment faces the difficulty that the people in the control group, shown no cue inside the experiment, have nonetheless encountered many campaign messages outside the experiment, either in their direct experience or in experienced mediated through news outlets or through conversations with other persons. An experiment that contrasts exposure to an authentic cue with exposure to no cue merely tests the marginal effect of a single additional exposure. For this reason such experiments ordinarily produce weak or null findings. We expected that the effects of exposure to one authentic campaign speech would be weak or null, and this expectation proved correct. However, exposure to a radically different kind of speech should be expected to produce more definitive effects.

We transformed two speeches by each candidate into distancing versions. The speeches were selected to address common themes: a radio speech during the campaign on crime by Clinton and a campaign speech on crime by Dole, and a campaign speech by Clinton on economic and tax policy and a campaign speech by Dole presenting his economic and tax programs. The texts were taken from the White House web page and from Dole's official campaign page.

Each text was rewritten to nominalize all verbs except forms of "to be" and "to have." These verbs, which are semantically almost empty; are minimally necessary to construct an English clause. All first-person singular pronouns were transposed into either the first-person plural forms or into third-person impersonal forms that did not substitute a new referent (e.g., "one is," "it is," but not "he, she is" or "they are"). All negative particles were transposed into negative prefixes attached to the head word negated by the particle in the original. "But" was deleted throughout.

Needless to say these transpositions produced extraordinarily awkward prose. For example, sentence (1), used by Dole to conclude both speeches, became sentence (2):

(1) Give us your votes and your support.

(2) This is an invitation for the gift of your votes and your support.

Before the transpositions were performed, each authentic text was condensed by omitting some sections in order to confine its length within a limit of three printed pages. Each of the distancing versions is, inevitably, six hundred to twelve hundred computer characters longer than the corresponding authentic text, or about fifteen to twenty percent.

The texts were headed with either "Edited Remarks by Senator Bob Dole" or "Edited Remarks by President Bill Clinton," laser-printed on ordinary white paper, and photocopied using a quality copier. Undergraduates were recruited who volunteered (with the intent of obtaining course credit) to find subjects in shopping malls. One of the authors also found subjects in a law firm where his wife is employed. The malls chosen by the undergraduate interviewers were located near their homes in and around Los Angeles. Undergraduate students of the authors' university were ineligible as subjects. Interviewers were instructed not to respond to questions or comments on the texts while the subjects were filling out the questionnaire. The texts were randomly shuffled before being given to the interviewers. Subjects were offered a payment of $5.00 to participate in a study of political attitudes, and their anonymity was guaranteed.

Each respondent in an experimental condition read either the authentic text or the distancing version of one speech. They filled out a questionnaire which began with demographic questions, questions about their attentiveness to politics, and questions about their voting intentions and party identification. The questionnaire then turned to whether they found the experience of reading the text interesting and enjoyable, to the political efficacy battery, to their thermometer ratings of the candidates (mentioned among a large number of filter persons), and finally to their evaluations of each candidate's competence on six substantive issues, including three discussed in the speeches and three inserted as filters. A large control group filled out the same questionnaire without reading any text, with the exception that the control group did not answer the questions about whether the texts were interesting and enjoyable to read.

The experiment was administered during the last two weeks of the 1996 campaign, ending the weekend before the election.

Results

Our results were more interesting than we expected. Table 1 displays partisan respondents' mean support for Bob Dole or for Bill Clinton, as measured by the absolute value of the difference in their thermometer ratings for the two candidates. (Nonpartisans are excluded.) The columns display the mean scores for inattentive and attentive partisans, while the rows display the experimental conditions of no exposure, exposure to ordinary language, or exposure to distancing language.
 
 

Table 1: Strength of Partisans' Support for Own Candidate, By Attentiveness and Type of Speech
 

Experimental Condition	Inattentive	Attentive
No Exposure	35.3	40.6
	(24.1)	(31.2)
	n=31	n=58
Ordinary Language	45.7	38.2
	(25.3)	(32.3)
	n=31	n=39
Distancing Language	25.6	51.7
	(24.1)	(31.9)
	n=27	n=43

 

In Table 1 the semantic priming hypothesis fails entirely. Among inattentive partisans, exposure to an additional campaign speech reinforces their preference for their own candidate only when they are exposed to ordinary language. Among these partisans, even at the very end of the campaign, exposure to a single message in ordinary language increases their support for their candidate by ten points (t=1.371, one-tailed p < .086). But exposure to the identical information in distancing language not only does not increase their support for their candidate: instead exposure to distancing language reduces their support by almost the same ten points. Exposure to distancing language makes their evaluations of candidates indistinguishable from evaluations by independents exposed to ordinary language or to no campaigning at all (see Table 3 below).

