Empirical Methods in CommunicationApplication Assignment #1
Experimental Research
Overview: You will briefly answer five essay questions that require you to dissect an
experiment by Palomares and Lee, that was published in a 2010 issues of the Journal of
Language and Social Psychology. Your answers must be typed and double spaced. Provide
a 1 inch margin on all sides of the page for written comments. Your paper must absolutely not
be longer than 1000 words; that is about 200 words per question. All word processors have a
word count feature. Please provide the word count of your essay (excluding your name and
other identifying information).
Due Date: May 10. Turn your paper in at the beginning of the class session. Electronic (email
attachment) submissions will not be accepted.
Questions
1. Palomares and Lee used a 2x2x2 between-subjects factorial design. (This kind of design
is referred to as between-subjects because each research participant was assigned to
only one experimental condition.) Explain why this design was necessitated by the
research hypotheses.
2. Experimental control is a critical feature of any experiment. If we wish to know the effects
of independent variables on dependent variables we must hold constant other variables
that would make causal inference challenging. What efforts were taken in this experiment
to achieve a high level of experimental control?
3. What efforts were made to determine if the manipulations of the independent variables
were valid?
4. Did this experiment measure any potential mediating variables? If so, what were these
variables? If no, why not?
5. The generalizeability of every experiment has limitations, and this study is no exception.
Briefly describe how our ability to generalize from the findings obtained are limited.
Evaluation
Your answers to these five questions will be equally weighted. Evaluation is based on the
thoughtfulness of your response and the quality of your writing.
Think of this essay as a take home exam. The work you submit must be your own. You may not
discuss this assignment or your ideas with other people. Doing so will be considered cheating.
Articles
Journal of Language and Social Psychology
29(1) 5
23
2010 SAGE Publications
DOI: 10.1177/0261927X09351675
http://jls.sagepub.com
Virtual Gender Identity:
The Linguistic Assimilation
to Gendered Avatars
in Computer-Mediated
Communication
Nicholas A. Palomares1 and Eun-Ju Lee2
Abstract
This research examined how individuals gendered avatar might alter their use of
gender-based language (i.e., references to emotion, apologies, and tentative language)
in text-based computer-mediated communication. Specifically, the experiment tested
if men and women would linguistically assimilate a virtual gender identity intimated
by randomly assigned gendered avatars (either matched or mismatched to their
true gender). Results supported the notion that gender-matched avatars increase
the likelihood of gender-typical language use, whereas gender-mismatched avatars
promoted countertypical language, especially among women. The gender of a partners
avatar, however, did not influence participants language. Results generally comport with
self-categorization theorys gender salience explanation of gender-based language use.
Keywords
gender-linked language, social identity, intergroup communication, message production,
stereotypes, prototypes
Gender-based communication is the focus of much scholarship. This work increasingly
emphasizes computer-mediated environments. Research, for example, has
examined how men and women communicate via e-mail (Colley & Todd, 2002), chat
groups (Koch, Mueller, Kruse, & Zumbach, 2005; Thomson, 2006), instant messages
1University of California, Davis, Davis, CA, USA
2Seoul National University, Gwanak-gu, Seoul, Korea
Corresponding Author:
Nicholas A. Palomares, Department of Communication, One Shields Avenue, University of California,
Davis, CA 95616, USA
Email: [email protected]
Downloaded from jls.sagepub.com at UNIV CALIFORNIA DAVIS on March 18, 2013
6 Journal of Language and Social Psychology 29(1)
(Fox, Bukatko, Hallahan, & Crawford, 2007), and other forms of computer-mediated
communication (CMC). One issue in this empirical arena is gender identity and its
manifestations in CMC. Scholars have argued, for example, that people perform masculinity
online as a means to reify their gender identities (Herrmann, 2007). Other
research has demonstrated that elevating the salience of gender identity prompted
women to reference emotions in e-mail more than men especially in mixed-sex interactions
(Palomares, 2008). Permeating this literature is a focus on the diverse, dynamic,
and sometimes transient nature of gender identity; how it differentially presents itself
in CMC given the circumstances; and the resultant communicative behavior of men
and women (Murachver & Janssen, 2007; Palomares, Reid, & Bradac, 2004).
Whereas the primary concern of this work is how a sex-consistent gender identity
affects communication, a relatively nascent interest is how people simulate a gender
identity online that they would not otherwise perform in offline settings (Herring &
Martinson, 2004; Hills, 2000; Rellstab, 2007). For example, a woman might pretend to
be a man in an online chat. The few instances of this research have studied strategic or
intentional portrayals of a different gender and focused on the communicative behaviors
people employ in these forgeries and if others can recognize a disingenuous gender
identity (Herring & Martinson, 2004; Hills, 2000; see also Thomson & Murachver,
2001); yet no known research has examined how more subtle cues might trigger the
enactment of a different gender identity online. We refer to this phenomenon as virtual
gender identity. Thus, we conducted an experiment to test if men and women would
linguistically assimilate a virtual gender identity intimated by (matched or mismatched)
gendered avatars representing them in text-based CMC. Specifically, our objective was
to determine if and how mens and womens gender-based language would emerge as a
function of gendered (i.e., masculine or feminine) avatars that represented them and
their interaction partner. In pursuit of this goal, we first review research on language
and gender, then present our theoretical orientation from which we deduce predictions,
and finally report an experiment that implemented and tested this rationale.
