How Gendered Communication Styles Affect Source Credibility:
Attributions in Computer-mediated Communication
A Research Project
Presented to: Dr. Gass; HCOM 308; Fall 2006
The advent of the digital age mandates that communication researchers take a special interest in computer-mediated communication (CMC). In particular, CMC presents scholars with an interesting context in which they can further test and refine communication theory. Gender is an important variable in virtually all contexts of communication, and CMC is no exception.
Communication research shows that a person tends to have more success persuading another person if the two people are opposite genders (Gass & Seiter, 2003, p. 102; Ward, Seccombe, Bendel, & Carter, 1985). For example, men tend to have more success in persuading women than women do. In addition, women tend to have more success in persuading men than men do. This phenomenon, known as the “cross-sex effect” in speech communication, became apparent after the 1970s, because studies at and before that time primarily featured men as confederates in persuasion studies (Gass & Seiter, 2003, p. 102; Ward et al, 1985).
However, these studies do not seem to identify what causes the cross-sex effect in persuasion. Specifically, the cross-sex effect does not yet discriminate between message-sender sex and the gendered language style of the message-sender. In face-to-face communication, a communicator’s sex is readily apparent to the communicators, so researchers cannot easily isolate message content from the message-sender’s sex. Thomson (2006) identifies gendered communication styles typical of both men and women. In addition, Thomson points out in his study of CMC that “people are very flexible in the language style they use” (2006, p. 167). Text-based computer-mediated communication has both the virtue and the vice of being a medium where a communicator can use a communication style typical of a member of the opposite sex without disclosing his or her true gender. For example, an Internet game, known as the Turing Game, features people pretending to be of the opposite gender. In a real-time text-based CMC environment, these individuals perform their pretend gender in the ‘e-presence’ of a panel of judges. These judges then vote what they believe is the gender of the contestant. According to Herring and Martinson, the Turing Game’s judges typically “base their assessments mostly on stereotyped content, leading to a high rate of error” (2004, p. 424).
Fortunately, one can attribute that high rate of error to the intention of the participants, which is deception. In other contexts, however, where a conversant does not have a motive to mask his/her gender, the rates of predictive error lower dramatically. For example, in a classroom study, confederates asked students to write an impromptu essay describing a picture, then the researchers studied the essays for signs of gendered language (Mulac & Lundell, 1986). “A discriminate analysis found that the frequencies of a set of 17 linguistic variables could be used to classify the participants’ gender with 87.5% accuracy” (Thomson & Murachver, 2001, p. 193).
Various scholars argue that CMC limits nonverbal and social cues. For example, “the reduced context cues perspective (Kiesler et al., 1984; Sproull & Kiesler, 1986) similarly assumes that CMC filters out elements that regulate interaction and impression formation between communicators” (Pena, 2006, p. 96). Additionally, Anderson and Emmers-Sommer present the “cues filtered out perspective” (2006, p, 154). This perspective “advocated that CMC was not conducive for forming close ties online due to the minimal social context and nonverbal cues inherent in CMC” (Anderson & Emmers-Sommer, 2006, p. 154). Furthermore, Gibbs, Ellison, and Heino agree that CMC offers “relatively limited nonverbal cues” (2006, pp. 154-155).
Other research indicates that proficient CMC users have developed ways to convey information that simulates nonverbal and social cues (Krohn, 2004). These proficient CMC individuals use text to form emotive icons, called emoticons, to convey emotion through the use of as few as two characters of text. For example, the emoticon =( could demonstrate sadness, and the emoticon =) could demonstrate happiness. However, although there are more complex emoticons (e.g. m( . . ) m showing a person using his/her fists in a combat stance), and adroit CMC participants use a variety of onomatopoeia, these techniques may or may not match authentic nonverbal communication.
While CMC significantly masks nonverbal cues, not all researchers agree that CMC masks social identity. In the context of discussion in Internet chat rooms, for example, Thomson argues, “there are frequently cues to social identity, such as the nature of the topic being discussed” (2006, p. 168), and, “participants shifted toward group norms when social identity was salient and away from group norms when personal identity was more salient” (2006, p. 168). Thomson also indicates that “self-stereotyped behavior is accentuated when individuals are anonymous compared with when they are individuated” (2006, pp. 168-169). In short, not only does CMC leave many signs of social identity intact, but also anonymity in CMC may increase self-stereotyped behavior, including amplifying gendered language (Thomson & Murachver, 2001, p. 196).
