Life in MiniatureDon't Sweat the Small Stuff!

Looking For Walthers Union ? Then This is Where You Want to Be

Variety of Language

Introduction:

The growth of the Internet has fostered hopes that it will promote active democracies around the world. These hopes are based on the multiple sources of information available to users on the Internet, which is believed to lead to better political choices, and the freedom to express one's views online, regardless of social status differences. However, the Internet also hosts various forms of deviant behavior, such as flaming of other discussants, malicious spreading of falsehoods, and hacking into private networks. These harmful acts may instead inhibit widespread participation and deliberation.

These conflicting situations arguably develop from a key structural characteristic of the Internet—the perceived anonymity of online users. Scott (1998) defined anonymity as "the degree to which a communicator perceives the message source is unknown and unspecified" (p. 387). Speakers in computer-mediated settings have the potential to adopt pseudonymous identities behind text-based personas. People can feel freer to share diverse ideas that they otherwise may suppress (Postmes, Spears, & Lea, 1998). However, anonymity may create difficulties in evaluating messages produced by unknown sources and may undermine credibility, at least in terms of the subjective experience of those reading the messages (Donath, 1999; Flanagin & Metzger, 2007). Thus, an investigation into the impact of user identity cues may indicate whether the very characteristic that makes the Internet a promising space for political discussion also works against its own potential.

An examination of Internet fora, or discussion boards, reveals that their different architectures provide varying degrees of anonymity by attaching users' information, such as demographic data or number of posts submitted to date, together with the actual discussion board message. These user details can be a type of status cue, providing background information about the source (Donath, 1999). The message text may also contain other identifying information. In the cue-lean world of text-based communication, the language that people use online becomes an important source of background information, possibly revealing their social position, gender, age, social class, and so forth (Gupta, 1998a). In the context of Singapore, the use of the colloquial variety of English, known as Singlish, may function as a heuristic cue that could affect the kinds of judgments that people make about speakers and their opinions. Our study examines the impact of such status cues in text-based computer-mediated communication (CMC) with respect to their influence on the perceptions of online discussants and the effects these cues may have in encouraging active participation in online political discussions.

Literature Review

Status Cues and Heuristic Processing in CMC

In a text-based CMC context with few identifying markers, the structural features of the medium may convey latent information compensating for the lack of nonverbal and visual cues, enabling heuristic judgments to be made about discussants. Two perspectives dominate the literature in this area: the cues-filtered-out perspective (CFO) and what might be called the available-cues-matter perspective (ACM). Early research supporting the first perspective (CFO) states that CMC is lacking in social presence, media richness, and social context cues. The social presence theory (Short, Williams & Christie, 1976) is defined as the feeling that interlocutors are sharing the same communicative space. Since CMC is primarily text-based, this theory states that a CMC user would have reduced perception of an audience, as compared to someone engaging in a face-to-face (FtF) interaction. Similarly, the media richness theory (Daft & Lengel, 1984) claims that the reduction of cues available in CMC makes it less well-suited for certain tasks involving complex negotiation of meaning, as compared to the FtF communication channel. Last, CMC is seen to lack regulating feedback, status and non-verbal cues, personalization, and social norms, resulting in disinhibition and greater equality of participation (Kiesler, Siegel, & McGuire, 1984). The above theories suggest that without the cues present in normal FtF interactions as a guide, message processing and interpretation are affected, reducing the efficacy of interpersonal communication.

Psychologically, inferences and evaluations drawn from messages are influenced by preconfigured schemas. These mental structures allow communicators to make sense of their world through organizing perceptual information into neat categories (Fiske & Taylor, 1991). When a situation involving patterns similar to past events occurs, existing schemas are activated. Consequently, there is a higher tendency for a person to evaluate a message or its speaker heuristically based on past events, instead of using high-level cognitive processing to make attributions about the current subject of consideration.

While the use of heuristics makes cognitive processing more efficient, the results may not always be accurate. Some negative consequences of using heuristics have been noted in stereotype studies, where incomplete or inaccurate assumptions are made about sources (Ross, Read, & Toglia, as cited in Bornstein, 1995). Analogously, negative judgments about arguments based on incorrect attributions about speakers will have detrimental effects on the quality of decision-making in online political deliberation. Johnson and Johnson (2000) note that political deliberation ideally helps citizens reach their best-reasoned judgment concerning which alternative will solve the problem at hand. For citizens to reach such judgments, it is important that their evaluations of related arguments are not distorted by irrational biases formed through heuristic evaluations.