Language and Gender in CMC
Research traditionally has emphasized gender differences claiming that men and
women tend to use dissimilar language independent of the context, personal proclivities,
or interaction partners (e.g., Lakoff, 1975; Mulac & Lundell, 1980; Tannen,
1990). The empirical evidence is somewhat compatible with this claim. Consistent
with stereotypes, for example, meta-analyses demonstrated that women used more
affiliative speech (e.g., references to emotion) and less assertive speech (e.g., direct
language) than men (Leaper & Ayres, 2007). Over time, however, the focus has shifted
away from gender and onto alternative (i.e., extragender) influences, especially those
present in CMC. Whether in e-mail, newsgroup postings, blogs, discussion groups,
online chats, or other computerized settings, the language of men and women largely
depends on the specific circumstances and features of the technology and context (e.g.,
Colley & Todd, 2002; Fox et al., 2007; Herring, 1993; Huffacker & Calvert, 2005;
Palomares, 2004, 2008, 2009; Savicki, Kelley, & Ammon, 2002; Thomson, 2006). In
Downloaded from jls.sagepub.com at UNIV CALIFORNIA DAVIS on March 18, 2013
Palomares and Lee 7
fact, the same aforementioned meta-analyses also found several factors that moderated
the gender effects often to an extent greater than gender alone (Leaper & Ayres,
2007). Language differences between men and women, thus, clearly exist, but they are
highly sensitive to extraneous factors that may increase, decrease, erase, or even reverse
the traditional gender-based patterns of use.
The emergence of three language featuresreferences to emotion, apologies, and
tentative languagehas been particularly vulnerable to contextual instability within
and across studies despite stereotypes and early conjectures that they are feminine
language forms. References to emotion, or language that includes any mention of a
feeling or emotion, have been indicted as typically associated with womens language
(Mulac, Bradac, & Gibbons, 2001). Yet other research has shown that men reference
emotion more than women (Mulac, Seibold, & Farris, 2000), that men and women use
them equally (Thomson, 2006), and that their use depends on the salience of gender
identity and dyadic sex composition (Palomares, 2008). Examinations of apologies
which some have construed as an indicator of politeness and a feminine language
style (Herring, 1993; Lakoff, 1975)have yielded a similarly diverse array of differences
and similarities between men and women (ONeill & Colley, 2006; Savicki,
Lingenfelter, Kelley, 1996; Tannen, 1990; Thomson, 2006). Tentative language signals
uncertainty, is typically associated with women (Herring, 1993; Lakoff, 1975), and like
apologies and references to emotion is contextually dependent (Brouwer, Gerritsen, &
De Haan, 1979; Carli, 1990; Palomares, 2008, 2009; S. A. Reid, Keerie, & Palomares,
2003; Tannen, 1990). We examined these three features because research frequently
employs them in CMC as stereotypically gender-based language forms.
Self-Categorization Theory
Notwithstanding inconsistent results among the three language features, an explanation
for the diverse collection of gender-based language manifestations is found in selfcategorization
theory (Turner, Hogg, Oakes, Reicher, & Wetherell, 1987).1 The basic
premise of the theory is that people mentally represent social groups as contextually
contingent prototypes or fuzzy sets of attributes that define in-group similarities in
contrast to out-group differences. People internalize the group prototype that is most
salient and relevanta state called depersonalized. Prototypes operate not only to
describe but also to prescribe, such that depersonalization provides a normative selfdefinition
for how one should perceive and behave in a certain context.
When applied to gender and language phenomena (cf. Palomares et al., 2004), the
theory maintains that if people interact devoid of a gender distinction, then ones gender
is irrelevant and gender-based language is unlikely to emerge; but if a gender categorization
is germane, then gender identity is applicable to ones self-construal, and people
will behave according to the activated prototypical norms (Palomares, 2008;
S. A. Reid et al., 2003). Gender-relevant interactions, thus, increase the salience of
gender identity so that the prototype of intergender relations has significant consequences
for language use. Self-categorization theory has been relatively successful in
attempts to explain and predict a diverse array of linguistic behavior for men and women
Downloaded from jls.sagepub.com at UNIV CALIFORNIA DAVIS on March 18, 2013
8 Journal of Language and Social Psychology 29(1)
in CMC. When sending an e-mail, for example, women referenced emotions significantly
more than men only if gender was salient because the prototype of gender salience
exploited supportiveness as a stereotypically feminine attribute (Palomares, 2008). We
formulated our expectations for the experiment based on self-categorization theory.
Performing Virtual Gender Identities
A limited number of studies have examined the online performance of a different gender.