The facet of CMC that limits nonverbal and social cues may permit research to refine the theory of the cross-sex effect of persuasion. For example, in face-to-face communication, if a female speaker chooses to use a male, perhaps assertive, communication style, others will have a difficult task of ignoring the fact that the speaker is female. In short, mediated communication before CMC (i.e. face-to-face, audio-only) did not allow communicators to separate nonverbal gender cues from verbal gender cues. The only exception to this rule is text-based communication. CMC may allow research to acquire a more intimate understanding of the cross-sex effect of persuasion. Does the persuader’s communication style matter, or does the effect merely originate from a communicator’s perceived sex? In pursuit of this research interest, I present the following research question:
How do gendered communication styles affect source credibility attributions in text-based computer-mediated communication?
Gendered communication styles will be the independent variable. Source credibility will be the dependent variable. Text-based computer-mediated communication will be a constant. Gender of both the sender and receiver of messages will be an important moderating variable.
At this time, certain terms need definitions. I will begin with gendered communication style. Thomson (2006) identifies gendered communication styles in terms of speech preferences. In general, Thomson points out “female-preferential features” of speech, which include:
Personal information, reference to emotion, asking a question, referring to a previous comment, agreeing with another’s statement, apology, compliment, self-derogatory statement, modal or hedge, intensive adverb, subordinating conjunction, stating an opinion, and statements of emphasizing similarities or solidarity with others. (2006, p. 169)
Thomson also identifies “male-preferential features” of speech as “giving a directive, disagreeing, insult, adjectives, statements emphasizing differences between group members” (2006, p. 169).
In order to better measure gendered language, the experiment will operationalize gendered communication by conducting a content analysis on the text in the CMC experiment. Researchers appear to have already constructed a framework for studying gendered language. Several studies define various language features, count the number of times that the language feature appears in a given text, and then correlate totals with the person who authored the text. Two studies utilize 13 separate language features (Thomson & Murachver, 2001; Thomson, Murachver, & Green, 2001). These studies utilize many of the same features above: adjectives, intensive adverbs, subordinating conjunctions, opinions, compliments, modals/hedges, apologies, insults, questions, references to previous message and answers, self-derogatory comments, emotion, and personal information. In addition, a later study utilizes 18 separate language features, adding six features to the list: agreement, disagreement, solidarity, differences, boasts, and directives (Thomson, 2006, p. 170). Thomson found that adjectives became a statistically insignificant language feature, and he removed it from the list (2006, p. 171). A scale count can measure these textual components as a variable, and tracking variations to the number of these language features can serve to make gendered language an independent variable. This scale count will not allow the experiment to determine that a text falls within a discrete category of female or male style. However, the ratio of male to female language features will allow the experiment to determine degrees to which a text expresses a stereotypical female or male style.
Second, I need to define text-based computer-mediated communication. Spitzberg (2006) identifies computer-mediated communication as “any human symbolic text-based interaction conducted or facilitated through digitally-based technologies.” Spitzberg includes some media in this category, such as cellular phone conversations and digital video-conferencing, but the inclusion of “text-based” into the research question narrows the context and excludes cellular phone conversations and video-conferencing. The justification for narrowing the CMC context to text-only is to examine mediated communication that filters nonverbal cues (i.e. speaking tone, body appearance).
Finally, the dependent variable, source credibility, needs to be measured. The most efficient and appropriate way to measure the attribution of source credibility will be to have the subject of the experiment complete the Measure of Source Credibility scale (McCroskey & Teven, 1999; see also Brann, Edwards, & Myers, 2005). Using that scale, each subject of the study will rate his or her conversation partner’s source credibility.
Many studies explore the effects of gender on communication. As a result, a number of scholars make conclusions that do not cohere with the rest of research. This literature review will begin by exploring the topic of gender and its relation to credibility and social influence. This review will examine the studies that support and contradict the assumptions of the cross-sex effect. In addition, this review outlines important studies that examine relevant CMC topics and communication theory.
Review of Theory
The majority of communication theory that relates to gender assumes that gender is an obvious facet during interaction. This review of theory will outline several communication theories, the correlations that those theories predict, and their limitations.
Social Role Theory
Carli reminds us that “the association of men with powerful, high-status roles has resulted in their generally gaining higher status than women” (Carli, 2004, p. 135). According to Alice Eagly’s Social Role Theory, “men and women are distributed differently into social roles” (Carli, 2004, p. 135). In traditional communication contexts (i.e., not CMC), roles are generated and adhered to every day, presumably because the factors of a person that produce expectations are salient to the people around a communicator. However, in CMC, those cues do not meet other communicators through the same channels. For example, in text-based CMC, a communicator could only discover the other person’s gender via the text that the other person generates.