The second perspective is available-cues-matter (ACM). Despite some evidence supporting the CFO perspective, it was challenged by some early Internet researchers (e.g., Rafaeli, 1986; Rice & Love, 1987), and more recent research has also argued against its theoretical claims. For example, the Social Information Processing (SIP) theory contests the deterministic CFO view by positing that despite the cue-lean nature of CMC, there are other social markers about a person available, which can be made salient, making the CMC environment in no way inferior to FtF situations in facilitating the development of relationships (Walther, 1992). The crucial mitigating factor is time, since it typically takes longer for text-based communication to evolve. In other words, given adequate time, users can manage relationships within the text-based constraints of CMC, developing substitutes to compensate for the deficiency of traditional non-verbal and paraverbal cues afforded by FtF exchanges. Walther (1996) also proposes the idea of hyperpersonal communication, in which "CMC has surpassed the level of affection and emotion of parallel FtF interaction" (p. 17).

Meanwhile, the Social Identity and Deindividuation Effects (SIDE) theory specifically challenges the unilateral "individuating" outcome proposed by the CFO perspective (Kiesler et al., 1984). If people have a range of identities which become salient depending on the social context and level of identifiability, the SIDE model argues that conformity to group norms will result if one's social identity is activated. However, if one's personal identity becomes salient, individuation will occur instead (Spears & Lea, 1994). Like the SIP theory, the SIDE model explores the influence of cues that are available in CMC and that affect the outcomes of online interactions. But while it examines the effect of online anonymity, it does not explicitly address which structural components of CMC determine the richness of a virtual social environment and is thus unable to either dispute or confirm if the medium is contextually barren.

The claims made by the earliy CFO theories and later the ACM perspectives of the SIP and SIDE models in the 1990s represent a historical progression, in keeping with the development of CMC technology and greater user experience. More recent theoretical developments highlight the importance of examining how source information may function as a heuristic cue in online communication. Mitra (as cited in Sparks, 2001) notes that "the anonymity of the IP address serves to disguise many of those social markers (age, gender, ethnic origin, accent, and so on) that in practice serve to either validate or disqualify the opinions of speakers in direct social interaction" (p. 73). These kinds of source cues may provide background information about the speaker's identity and enable heuristic judgments to be made about the speaker and the message.

Language and Linguistic Style

Since the majority of CMC occurs in a text-based format, it is reasonable to infer that language variables would become an important source of heuristic cues, independent of content or topic of discussion. Some of the language variables that have been examined in the face-to-face context include pragmatic and syntactic codes and standard discourse schemas (Liu & Ginther, 2001). Smith, Siltanen and Hosman (1998) have looked at how powerful and powerless language styles affect evaluations of a speaker's authoritativeness, sociability, and similarity to the receiver, while Adkins and Brashers (1995) examined the impact of such language styles on attractiveness, credibility, and persuasiveness in CMC. Some studies of perceptions of CMC language variables have focused on politeness/impoliteness or grammar use (Jessmer & Anderson, 2001). These studies suggest that variations in language styles influence the audience's perception of the speaker or writer. It follows that participants may use language as a heuristic cue to evaluate informativeness, persuasiveness, and credibility in online discussions.

In the multilingual Singapore context, the Singapore Government has promoted the use and learning of English alongside native mother tongues since the late 1960s, both as a tool for economic survival and as the "ethnically neutral" language in a country where racial relations require careful management (Ho & Alsagoff, 1998). Despite its acknowledged importance, English, as the language of the Western world, is not fully accepted as a natural native language. It has been seen as "an objectionable representation of the national character" (Llamzon, 1977, p. 41).

If Singapore English is to be "regarded as a language which is truly Singaporean" rather than a foreign tongue, Llamzon (1977) argued that it was necessary for it to be indigenized. Colloquial Singapore English (CSE) and Singapore Standard English (SSE) developed as two possible indigenous varieties. CSE, more commonly referred to as Singlish, is primarily a contact language,1 with a grammatical structure significantly different from SSE. Its extreme forms feature a truncated sentence structure, with frequent instances of dropped articles, subject deletion, use of pragmatic particles like 'hor' and 'lah', and zero copula.2 SSE, on the other hand, is a culturally-located variety of English (Gupta, 2001), with Standard English grammar plus a few local words to express local concepts.3

Borrowing from the diglossia model of language (Ferguson, 1959), which describes the situation where two varieties of a language are used concurrently within a community, we may consider SSE to be the high (H) variety. The H is defined as a more standardized form of the language, which is most likely not natively spoken but may be learned in addition to the native variety. Singlish would be the low (L) variety, where the L generally refers to a regionalized dialect of the language. For this study, the terms Standard English and Singapore Standard English are used interchangeably to refer to a language style distinguished by the use of standard grammar, with a few local words or concepts added to its lexicon.

The use of a colloquial or a standard style could lead others to form different kinds of impressions about the speaker. Gupta (1998a) argued that for Singapore English, similar to most languages, the speakers' style conveys information to others within the community about their social position, and speakers provide such information either consciously or semi-consciously. Empirical support linking language style and social position has been found by Kuo (1977), who noted that proficiency in and household use of English in Singapore were positively correlated with economic prospects, social mobility, educational level, and household income.