While the semantic priming hypothesis fails, the identity bonding hypothesis receives strong support. As expected, a single exposure to ordinary language exerts a marginal effect relative to no exposure within the experiment among people who, at the end of the campaign, have already encountered many campaign messages, but when exposure to ordinary language that confirms sharing of identity is compared to exposure to distancing language that disrupts it, a twenty-point gap emerges between the affect expressed by inattentive partisans in the ordinary and in the distancing condition. This gap is not marginal at all (t=2.57, one-tailed p<.006).
 
 

Table 2: Impact of Ordinary and Distancing Speech on Partisans' Candidate Evaluations: ANOVA Results
 

Variable	F	p
Distance	0.36	0.699
Attentiveness	3.84	0.051
Distance*Attentiveness	5.47	0.004

n =230
 
 

While we predicted a large gap in the affect expressed in the two experimental conditions by the inattentive partisans, we were surprised by the impact of distanting language on attentive partisans. Table 2 presents the results of a 3 X 2 analysis of variance (corresponding to the three levels of the distancing manipulation and the two levels of attentiveness) of the partisans' net thermometer scores. As only the identity bonding hypothesis predicts, attentiveness interacts with distance. In conformity with the typical finding of campaign research, our experiment finds that exposure to campaigning in ordinary language reinforces inattentive partisans while failing to influence attentive partisans, who are already committed to one candidate. Unexpectedly, however, distancing language also reverses the typical finding of campaign research about attentive partisans. Disrupting identification among the inattentive, distancing language produces among attentive partisans a powerful, thirteen-point reinforcement of preference for the favored candidate.

We do not know precisely why our distancing texts unexpectedly reinforced attentive partisans. The literature on distancing language does not supply much guidance, for it does not specify the exact cognitive mechanism by which distancing language disrupts social identification. One possible explanation derives from "depth of processing" psychology. It has been pointed out (Givon 1995) that ordinary language permits maximally efficient cognitive processing by locating related concepts together in space or time. If ordinary language is maximally efficient, distancing language would compel the audience to do additional processing in order to construct an equally elaborate cognitive model of the text (van Oostendorp 1994). It may be that this excessive cognitive burden alienates an inattentive audience from a distancing text. Inattentive and attentive partisans differ in their willingness to do the work of processing a political text. In general, when people process a stimulus more deeply, they experience more strongly whatever affect the stimulus invokes. Thus if attentive partisans encountering distancing language process it more deeply, they strengthen their identification with their preferred candidate, while inattentive partisans, predisposed anyway to process political cues superficially, may be further repelled by the extra effort necessary to process distancing language. Then they would experience their already weak affect with reduced intensity.

The identity bonding hypothesis would be refuted if ordinary language motivated independents to identify with either candidate. As shown in Table 3, this refutation does not occur in our experiment. When exposure to ordinary language is compared with no exposure, independents' evaluations are unaffected. Exposure to ordinary language actually reduces the difference in the independents' evaluations of the two candidates, but the result does not verge on statistical significance. There is a slight tendency for independents to respond to distancing speech in much the same way as attentive partisans do. When contrasted with exposure to ordinary language, exposure to distancing speech modestly increases independents' support for one candidate (t=1.394, p<.165). This tendency is especially pronounced among the attentive independents who were exposed to distancing speech. While the numbers are far too small for reliable analysis, it is plausible that the attentive independents' reactions are the opposite of attentive partisans' reactions. While attentive partisans who read distancing language may reconcile the contradiction between frustration with the text and support for the candidate by increasing their support for their preferred candidate, independents have no preferred candidate. The (very few) attentive independents in our sample respond to distancing texts by increasing their support for the candidate opposing the one whose revised speech they read. Attentive partisans expressing an intent to vote for either candidate, by contrast, move massively in favor of their preferred candidate regardless of whose speech they see when reading the distancing version of the text.
 
 

Table 3: Evaluations of Bob Dole and Bill Clinton by Experimental Condition: All Partisans versus All Independents
 
 
 

Experimental Condition	Partisans	Independents
No Exposure	38.8	28.4
	(30.6)	(28.2)
	n=89	n=34
Ordinary Language	41.5	23.4
	(29.5)	(25.0)
	n=70	n=22
Distancing Language	41.6	34.6
	(31.6)	(26.0)
	n=70	n=33

 

Discussion

In sum, the results favor identity bonding over semantic priming as the likely mediator of the classic reinforcement effect. In fact, we failed to detect any traces of semantic priming. Contrary to the hypothesis of semantic priming, inattentive partisans given information about the candidates' positions on salient issues were no more enthusiastic about their candidate than others given no such information. Consistent with the identity bonding hypothesis, inattentive partisans exposed to the distancing version of campaign speech were less polarized than their counterparts exposed to the ordinary language version of campaign speech. In addition, as further predicted by the identity bonding hypothesis, independents were generally unaffected by the distancing manipulation.