Expectation States Theory
Expectation States Theory states that “gender acts as a diffuse status characteristic, a general attribute that is associated with an individual’s relative status in society” (Carli, 2004, p. 135). Carli also reports that diffuse status characteristics include gender, race, degree of physical attractiveness, and education (2004, p. 135). Building on Social Role Theory, we can conclude that diffuse status characteristics are basic cues for building social roles. In general, we could scarcely consider more basic cues for social roles than gender, race, physical attractiveness, and education. Also, in considering the CMC context, text-based communication seems especially apt to masking these diffuse status characteristics. Of course, the existence of these characteristics will leak through a CMC medium. A communicator may disclose, intentionally or unintentionally, overtly or discretely, any of these characteristics. As communicators tend to be more consciously aware of overt disclosures in face-to-face communication, we can assume that conversants would be more consciously receptive to overt disclosures in CMC.
According to Berger et al (as cited by Carli, 2004, p. 135), people generally attribute a high-status individual with more competence than a low-status individual. Carli adds that “people seek the opinions of high-status people and yield to their influence more than to people of low status” (Berger et al, 1977 as cited by Carli, 2004, p. 135). According to Carli, this tendency turns into a self-fulfilling prophecy. The question for CMC is whether diffuse status characteristics are salient enough in CMC to influence source credibility attributions. This question may hinge on whether those characteristics manifest themselves in the CMC medium.
Review of Research
In addition to theory, research experiments are important to the influence of gendered language on source credibility in CMC. Research that applies to this topic will fall into two categories. On one hand, most traditional research on gender and social influence surveys how men and women interact in face-to-face communication. The theories reviewed above rely on decades of research that falls into this category. On the other hand, recent research in CMC affords detailed experiments on gender and social influence. Furthermore, while most of this CMC research does not disconfirm current theory, it suggests areas where more research can further refine those theories.
Carli (2004, p. 138) reports a meta-analysis of studies, all of which study the degree that men and women separately influence others. Lockheed conducted this meta-analysis of 29 studies in 1985, and concluded that “men exert greater influence and exhibit more leadership behaviors than do women” (Carli, 2004, p. 138). In addition, this difference is apparently not due to performance differences or message differences. According to Propp, “In group interactions members were more likely to attend to ideas contributed by men and to use them in solving group problems than to the identical ideas contributed by women” (1995 as cited by Carli, 2004, p. 138).
These “identical ideas” may leave room for interpretation, leaving researchers wondering whether a communicator can moderate an identical idea with language styles and minor differences in word choice. However, other studies seem to point to the fact that the male and female agents may have well used the same words. According to Carli:
Research has shown that men remain more influential than women, even when the persuasive messages of the male and female agents are manipulated to be identical (Altemeyer & Jones, 1974; DiBerardinis, Ramage, & Levitt, 1984) or when the performances of the male and female agents are manipulated to be equally good (Schneider & Cook, 1995; Wagner, Ford, & Ford, 1986). (2004, p. 138)
These studies all clearly support the accepted fact that gender affects social influence. However, these research experiments do not occur in environments where the genders of the interactants are hidden or obscure. We have to assume that if the gender of the communicators is salient, then that variable will eclipse the effect of gendered language, which may moderate social influence.
While some researchers believe that men are more persuasive than women, regardless of the situation, the cross-sex effect is a phenomena that communication researchers have known about for decades. According to Ward et al, the first person to devote scholarly attention to the cross-sex effect was Alice Eagly (1985, p. 269). There are several competing explanations for the cross-sex effect. Shaffer provides one explanation, arguing that “communicator sex may influence sex differences in opinion change though interpersonal attraction” (1975 as cited by Ward et al, 1985, p. 269). Eagly (as cited by Ward et al, 1985) contends:
Normative pressures governing the expression of opinions may vary with the make-up of the communicator-recipient dyad. Female recipients may follow a norm of deference to male authority and consequently manifest more opinion change when exposed to a male communicator, whereas males may adhere to a norm of chivalry and hence show more opinion change when exposed to a female communicator (Eagly, 1978:97). This explanation stems from the theory that greater female influenceability in cross-sex contexts stems from role-related expectancies derived from status inequalities found in the larger society. (p. 269)
Any of these explanations, if true, would lead research to conclude that gendered language does not affect source credibility attributions in CMC, unless the gendered language of one person in a dyad caused the other person to make a conclusion about the other’s gender.