Singapore government discourse seems to support the belief that the use of Singlish communicates social inferiority, with users often perceived as being less competent and professional in the global marketplace. Gupta (1998b) cites a talk by Singapore's Minister Mentor Lee Kuan Yew in August 1994, in which he pointed out that the use of Singlish confused native English speakers and connoted a negative image of Singapore. However, Lee's belief that using Singlish would cause speakers to appear somewhat unappealing is not entirely unfounded. In Ferguson's (1959) description of diglossia, he points out that speakers tend to regard the H variety of language (in this case, Standard English) as superior to the L (Singlish); that H is "more beautiful (and) logical" (p. 237). As such, we would expect to see the following effects when colloquial and standard language styles are used:

 

H1:

Discussants who use Singlish will be perceived as being a) less informative and b) less persuasive, and they will have c) lower source credibility than discussants who use Standard English.

Source Expertise

Source expertise is another status cue that may be used to make heuristic judgments about a speaker. Several studies examined its effects on persuasion (Eastin, 2001; Petty & Cacioppo, 1986; Smith et al., 1998). These studies manipulated levels of expertise through titular changes conferred to the same source, demonstrating that people use source expertise heuristically to make evaluations about the message content, and that attitude change is greater when messages are attributed to an expert source. However, the concept of status cues in the discussion fora arena remains a relatively new area of study, because its dimensions are not always clearly delineated. It has been used as a component within the multi-dimensional construct of credibility in previous research (see McCroskey, 1966; Sternthal, Phillips, & Dholakia, 1978) and also as a predictor of credibility in other studies on persuasion (Eastin, 2001; Smith et al., 1998). Here, we used Hoffman's (1998) conceptualization of expertise, where the accumulation of skill through experience in the field is important in differentiating an expert from a novice. Hoffman proposed that experience gives a person practice in the particular domain of expertise, and "with practice, a skill loses the quality of being conscious, effortful, deliberate, and linear, and takes on the quality of automatic pattern recognition" (p. 84).

Applying this definition to the online discussion forum context, we can identify several characteristics that may provide an indication of experience or expertise. A brief sampling of Singaporean and foreign web-based discussion fora on current affairs revealed that a fair number make use of a similar architecture, in which identifying information about the forum contributor is presented in a sidebar next to the message text. The information that may indicate experience includes: (a) the number of posts made by a discussant, (b) the date that the discussant joined the forum, (c) a designative label (e.g., Expert or Novice), and (d) a star rating (on a scale of one to five). The number of posts indicates the level of participation of the discussant in the particular forum, while the date joined specifies the duration of membership in the discussion board. The hierarchical label demarcates the relative position of each discussant within that forum. The star rating is used in many review sites on the Internet that rate the quality of products, persons, or experiences. By contrast, this rating method when used in online forums corresponds to the number of posts by a discussant (see, e.g., the Liberty News Forum, http://www.libertynewsforum.com/, and sgForums.com, http://politics.sgforums.com/). The star ratings might be used heuristically to indicate both the quality of a discussant's postings and the level of experience in that forum. The prevalent usage of some kind of experience/expertise indicator in a number of fora raises the possibility that people might make use of these cues to form perceptions of the discussant, which could consequently affect their evaluation of message content.

Hoffman (1998) further points out that experts are different from novices, in that they possess "an articulated, conceptual, and principled understanding" (p. 83) of the area of expertise. An expert knowledge structure is also different from a novice's, in that it clearly organizes the information related to the domain and is so extensive that abstraction and generalization are possible on a conceptual level. In line with this characterization, it is logical to suggest that people tend to invoke heuristics established through past experience with experts when presented with the above-mentioned indicators on discussion boards. Therefore, discussants who are labeled Experts would be perceived to be more informative than discussants who are labeled Novices, regardless of the knowledge presented.

Similarly, expertise may be linked to persuasiveness. For example, Smith et al. (1998) found that expertise significantly predicted changes in attitude, with high expertise producing a stronger effect and thus being more persuasive. Expertise might also affect perceptions of online discussants' credibility. Slater and Rouner (1996) consider it self-evident that messages that come from expert and objective sources are seen to be more credible than those from inexpert and biased sources. Eastin's (2001) study was one of the first to examine the effects of different levels of source expertise on credibility in the domain of CMC. He found that high expertise sources are perceived to be significantly more credible than low expertise sources. Based on the research cited above, we expect the following:

 

H2:

Discussants who have status cues indicating they are Novices will be perceived as being a) less informative and b) less persuasive, and as having c) lower source credibility than discussants who have status cues indicating they are Experts.

