Back to Vol. 2, No. 4 Table of Contents
[Conversation]
Collaborative U CMC Play E-Commerce Symposium Net Law InfoSpaces


On-Line Smiles: Does Gender Make a Difference
in the Use of Graphic Accents?

Diane F. Witmer

Department of Communication
Purdue University

Sandra Lee Katzman

Interac Co Ltd

Table of Contents

Abstract

In the gender-bending world of computer-mediated communication (CMC), is it possible to determine the gender of a message sender from cues in the message? This study addresses the question by drawing on current literature to formulate and test three hypotheses: (i) women use more graphic accents than men do in their CMC, (ii) men use more challenging language in CMC than do women, and (iii) men write more inflammatory messages than do women. Results indicate that only the first hypothesis is partially supported and that women tend to challenge and flame more than do men in this sample group. The authors also discuss implications and pose questions for additional research.

Introduction

"On the Internet, nobody knows you're a dog." Thus reads the caption of a Peter Steiner cartoon (July 5, 1993: 61). The cartoon depicts a dog at a computer, commenting on the anonymity of computer-mediated communication (CMC). The dog's words express awareness of a key characteristic of virtual reality: that CMC can mask personal characteristics and identities of cyberspace travelers to create personal anonymity in a public arena.

Impersonating a member of the opposite sex is a fairly common practice in the world of CMC. Bruckman (1993) reports that both men and women are aware of extra attention female personae attract in virtual communities called MUDs (multi-user domains). As a result, men often log onto MUDs as women (Bruckman, 1993), and Curtis (1992) reports that the most sexually aggressive of the "female" MUD inhabitants often are men. How does one know whether electronic exchanges are occurring with a man or with a woman? This paper considers the possibility that the ways in which men and women communicate via computers may differ, a concept that has made the cover of Newsweek (Kantrowitz, 1994; Tannen, 1994), and that those differences may be discernible in the typed symbols through which computer users attempt to create shared (or unshared) meaning.

Users create relatively consistent personae and assume social or organizational roles (e.g., Barlow, 1990; Matheson, 1991; Reid, 1991; MacKinnon, 1992; Herring, 1993; Kramarae and Taylor, 1993), perhaps not always intentionally. Herring (1993), for example, reports gender differences in the discourse of academic computer news list participants, including quantity (males tend to send long messages), topic selection (men most frequently discuss issues; women most often contribute to personal discussions), and style (e.g., strong assertions by males, attentuated assertions by females; self-promotion by males, apologies by females; challenges by males, supportive remarks by women). Might computer-mediated discourse, then, enable a reader to discern the gender of an author in CMC?

Ever since linguist Lakoff's (1973) work indicated that women use a language that is powerless, scholars have scrutinized the verbal communication styles of women and men. Quina, Wingard, and Bate (1987) confirm Lakoff's feminine language, and characterize it as higher in social warmth than masculine language. Mulac and Lundell (1986) indicate that certain language features potentially serve as linguistic gender markers, including aesthetic quality and emotional expressiveness. Bate (1988) summarizing the work of Eakins and Eakins (1978), DuBois and Crouch (1976), and Berryman-Fink (1978), concludes that gendered language styles include adjectives and adverbs to qualify or equivocate feminine discourse, and active verbs without qualifiers in masculine language (p. 95). Tannen (1990), too, indicates that women's communication tends to be more supportive and rapport-building. In contrast, she states that men typically are more report-giving and informative. Do these differences exist on a computer screen or printout? If so, they might be manifested in some of the specialized symbols that are familiar to those who engage in CMC.

Most of the literature that concerns communicative gender differences focuses on face-to-face verbal or written language, but relatively little deals with CMC. Beniger (1987) posits that technological advances are blurring the distinctions between interpersonal and mass communication, and Herring (1993) extends the concept of gendered interpersonal discourse to CMC. It is obvious, though, that in the context of a computerized medium, social structures and discursive practices are created with a narrowly-defined set of symbols. CMC systems usually support only a "low-end ASCII" character set. This means the communicator is restricted to American upper and lower case letters and numerals, and some commonly-used mathematical and punctuation symbols (e.g., "$," "%," "()," "+," etc., omitting umlauts and other European characters). Some of the aggregate symbols are called "emoticons" or "smiley faces," which are intended to represent a human face, using a colon for eyes, a hyphen for a nose, and a parenthesis for a smile or frown.

smile :-)
frown :-(


Grappling with terminology for these new symbols, Stanford doctoral candidate in communication Dennis Kinsey (1993) suggested the phrase, "graphic accents" (GAs), to one of the authors. The term is used in this paper to refer generically to emotional, artistic, and directional devices. Because of their potential to add expressiveness, emotion, and aesthetics to written discourse, GAs are the focus of this study as possible gender markers in CMC.