Few experiments confirmed the cross-sex effect in 1985 (Ward et al, 1985, p. 270). In response to those mixed results in the research, Ward et al make one requirement among others in order to maintain the validity of the above explanation. According to Ward et al, “The persuasive messages must be presented by communicators who are physically present rather than presented through more remote channels of communication such as written messages or audio tape recordings” (1985, p. 270). Audio tape filters visual cues, and written messages filter not only visual cues, but also all nonverbal cues. Therefore, Ward’s et al analysis indicates that a cross-sex effect in CMC should either not exist, or be greatly mitigated. However, since gendered language may cue communicators to make a conclusion regarding the gender of their interactant, this effect will still appear.
Three research articles that focus on CMC are particularly relevant to the present inquiry. All three of these articles were published in the last five years, and all three of them featured Rob Thomson as their main author.
In 2001, Thomson and Murachver argued that content analysis could predict the gender of a communicator in CMC. In this study, the researchers applied 17 linguistic variables to predict a participants’ gender with 87.5% accuracy (Thomson & Murachver, 2001, p. 193). Although 87.5% is a high rate of accuracy, it is far from deterministic. Thomson and Murachver recognize this, and attribute the indeterminacy partially to variations from person to person (2001, p. 194). In addition, Thomson and Murachver argue, “When conversing with a member of the other sex, both women and men change some aspects of their language style towards the gender-preferential style of their partner” (2001, p. 194).
In an effort to isolate what causes a communicator to change his or her communication style, Thomson, Murachver, and Green conducted another study in 2001. They paired subjects into e-dyads and recorded their conversations via e-mail, although the subjects did not know the gender of their partner. This time, Thomson et al conclude:
Analysis revealed that the netpal’s language style, and not the participants’ gender, predicted the language used by participants in their e-mail replies. Female and male participants used the gender-preferential language that matched the language used by their netpals. (2001, p. 171).
In this study, Thomson et al conclude that gender and social identity may be particularly salient in CMC, due to the absence of many other cues (2001, p. 171).
As a follow-up to this topic, Thomson published another research experiment in 2006. This experiment centered on the notion that the language features that a communicator uses depend on a number of factors. Thomson’s experiment explored the possibility that the topic of discussion could affect the gendered language in CMC discussion. Thomson’s experimental analysis concludes that the topic affects gendered language. In particular, “There was a higher frequency of female-preferential features in discussions about male-stereotypical topics and a higher frequency of male-preferential features in discussions about male-stereotypical than in discussions about other topics” (Thomson, 2006, p. 172). In addition, “Discussions about neutral topics, however, showed a low frequency of both female and male features rather than a mixture of both” (Thomson, 2006, p. 172). This latter phenomenon indicates that topic does more than empower a person to talk about something of which they are expert (Thomson, 2006, p. 172). This study confirms that topic can operate as an independent variable in a research experiment. It also suggests that the current study should introduce topic as an additional moderating variable.
Participants and Design
The experiment will attempt to recruit speech communication students on a volunteer basis. The recruiters will ask the students to participate in a research experiment that will explore how people communicate using computers. During the recruitment process, the experimenter will inform the students that participation in the experiment will be completely voluntary and confidential, and participants will have the option to withdraw from the experiment at any time. An experimenter will ask each participant to read a consent form with this information, sign it, then return it to the experimenter. This consent form will describe the cross-sex effect as scholars currently understand it and what information the experiment promises to contribute to that phenomenon. After the experiment concludes, the experimenters will disclose the details of the experimental design, including the data collected, and the measurement techniques. The experimenters will give the students an opportunity to ask questions. In addition, if a dyad consents to disclosing their identities to each other, the experimenters will disclose that information to the dyad partners.
When the students consent to the experiment, the experimenters will divide the students into dyads. The experimenters will randomly assign each participant to one of two groups. The first group will be same-sex dyads. The second group will be cross-sex dyads.
The experiment will utilize Internet e-mail as a mode for the dyads to interact. After the experiment assigns participants to their dyads, each partner will be assigned an email address and access to a web-based email client (which will record the emails for coding). The participants will also receive training on how to use the email client. The name of the email address will be a random word, chosen to obscure the identity and gender of the dyad partner. The experimenters will inform the participants that they will correspond with their partner exclusively through e-mail. In addition, the experimenters will inform the participants that the design of the experiment requires the participants not to disclose either their real names or their gender to their partner. The experimenters will ask the participants to exchange as many emails as possible for a period of two weeks.
After this period, each participant will rate his or her partner using the Measure of Source Credibility scale (McCroskey & Teven, 1999; see also Brann, Edwards, & Myers, 2005).