Status Cues and Participation in Online Discussions

For the Internet to function well as a new public sphere, it should allow for widespread participation in the political process (Habermas, 1989). If structural variations in cues online cause related variations in willingness to participate, these must be taken into consideration when evaluating the potential of the Internet as a public sphere. CMC studies in organizational and social contexts have shown that lack of status cues can result in greater equality of participation online (Sproull & Kiesler, 1986). Although he disagrees with the CFO perspective on the issues of cue salience, Walther (1992) also found that initial interactions online are characterized by high equality and low relational dominance. Thus, regardless of which perspective holds, it would seem that the Internet is a platform conducive to encouraging equal participation in political debate, even if this equality is transitory. These perspectives do not, however, predict the effect of different status levels on participation online.

The literature on Singlish would seem to suggest that the informal climate created by the presence of a colloquial language cue would encourage participation. Bokhorst-Heng (1998) highlights a Business Times report which comments on how Singlish is "becoming trendy among young professionals as a familiarity marker and an act of identity" (p. 304). Studies show that Singlish is also used in informal online discussions (Fang, 2007; Gupta, 2006). Variations in particular language styles might thus foster affinity or a sense of belonging between lurkers and discussants, encouraging the lurkers to come forward and share their perspectives. We therefore posit the following hypothesis:

 

H3:

When discussants use Singlish, participants will be more willing to participate in the discussion than when discussants use Standard English.

Concerning the impact of different levels of expertise on willingness to participate, there is little research to suggest how these two variables might be related. While it can be expressed as a research question, we tentatively proffer a directional hypothesis, based on the earlier hypotheses made about the influence of the expertise variable. To the extent that a low expert source is perceived as less informative, persuasive, and credible than a high expert source, we feel that it is likely that people would be less willing to participate in online discussions with novices than with experts.

 

H4:

When discussants have status cues indicating they are Novices, people will be less willing to participate in the discussion than when discussants have status cues indicating they are Experts.

Beyond willingness, it is important to understand the reasons why people would participate, in order to gain a fuller understanding of the dynamics of online political discussions. Research by Stromer-Galley (2003) suggested two possible competing motivations for participating in political conversations online. A desire for homophily may motivate people to join discussions and interact with others who have similar interests, as a form of reinforcement of one's views. On the other hand, a desire for diversity might lead people to seek out discussants unlike themselves, to encounter different opinions and therefore expand their worldview. Stromer-Galley suggested that both perspectives have potentially unfavorable consequences for political talk online. Exposure to similar perspectives may lead to group polarization or the radicalization of one's initial position, while a coming together of diverse perspectives may simply result in an irrational jumble of opinions. On a more optimistic note, if diversity is the motivation behind political participation, it implies that political discussion participants are being exposed to a wider range of opinions, possibly aiding deliberative conversation. Thus, we investigate how the desires for homophily and diversity might vary in relation to status cues in political discussions online. We also examine change in desires for homophily and diversity over time, regardless of the discussion topic.

 

RQ1:

How do language and expertise affect the desires for homophily and diversity in online political discussions?

Method

Participants

Eighty undergraduates (38 women, 42 men), with a mean age of 21.5, were recruited through email advertisements. They were offered an incentive of $10 for completing the study. Non-Singaporean respondents were screened from the sample to avoid possible confounds brought about by their unfamiliarity with local political issues and discussions.

Design

The experiment had a mixed design, with two between-subjects factors (language and expertise), and one within-subject factor (discussion issue), which were fully crossed. This yielded a 2 (Standard English versus Singlish) x 2 (high versus low expertise cues) x 2 (Unions versus HDB issue) factorial design. Participants were randomly assigned to one of the between-subjects condition, with gender balanced across conditions. The order of presentation for the issues to the participants was also balanced.

Stimulus

Participants were presented with two political discussion message threads extracted from existing online Singaporean fora. One topic concerned a recent dispute between the management and the pilots' union of Singapore Airlines. The other topic dealt with a recent change in the Singapore Government's public housing policy, requiring buyers of new Housing Development Board (HDB) flats to pay a 10% deposit upon acquisition. The topics were chosen based on the likelihood that the participants' involvement with them was low, drawing from the theory that low-involvement issues tend to invoke heuristic, rather than high-level, cognitive processing (Eagly & Chaiken, 1993; Petty & Cacioppo, 1986). The discussions were edited to comparable lengths in terms of average number of words per discussant. They were manipulated such that only four different discussants' postings made up the entire thread, and one particular discussant's postings constituted half of the discussion, in terms of total number of words. Participants were asked to make their evaluations of this predominant discussant. Language and expertise cues were manipulated only for this discussant, with the attributes of the rest being constant across treatments.