Hypotheses

The research on gender differences in verbal communication is methodologically and paradigmatically uneven, but often written from a critical theory perspective. This study distinguishes between sexist discourse, which often focuses on power imbalances or domination, and discourse that simply may disclose or indicate the gender of the author. Quina et al. (1987) report that nonfeminine linguistic style is perceived as lower in social warmth than feminine discourse, but Epstein (1986) proposes that gender differences are context-specific and superficial. Ganong and Coleman (1985) posit that feminine-stereotyped communication is more expressive of emotion than masculine discourse, and Ivy and Backlund (1994) support the idea that women are more expressive both verbally and nonverbally than men. Herring's (1993) findings of gender differences in academic CMC are congruent with Bate's (1988) synthesis of gendered language. Herring proposes that in general, women's computer-mediated language includes "attentuated assertions, apologies, explicit justifications, questions, personal orientation" and supportiveness (p. 8). Men, on the other hand, communicate with "strong assertions, self-promotion, presuppositions, rhetorical questions, authoritative orientation, challenges [to] others, and humor/sarcasm." An additional dimension of feminine language is aesthetic quality, (Mulac, Bradac, and Mann, 1985; Mulac and Lundell, 1986), which is identified, in part, as evidence of beauty or ugliness in the discourse. GAs, then, may serve as methods of adding emotional or emphatic expressiveness to the computer-mediated text. In aggregate, they might be used to help express emotion, equivocate, and (perhaps) add an aesthetic element to feminine discourse. This evidence leads to the hypothesis that:

H1 Women use more graphic accents than
men in their computer-mediated discourse.

Because the nature of an e-mail message may affect the extent to which GAs are used, it also is useful to look at other message characteristics. If men's language is more assertive than women's (Tannen, 1990; Bate, 1988), we may see more challenges or "flames" (hostile or abusive language) in their e-mail messages. If so, challenging or flaming messages may include fewer GAs. This evidence suggests the following hypotheses:

H2 Men use more challenging language than
do women in computer-mediated discourse.
H3 Men flame more often than women
in computer-mediated communication.

To test the hypotheses, data were collected and statistically analyzed as described in the next section of this report.

Development of the Study and Methodology

This study originated as both part of the creation and the result of an international computer-supported collaboration called simply "ProjectH." Sudweeks and Rafaeli (1996) explain in their detailed report of ProjectH that a quantitative methodology was chosen in an effort to dovetail with laboratory-based CMC studies. ProjectH members purposely created a codebook as broad as possible to accommodate a diverse range of research questions and hypotheses. To some extent, then, the study reported here is method driven (Rafaeli, Sudweeks, Konstan, and Mabry, 1994), although one of the authors was interested in GAs early on and helped write the portion of the ProjectH codebook that focused on GAs. The codebook included, in part, three questions on GAs, the gender of the message senders (where discernible), and codings of message content.

Sampling

Rafaeli et al. (1994) provide a detailed description of the sampling process in the ProjectH Technical Report. In brief, the data were collected over a period of months, beginning with a randomly-selected Monday (March 15, 1993), from publicly-posted newsgroups and special interest groups on the Internet and CompuServe, which is an online subscription service. After excluding groups that were inappropriate for study (e.g., languages other than English, extremely low volume, etc.), groups were selected at random, and 100 messages or three days' worth of postings, whichever was greater, were gathered from each group, starting on random Mondays. ProjectH members collected a total of 3000 messages, with the intent of having each block of 100 messages coded by two coders, in order to determine reliability. Rafaeli, et al. (1994) describe the outcome:

For various reasons, 40 per cent [sic] of potential coders were unable to code. Of the 37 lists (batches of 100 messages) distributed, 20 were single coded, 12 were double coded, and 5 were not coded. Of the 32 coded, 4 were unfinished, giving a final tally of 20 single coded and 10 double coded lists. The database(s), therefore, has a total of 4000 messages from fully-completed lists, of which 3000 are unique.

Since no single person coded all the messages, and since only 1,000 of the messages were coded by two persons, it was necessary to infer interrater reliability. Ten pairs of ProjectH members coded 100 identical messages, totaling 1,000 messages from ten lists, which comprised one-third of the 3,000-message sample.

Cohen's kappa was computed for each pair of coders on the variables of interest to this study. Kappa ranged from extremely low (approaching zero) to 1.0, and empty cell rows or columns in the contingency matrix rendered kappa incalculable for several variables. Other coding pairs had high levels of agreement, often approaching 100 percent, but the empty cells produced unreadable kappa values. In general, however, the coding for those variables seemed to be the most consistently reliable between individual coders. The six variables that indicated high interrater reliability were the only ones retained for analysis. Low interrater reliability required that a number of potentially useful variables be discarded, including emodevice and categories intended to determine levels of humor, apology, and degree of formatting in the messages. Details of reliability determination appear in in Witmer and Katzman (1997). Once reliability was determined, the duplicate sample batches were randomly eliminated from the database, leaving a total of 3000 unique messages.