Our identity bonding hypothesis supplies a close fit to the attitudes of inattentive partisans, but it requires further elaboration if we are to understand why distancing language so strongly reinforces the preferences of the attentive partisans. We have proposed one possible mechanism by which this reinforcement may occur, suggesting that because a distancing text demands deeper processing to form an equally elaborate comprehension of the text, those who are willing to do the additional cognitive work find themselves more strongly affected. But there are other possibilities, and we believe that this result, as intriguing as it was unexpected, deserves further consideration.

More generally, the evidence suggests that the ties that bind partisan voters to candidates during campaigns are woven more by the dynamics of intergroup relations and the tendency of group members to prefer one of their own than by the transmission of information about the candidates' positions on policy issues. A supposition that group identity is primary certainly helps to resolve a paradox in the conventional wisdom concerning the role of information in voter choice. There is ample evidence that the typical voter is information poor but attitude rich (see, for example, Kuklinski and Ferejohn, 1991). For the relatively uninformed electorate, partisanship is the meaningful signal and the candidates' ability to evoke shared partisan identity is crucial to "holding the base." This result should not be taken to mean that the semantic or informational content of campaign speech is irrelevant to electoral support; far from it. Texts with different informational meanings may be more or less evocative of partisan identity. For instance, candidates who campaign on "owned" issues more effectively attract the support of their fellow partisans (Petrocik, 1996; Ansolabehere and Iyengar, 1995; Iyengar and Valentino, 1998). Our results suggest that issue ownership is good strategy because it more strongly conveys a candidate's group-relatedness. Future work will need to tease out the interactions between linguistic distance and semantic content in political speech.

Lastly, this research draws attention to the potential of linguistics as a discipline capable of generating novel hypotheses and techniques for the study of politics. Campaigning is mainly semiotic activity, in which linguistic signs are prominent. A widespread view of campaign advertising that only visuals matter ignores the background of linguistic activity that alone enables viewers to interpret visuals of candidates cursorily labeled with linguistic items such as "pro-choice" or "pro-life." These brief labels have developed into condensation symbols, capable of cuing whole complexes of linked attitudes in viewers, only because the viewers have previously participated in extended reading and conversation informed by public utterances on or on behalf of prominent politicians, often including the candidates portrayed by the visuals. In the absence of these extensive prior linguistic interactions, visuals labeled in this manner might not exert any effect at all. Viewers would not know whom the visuals showed nor what the labels meant. Neither "pro-choice" nor "pro-life" refers to any particular attitude toward state policy on abortions except as the result of extensive linguistic exchanges -- as is evident in the circumstance that before the last two decades, neither "pro-choice" nor "pro-life" even existed as stable lexical items. Ansolabehere and Iyengar (1996, 102-105) have shown that one can present identical visuals, switch the labels, and produce noticeable effects on viewers.

The discipline of linguistics is rich in concepts useful for studying campaigns and other political phenomena, particularly concepts drawn from the new subfield called "pragmatics," the study of language in use. Metaphor, treated as a pragmatic concept, has been used to study what Petrocik (1996) would call "issue ownership" by the two major parties (Lakoff 1996) and to distinguish political communications that are persuasive from those that are memorable (De Landtsheer and De Vrij 1988). Anderson (1998) has contrasted authoritarian and electoral politicians' use of modal verbs (those expressing possibility, obligation or necessity, e.g., "may," "ought," "must"), as the authoritarians try to depress and the electoral politicians try to encourage partisan identification by either diverting or attracting attention to themselves as reasoning agents who either do not or do construct a relationship with the audience. These and other concepts from the discipline of linguistics have received too little attention in the conventional study of campaigns.

Table 4 shows the distribution of respondents by educational attainment and race, while Table 5 shows the distributions by sex and mean age (where age calculated by subtracting the year of birth from 1998). While we make no pretense that the sample is randomly drawn from any population, the respondents are reasonably representative of the Los Angeles population. Citizens of Hispanic ethnicity are rather underrepresented for Los Angeles, while well educated whites are overrepresented.

Table 4: Distribution of Respondents by Race and Education
 
 
 

	some high school	high school graduate	some college	college graduate	post-graduate	Total
Asian	0	0	12	15	5	32
African-American	0	3	18	13	4	38
Hispanic	0	6	17	7	1	31
White	6	13	61	79	42	201
Other	2	3	4	7	1	17
Total	8	25	112	121	53	319

Missing data: 1
 
 

Table 4: Distribution of Respondents by Age and Sex
 
 
 

	Mean Age	Std. Dev	n
Males	36.6	12.5	156
Females	37.2	12.0	161
Total	36.9	12.2	317

Missing Data: 3