For the manipulation of the language variable, two versions of postings by the predominant discussants were produced, namely a Standard English and a Singlish version. The Standard English version utilized grammatically correct English and standard sentence structures. The Singlish version strongly featured the characteristics of CSE, such as non-standard grammar structures and local jargon. A linguistic expert specializing in the study of Singlish vetted the stimulus materials to ensure that variations in language manipulations were consistent, natural, and did not affect the content of the messages presented. Manipulations of the language used by the predominant discussant necessitated variations in word counts between conditions, but these were kept to a minimum. The postings of the non-dominant discussants were presented in Standard English.

For the manipulation of the expertise variable, the discussion remained the same, but the sidebar information was altered. Specifically, four kinds of identifying information were used to indicate level of expertise of the dominant discussants: the number of posts4 (greater than 1000 for Experts, fewer than 20 for Novices), a hierarchical label (Veteran or Newbie), the date joined (a date in 2001 or a date in 2003), and a star rating (five stars versus one star). Postings made by non-dominant discussants were displayed with median attributes (that is, a few hundred posts, the neutral hierarchical label Member, a random date in 2002, and a three-star rating).

Procedure

The experiment was conducted in two phases. The first phase was carried out in five sessions over three consecutive days. Participants filled out a questionnaire measuring basic demographic information, covariates, and motivations for participating in online political discussions, namely the desires for homophily and diversity. They were then scheduled for the experimental phase and thanked for their participation. The phase one sessions lasted about 30 minutes each.

The second phase was conducted roughly one week later, in seven sessions over three consecutive days, in groups ranging from six to 24 participants per session. Participants were requested to read all the postings from the first discussion forum on a web browser before proceeding to answer questions regarding the predominant discussant. The procedure was repeated with the second discussion thread. The second phase questionnaire included manipulation checks for language and expertise and questions on the study participants' motivations for participating in online political discussions. The sessions were all completed in about 30 to 40 minutes.

Measures

This study examined the impact of status cues on five general dependent variables (see Appendix). The first three variables examined perceived attributes of the predominant discussant in the forum discussion. Perceived informativeness and persuasiveness were measured using composite indices developed by Ng and Detenber (2005). The indices were seven-point, Likert-type, composite measures with 10 and eight items, respectively, for each variable. Source credibility was measured using McCroskey's 12-item Source Credibility Scale (taken from Rubin, Palmgreen, & Sypher, 1994), comprised of seven-point semantic differential scales. The scale is divided into two sub-dimensions: authoritativeness and character. The fourth and fifth dependent variables concerned the participants' desire to contribute to online political discussions. Willingness to participate was measured using a 10-item, seven-point Likert-type scale developed by Ng and Detenber (2005). The competing motivations for participation were the desire for homophily, measured using a four-item, seven-point Likert-type scale, and the desire for diversity, measured using a three-item, seven-point Likert-type scale (Ng & Detenber, 2005).

In examining the dependent variables, the possible mitigating effect of other factors needs to be considered. For example, Sternthal et al. (1978) noted that high credibility sources can be more persuasive than low credibility sources when issue involvement is low. Therefore, to control for this, involvement was measured for each of the two issues using selected items modified from the Personal Involvement Inventory (Ng & Detenber, 2005). Similarly, political efficacy may influence participants' perceptions of discussants, as well as their willingness to participate in online discussions. Political efficacy generally refers to "the feeling that individual political action does have, or can have, an impact upon the political process" (Acock, Clarke, & Stewart, 1985, p. 187); it is often broken down into two dimensions: internal/self and external/system. In Singapore, an increasing proportion of citizens believe that it is ineffectual to participate in local politics (Ooi, Tan, & Koh, 1998); thus it is especially important to consider the mediating effect this factor may have on participants' willingness to contribute to online political discussions. Therefore, two dimensions of political efficacy, system and self-efficacy, were measured as controls using scales taken from Craig, Niemi, and Silver (1990; see Appendix). Finally, to control for the possible differences between frequent users and non-users of political discussion fora, the respondents' frequency of discussing political issues online was measured using a six-item Likert-type scale from Ng and Detenber (2005).

Data Reduction and Analysis

The manipulation check items for the language and expertise variables were combined into separate composite measures. Composite measures were computed for the following control variables: unions issue involvement, HDB issue involvement, self-efficacy, and system efficacy. Individual items for the dependent measures of informativeness, persuasiveness, the two dimensions of source credibility (authoritativeness and character), and willingness to participate were also combined to form five composite measures. Reliabilities for all the composite scores were good for both issues. Table 1 contains the reliabilities for the control measures, while Table 2 reports the reliabilities for the manipulation checks and dependent variables composite measures.

 

Cronbach's ?

Unions Issue Salience

.90

HDB Issue Salience

.92

Political Efficacy: Self

.75

Political Efficacy: System

.66

Table 1. Reliabilities for control measures

 

Cronbach's ?