Results

Descriptive statistical analysis revealed that of the 3,000 messages, gender could be determined in only 2,599. Of these, the overwhelming majority of message authors, 83.6 percent, were male, and 16.4 percent were female. Of the 3,000 messages, 396 (13.2 percent) included GAs (emoticons or articons). The genders of 53 authors (13.4 percent) of GA-containing messages were unknown. Of the remaining 343 messages that contained GAs and for which the gender of the author could be determined, females posted 102 (29.7 percent), and males posted 241 (70.3 percent).

To test the hypotheses, a 2(gender) x 2(flame) x 2(challenge) factorial ANOVA and t-tests were conducted on the 343 GA-containing messages for which the author's gender could be identified. Since there were unequal numbers of males and females, the "regression approach" in SPSS was used for the ANOVA, which is the most conservative method of adjusting for unequal n's (Tabachnick and Fidell, 1989: 339-342).

Results of the t-tests indicated that females used GAs significantly more often than males [t(341)=4.51, p=.0000], included more challenges in their messages [t(341)=5.66, p=.0000], and flamed more than men [t(341)=6.45, p=.0000]. A table summary of results appears in Witmer and Katzman (1997).

Results of the factorial ANOVA on GAs indicated significant main effects for flaming [F(1,335)=36.724, p.=.0000] but not for gender [F(1,335)=3.361, p=.068] or challenge [F(1,335)=2.857, p=.092]. It also revealed a two-way interaction between gender and flaming [F(1,335)=8.950, p<.001] and a three-way interaction between gender, flaming, and challenge [F(1,335)=7.041, p=.0000].

Discussion

The first hypothesis, that females use GAs more than males, is partially supported by the data. Neither males nor females use GAs to any great extent, as indicated by GAs being identified in only 396 of the 3,000 messages--13.2 percent of the total sample. The results indicate, however, that the computer users who do include GAs in their discourse tend to be women. The use of articons and emoticons primarily by females suggests that the aesthetic quality reported in Mulac, Bradac, and Mann (1985) and Mulac and Lundell (1986) and (possibly) expression of emotion translates to the computer-mediated environment.

The data do not support the second and third hypotheses, that men use more challenging language and flame more often than women. Although the literature indicates that men might be more likely to flame or challenge, the opposite is true in this sample population. The unexpected difference between women and men in flames and challenges may be a result of a status-leveling effect of CMC (Sproull and Kiesler, 1991). One explanation for this might be that women feel more at ease in the relatively anonymous electronic environment. Another possibility is that the women who currently engage in CMC already are involved in male-dominated endeavors such as high-tech organizations and academia. These women may not represent the discursive praxis of women in general nor their places in a still male-dominated power structure. Finally, it is possible that if women are more expressive of any emotion than men, a comparison between their uses of supportive or equivocal language also will indicate significant differences. Since the interrater reliability for apology was not high enough to include it in this study, we don't know if women apologize as well as challenge in CMC.

This study has several limitations. First, because of the way in which the codebook and method for ProjectH evolved, the data do not include codings for some variables that might inform this study. Second, while a content analytic approach has the advantage of being unobtrusive and permitting the collection of a large sample, it also has the disadvantage of forcing the researcher to guess or infer why some phenomena occur (e.g., are emoticons used to attenuate or emphasize message content; are they simply manifestations of playfulness; do they serve as a form of shorthand; are they substitutes for writing skills?). Third, interrater reliability remains problematic. While many of the paired codings seemed respectably congruent, the unprecedented ProjectH collaboration presents new methodological challenges.

In spite of the limitations, this study raises some questions about the ways in which women and men communicate in the computer-mediated environment. Why do women tend to challenge and flame in this group of CMC users, rather than use the supportive and qualifying language we expect in face-to-face settings? What purpose, beyond being cute, does the smiley serve? How do keyboard, writing, and computer skills affect the use of graphic accents? It seems clear that the information superhighway is paving a road to new forms of discourse, offering communication scholars a rich field for study.

References

About the Author

Diane F. Witmer (PhD, MA, MS, University of Southern California, BS, University of La Verne) is an assistant professor of communication at Purdue University. Her practical experience includes both corporate and not-for-profit public relations, and she holds Public Relations Society of America (PRSA) accreditation. Dr. Witmer's research interests include computer-mediated communication and organizational communication, and she has been instrumental in the development and maintenance of a several World Wide Web sites. She is an active member of the International Communication Association, Speech Communication Association, Central States Communication Association, Public Relations Society of America, and Communication Institute for Online Scholarship.
Address: Department of Communication, Purdue University, Liberal Arts & Education Building, 2114, West Lafayette IN 47907-1366, USA.

Sandra Katzman, an independent journalist and communication researcher, teaches English in Japan. She is a pioneer of the University of California at Santa Cruz (BA cum laude English Literature), and studied social science at Stanford University (MA Communication). Her editorial opinion articles appear in The Los Angeles Times and The Sacramento Bee.
Address: Interac Co. Ltd, Fujibo Bldg, 2-10-28 Fujimi, Chiyoda-ku, Tokyo 102, Japan.