Coding the gendered language that each participant uses will require a content analysis, targeting Thomson’s (2006, p. 170) 18 language features. In order to obtain an accurate content analysis, two independent analysts will tally the occurrences of language features. The experiment will only consider the occurrences that both of the analysts tally. In addition, the analysts will count the total number of words that each participant used in the emails. This content analysis will supply the experiment with a number of male stereotypical language features, a number of female stereotypical language features, and a number of email words. Therefore, each participant will exemplify a certain degree of male stereotypical and female stereotypical language per email word.
Coding the overall source credibility attributions is a matter of deriving a mean average of the McCrosky’s and Teven’s 18 inventory areas of ethos/source credibility. Individual mean averages within competence, goodwill, and trustworthiness may provide useful results for the experiment.
Anderson, T. L., & Emmers-Sommer, T. M. (2006). Predictors of relationship satisfaction in online romantic relationships. Communication Studies, 57(2), 153-172.
Brann, M., Edwards, C., & Myers, S.A. (2005). Perceived instructor credibility and teaching philosophy. Communication Research Reports, 22(3), 217-226.
Carli, L. L. (2004). Gender effects on social influence. In J. S. Seiter, & R. H. Gass (Eds.), Perspectives on persuasion, social influence, and compliance gaining (pp. 133-148). Boston: Pearson Education.
Gass, R. H., & Seiter J. S. (2003). Persuasion, social influence, and compliance gaining. (2nd ed.). Boston: Pearson Education.
Gibbs, J. L., Ellison, N. B., & Heino, R. D. (2006). Self-presentation in online personals – the role of anticipated future interaction, self-disclosure, and perceived success in Internet dating. Communication Research, 33(2), 152-177.
Herring, S. C., & Martinson, A. (2004). Assessing gender authenticity in computer-mediated language use – evidence from an identity game. Journal of Language and Social Psychology, 23(4), 424-446.
Krohn, F. B. (2004). A generational approach to using emoticons as nonverbal communication. J. Technical Writing and Communication, 34(4), 321-328.
McCroskey, J. C., & Teven, J. J. (1999). Goodwill: a reexamination of the construct and its measurement. Communication Monographs, 66(1), 90-103.
Mulac, A., & Lundell, T. L. (1986). Linguistic contributors to the gender linked language effect. Journal of Language and Social Psychology, 5(2), 81-101.
Pena, J., & Hancock, J. (2006). An analysis of socioemotional and task communication in online multiplayer video games. Communication Research, 33(1), 92-109.
Spitzberg, B. H. (2006). Preliminary development of a model and measure of computer-mediated communication (CMC) competence. Journal of Computer-Mediated Communication, 11(2), 629-666.
Thomson, R., Murachver, T., & Green, J. (2001). Where is gender in gendered language? Psychological Science, 12(2), 171-175.
Thomson, R., & Murachver, T. (2001). Predicting gender from electronic discourse. The British Journal of Social Psychology, 40, 193-208.
Thomson, R. (2006). The effect of topic of discussion on gendered language in computer-mediated communication discussion. Journal of Language & Social Psychology, 25(2), 167-178.
Ward, D. A., Seccombe, K., Bendel, R., & Carter, L. F. (1985). Cross-sex context as a factor in persuadasibility sex differences. Social Psychology Quarterly, 48(3), 269-276.
APPENDIX: Measure of Ethos/Credibility
Instructions: Please indicate your impression of your email partner by circling the appropriate number between the pairs of adjectives below. The closer the number is to an adjective, the more certain you are of your evaluation.
Intelligent 1 2 3 4 5 6 7 Unintelligent
Untrained 1 2 3 4 5 6 7 Trained .
Inexpert 1 2 3 4 5 6 7 Expert .
Informed 1 2 3 4 5 6 7 Uninformed
Incompetent 1 2 3 4 5 6 7 Competent .
Bright 1 2 3 4 5 6 7 Stupid .
Cares about me 1 2 3 4 5 6 7 Does not care about me .
Has my interests at heart 1 2 3 4 5 6 7 Does not have my interests at heart
Self-centered 1 2 3 4 5 6 7 Not self-centered .
Concerned with me 1 2 3 4 5 6 7 Unconcerned with me .
Insensitive 1 2 3 4 5 6 7 Sensitive.
Not understanding 1 2 3 4 5 6 7 Understanding .
Honest 1 2 3 4 5 6 7 Dishonest
Untrustworthy 1 2 3 4 5 6 7 Trustworthy .
Honorable 1 2 3 4 5 6 7 Dishonorable
Moral 1 2 3 4 5 6 7 Immoral
Unethical 1 2 3 4 5 6 7 Ethical .
Phoney 1 2 3 4 5 6 7 Genuine