 

Unions Issue

HDB Issue

Language Manipulation Check

.90

.90

Expertise Manipulation Check

.77

.78

Informativeness

.85

.89

Persuasiveness

.90

.92

Source Credibility: Authoritativeness

.90

.84

Source Credibility: Character

.73

.77

Willingness to Participate

.92

.93

Table 2. Reliabilities for manipulation checks and dependent measures

To determine which control measures were significantly related to the dependent variables and therefore should be included as covariates in the ANOVA models, correlation analyses were run. For the Union issue, self-efficacy, r(80)=.27, p=.01, and issue involvement, r(80)=.28, p=.01, were significantly related to the willingness to participate. For the HDB issue, the composite measure for discussion participation, r(79)=.24, p=.03, was related to the measure for authoritativeness. In addition, self-efficacy, r(80)=.38, p<.001, and issue involvement, r(80)=.23, p=.04, were related to willingness to participate. Hence, each control measure was used as a covariate only in specific analyses, as noted below. The data were analyzed using repeated measures analysis of variance (ANOVA), with issue (Unions or HDB) as the repeated measure, and language and expertise as the between-subjects factors.

Results

Manipulation Checks

Mixed-model ANOVA analyses were run to determine whether there was a perceptible difference in the treatment levels of language and expertise. The overall model for language was significant, F(1, 78)=47.21, p<.001; Standard English (M=4.14, SE=.11), Singlish (M=3.09, SE=.11). For the individual issues, the language manipulations in both discussions were also significant, F(1, 78)=30.87, p<.001 for Unions; F(1, 78)=25.78, p<.001 for HDB.

In the expertise manipulation check, the overall model was significant, F(1, 78)=9.34, p=.003; Novice (M=4.30, SE=.10), Expert (M=4.73, SE=.10). For the individual issues, the expertise manipulation in the Unions discussion was significant, F(1, 78)=16.45, p<.001.

Although the means were in the proper direction, the difference was not significant for the HDB discussion, F<1. Given this somewhat "weak" manipulation, the results should be interpreted cautiously for the expertise variable. Table 3 summarizes the means by issue for the language and expertise manipulation checks.

 

 

Issue

 

 

Unions

HDB

 

 

M

SE

M

SE

Language

Standard English

4.43

.14

3.86

.14

 

Siglish

3.33

.14

2.86

.14

Expertise

Novice

4.11

.12

4.49a

.13

 

Expert

4.71

.12

4.65a

.13

Table 3. Means for manipulation checks

Note
: Means denoted with superscript do not differ significantly across levels at
p
< .05.

Informativeness

Hypothesis 1a postulated that discussants who used Singlish would be perceived as being less informative than discussants who used Standard English, while hypothesis 2a proposed that discussants who were rated as Novices would be perceived as being less informative than discussants who were rated as Experts. Results of the ANOVA model indicated that neither language nor expertise significantly affected informativeness, F(1, 76)=2.74, p=.10, for language; F<1, for expertise. Nor was the interaction between language and source expertise significant, F(1, 76)=3.34, p=.07. Therefore, hypotheses 1a and 2a were not supported. Power analyses indicate that with 80 subjects, there was a 37% chance of detecting a main effect of Singlish and a 44% chance of detecting an interaction effect between Singlish and source expertise.

Persuasiveness

Hypothesis 1b proposed that discussants who used Singlish would be perceived as being less persuasive than discussants who used Standard English, and hypothesis 2b predicted that discussants rated as Novices would be perceived as being less persuasive than discussants rated as Experts. Results showed that neither language nor expertise was significantly related to the perceived persuasiveness of the assessed discussants, F(1, 76)=2.26, p=.14, for language; F<1, for expertise. Language and expertise interactions were not significant, either, F(1, 76)=1.86, p=.18. Hypotheses 1b and 2b were thus not supported. Power analyses indicate that there was a 32% and 27% chance of detecting the main effect and the interaction effect, respectively.

Source Credibility

Hypothesis 1c suggested that discussants who used Singlish would be perceived to have lower source credibility than discussants who used Standard English. Hypothesis 2c postulated that discussants who were rated as Novices would be perceived to have lower source credibility than discussants who were rated as Experts. A repeated measures ANCOVA analysis was run for the authoritativeness dimension of source credibility, with discussion participation as a covariate, while an ANOVA model was constructed for the character dimension. Language did affect perceptions of authoritativeness, but the relationship was right at the level of statistical significance, F(1, 74)=3.94, p=.05; Standard English (M=4.71, SE=.11), Singlish (M=4.40, SE=.11). However, authoritativeness was not related to the expertise variable, F<1, nor was there a significant interaction between language and expertise, F(1, 74)=2.30, p=.13. A power analysis revealed that there was a 32% chance of finding an effect with 80 subjects.

There were no significant main effects for either independent variable on the character dimension of source credibility, Fs<2. However, there was a significant interaction between language and expertise, F(1, 75)=12.23, p=.001. In the Standard English condition, Novices had a higher character source credibility (M=4.65, SE=.13) than Experts (M=4.15, SE=.12). In the Singlish condition, the results were in the opposite direction: Experts were rated with a higher character source credibility (M=4.42, SE=.12) than Novices (M=4.04, SE=.13; see Figure 1). Thus hypothesis 1c was partially supported, but only with respect to the authoritativeness dimension. Hypothesis 2c was provisionally supported in the Singlish condition. However, it was counter-hypothetical in the Standard English condition, since Novices were rated with higher character source credibility than Experts. Follow-up tests on the interaction showed that in the Novice condition, there is a significant difference in Singlish's effect on character source credibility, F(1, 36)=10.11, p=.003, but no significant difference in the Expert condition, F(1, 39)=2.70, p=.11.

Figure 1. Plot of expertise condition against scores of character source credibility, grouped by language condition

Willingness to Participate

Hypothesis 3 predicted that when discussants used Singlish, participants would be more willing to participate in the discussion than when discussants used Standard English. Hypothesis 4 proposed that when discussants were rated as Novices, participants would be less willing to participate in the discussion than when discussants were rated as Experts. The ANCOVA model revealed that willingness to participate was not significantly related to the independent variables, F<1 for both language and expertise. There was no significant interaction found between language and expertise, F(1, 73)=2.03, p=.16. Thus hypotheses 3 and 4 were not supported. Power analysis showed that with 80 subjects, there was a 29% chance of finding the interaction effect between Singlish and source expertise on willingness to participate.

Motivations for Participation

RQ1 sought to investigate if the desires for homophily and diversity were affected by language and expertise. No significant main effects were found. However, for the homophily motivation, the interaction between phases and language was significant, F(1, 75)=9.44, p<.01. There was no difference between phase one and phase two scores in the Standard English condition, but differences were found in the Singlish condition (phase one, M=4.57, SE=.14; phase two, M=4.07, SE=.16). In other words, the use of Singlish by predominant discussants caused the desire for homophily to decrease over time. However, there was a difference in phase one scores for homophily between the conditions. A one-way ANOVA model revealed that this difference was significant, F(1, 78)=4.17, p=.05, Standard English (M=4.19, SD=1.00); Singlish (M=4.57, SD=.67). Thus, results should be interpreted with caution, because random assignment failed to produce equivalent groups at phase one. Over time, there were no main or interaction effects found for diversity between language or expertise, F<1.

Discussion

The results of this study suggest that language and expertise cues in a CMC context do not have quite the impact that was anticipated. In fact, the effects were very limited. Specifically, the status markers of language and expertise did not influence perceptions of a discussant's informativeness (H1a and H2a) or persuasiveness (H1b and H2b). They also did not affect Singaporean subjects' intentions to participate in the online discussions (H1d and H2d) or influence their motivations to do so (RQ1). The language and expertise cues did have some influence on perceptions of source credibility (H1c and H2c), but the relationships were not always straightforward. A brief discussion of the significant findings in the study follows.

For source credibility, authoritativeness was found to be related to language style. As predicted, discussants using Standard English were rated as being more authoritative than discussants using Singlish. The individual items in the McCroskey scale suggest that authoritativeness is related to the intellectual competence of the source (Rubin et al., 1994). Thus, the use of H-variety Standard English, which, according to Ferguson (1959) is associated with logic and the expression of "important thoughts," most likely causes respondents to perceive the source as being more competent. Somewhat surprisingly, however, the labeled expertise of the discussant did not affect perceptions of authoritativeness or intellectual competence. In this particular online context, it appears that how one "speaks" or writes is more influential than the degree of competence ascribed to individuals by various structural features of the website.

For the character dimension, a significant interaction effect was found between language and expertise. When discussants identified as Novices wrote using Standard English they were evaluated more favorably (i.e., more honest, friendly, pleasant, unselfish, nice, and virtuous) than Novices using Singlish. There was no significant effect of language when the discussants were identified as Experts. What this suggests is that Novices benefit from the use of Standard English, but Experts do not, nor do they suffer from using Singlish. It is possible that the Novices using the formal language style may have created some violations of expectations among the participants, leading to higher character evaluations. In contrast, if one is regarded as an Expert, language style does not affect attributions of character.

This interaction may be a consequence of the sequential processing of heuristic cues. In this case, the participants most likely encountered the expertise cue first. The label and other explicitly stated information conveying expertise (or lack thereof) are immediately apparent, or potentially so, upon first seeing the discussion forum. Having seen the expertise label, some expectations about the labeled discussant are set up by the participants based on heuristic processing. By contrast, the language cue is not immediately apparent, as a participant would need to read and process each posting before they are able to perceive the cue's presence. It is likely that the language cue is processed after the expertise cue and potentially modulated by it. In the case of the present findings, the label Expert seems to override any effect of language style, but for those labeled Novice, language style does indeed matter in terms of attributions of character.

The lack of significant effects from the status cues runs counter to previous research (e.g., Adkins & Brashers, 1995; Eastin, 2001), which found that characteristics such as powerful and powerless language styles and perceived expertise affected source credibility and message persuasiveness. A possible explanation is that the context of an online political discussion can have a strong influence on lurkers (the role assigned to the participants in our study). Their perceptual focus may be diverted toward the quality of argument and other substantive points of message content, rather than assessing information conveyed by status cues. Even so, the findings have some implications for theories relating to expertise, language, and political communication in computer-mediated environments, as discussed below.

The findings imply that the mental processing of expertise cues in CMC is different from that in face-to-face contexts. Studies of the latter have found that higher source expertise has significant effects on persuasiveness (e.g., Smith et al., 1998; see Sternthal et al., 1978, for a review). The absence of a relationship between expertise and perceptions of informativeness and persuasiveness in the present study suggests that the interpretation of status cues online is not straightforward or easily discerned. It may be that the Expert title does not specifically identify the kind of knowledge or experience held by the discussant, as opposed to past conceptualizations of expertise that used titles that were rich in social information. For example, the title of 'doctor' implies that the person has had extensive formal education and is also experienced in the area of medicine. By contrast, it is impossible to tell if an 'expert' possesses a high level of education, experience, or the kind of specialized knowledge relevant to the topic of discussion.

Our findings also suggest that the language cue does not have a pervasive or uniform impact on all the perception variables examined. One explanation as to why variations in language style did not consistently affect the perception variables is the possibility that the participants in the study, being Singaporeans and familiar with Singlish, had internalized the local language style. They might thus unconsciously discount this as a heuristic cue, choosing instead to evaluate the message content for its own sake. This explanation could be tested by comparing the evaluations of the discussant by a Singaporean versus a foreigner who is not familiar with the local language style. The uses of Standard English and Singlish could also be tested across different communication contexts, such as online and offline, or by varying the topic and degree of formality normally associated with a given context.

It is also possible that the use of truncated or ungrammatical styles is expected in the context of Internet communication. This is consistent with the linguistic concept of language register, which specifies what language style is appropriate and expected to be used in a particular setting (Halliday, 1978). The development of the abbreviated style of writing that is commonly seen in Internet-based interpersonal communication, alongside more formal writing seen on corporate and educational websites, could indicate that both formal and informal registers operate in the online context. Thus, when an ungrammatical or colloquial writing style is used, it may not be seen as an anomaly to normal Internet communication, and it may subsequently be discounted as information relevant to impression formation.

Limitations and Directions for Future Study

The concept of expertise in CMC contexts has not been adequately addressed by existing research, and the present study has its limitations as well. While the characteristics of source expertise we examined may have been ecologically valid, they may not be strong representations of the variable. Furthermore, the discussants' number of postings, dates joined, star ratings, and the designative labels were all manipulated concurrently, so their individual impact could not be assessed. More research is needed to investigate what these structural attributes of discussion fora mean to online readers. Specifically, a clearer explication is needed of the readers' criteria for being an expert, the ways in which these cues may be shown, and the impact these cues can have on the perceptions of the contributor and the message content.

With the exception of a very small number of studies (e.g., Gupta, 2006), the use of Singlish online remains largely unexplored. This presented another obstacle, since while there are established rules for Singlish, it was difficult to translate consciously in a natural fashion from Standard English into Singlish and vice versa. There was a trade-off between distinctively manipulating one discussant's language style and ecological validity. It was unlikely that a single discussant would write very differently from other discussants in the same forum, as discussants tended to adapt to each other's writing style. If the other discussants decided to adapt to the style of the original discussion poster, it would be difficult to make a naturalistic manipulation based on judgments in language style alone.

To address the lack of information on the use of colloquial styles online, more formal investigation into norms of usage is necessary, particularly into the purposes for employing such styles online, the degrees to which they can be varied, and how they are processed by other participants. More research is needed to discover what language register is seen as appropriate for Internet communication, and if this varies by format (e.g., discussion board or web page), function (e.g., instructional or relational), context (e.g., the message content or topic), or intended audience (e.g., students or professionals). Although this study did not find much evidence for the impact of language style in CMC, we believe that it is still an important concept to investigate further.

Acknowledgments

This project was supported in part by a grant from the Singapore Internet Project. An earlier version of this paper was presented to the Communication and Technology Division of the International Communication Association at the annual conference held in New York, NY, May 2005. Correspondence should be addressed to the fourth author.

About the Author