The earliest scholarship highlighted intentional gender swapping on the Internet
(e.g., a man posing as a woman) and documented and described its natural occurrence.
People gender swap, for example, in text-based multiuser dungeons and similar online
groups for a range of reasons (Berman & Bruckman, 2001; Bruckman, 1993; Danet,
1996; Donath, 1999; McRae, 1995; Menon, 1998; Rheingold, 1993; Turkle, 1995; Van
Gelder, 1996). Assuming a different virtual gender identity has several sociological
and psychological implications (Herrmann, 2007; Kendall, 2000; E. M. Reid, 1991,
1995; Rellstab, 2007; Rodino, 1997) especially considering that a substantial portion
(40% to 60%) of online social-site members typically do so for some of their time
online (Roberts & Parks, 1999). Relatedly, Internet users can strategically ambiguate
their gender often via gender-neutral pseudonyms (Bechar-Israeli, 1995; Van Gelder,
1996). Gender equivocation, however, is more common among women than men
(Jaffee, Lee, Huang, & Oshagan, 1995; Jazwinski, 2001), likely because it assuages
gender biases that can occur in face-to-face interactions (Flanagin, Tiyaamornwong,
OConnor, & Seibold, 2002; Koch et al., 2005). Research has also examined the detection
of real (Koch et al., 2005; Nowak, 2003; Thomson & Murachver, 2001) and false
(Herring & Martinson, 2004; Hills, 2000) gender identities in CMC.
Whereas most research on virtual gender identities has recorded its natural occurrence,
objectives, implications, and detection, recent examinations have studied the
communicative behaviors people manipulate when intentionally performing a false
gender. Such research has found that people seem to have control over macro forms of
communication (e.g., topic) more than molecular forms (e.g., tentative language). For
example, if told to pose as a different gender when interacting with an unknown partner
via e-mail, participants typically exploited gender-stereotypical topics while
having relatively little control over gender-typical syntactic and lexical choices (Hills,
2000). Likewise, in synchronous CMC, people successfully altered their topical content
when intentionally performing a different gender but ineffectively changed
molecular forms of communication; in fact, their molecular features actually gave
cues to their true gender despite their effective topic manipulations (Herring &
Martinson, 2004). Our experiment advances past research by not overtly instructing
people to communicatively perform a different gender identity. Instead, we manipulated
gendered avatars to test if people would automatically assimilate their language
to a virtual gender identity without explicit direction to do so.
A gendered (i.e., masculine or feminine) avatar can heighten the salience of gender.
Avatars are graphical self-representations in a computer-mediated environment that can
Downloaded from jls.sagepub.com at UNIV CALIFORNIA DAVIS on March 18, 2013
Palomares and Lee 9
reveal social information in an otherwise cue-limited setting (Blascovich et al., 2002).
Interacting via avatars, for example, can impart levels of trust and intimacy similar to
an audiovideo mode of mediated communication but more than text-only communication
(Bente, Rggenberg, Krmer, & Eschenburg, 2008). Gender inferences of
anonymous others depend on their avatars even if avatar representations are known to
be arbitrary (Lee, 2007a). People prefer avatars that closely represent themselves over
less accurate digital representations, especially in terms of gender (Nowak & Rauh,
2005). In fact, avatars have behavioral consequences by inducing avatar-consistent communication:
In line with attractiveness stereotypes (cf. Langlois et al., 2000), intimacy
(e.g., self-disclosures) was greater for people represented by attractive than less attractive
avatars (Yee & Bailenson, 2007, Experiment 1). Likewise, in a second study that
capitalized on confidence stereotypes of tall people (cf. Young & French, 1996), participants
who assumed an avatar taller than their negotiation partners avatar were more
likely to decline their partners unfair offer than if their avatar was shorter. Given that
people heed avatars cognitively and behaviorally, a gendered avatar might affect
gender-based language because it yields a gender self-definition germane. According
to self-categorization theory, however, these linguistic consequences would depend
on the nature of the avatar and its ramifications for gender salience: A gendered selfrepresentation
in CMC will intimate the prototype for gender-based linguistic behavior.
Specifically, masculine avatars will implicate male-linked language norms, whereas
feminine avatars suggest female-typical language norms. As a result, people linguistically
assimilate to these communicative norms.
These effects, however, are likely more robust for women than men. Women tend
to be more responsive to gender salience than men are (Palomares, 2008; S. A. Reid
et al., 2003), and they tend to identify with their gender more strongly than men do
(Cameron & Lalonde, 2001). In fact, men were less likely than women to take a gendered
avatar into account when inferring an anonymous partners gender (Lee, 2007a).
Women also are more accurate when decoding others nonverbal communication and
are generally more sensitive to it than men are (Hall, 2006). Because women tend to
be particularly reactive to visual communicative stimuli and gender salience, we
expect a woman to use more stereotypically feminine language when her avatar is
consistent (i.e., feminine avatar) than inconsistent (i.e., masculine avatar) with her
true gender; yet the effect of this corresponding pattern for men will likely be less
extreme if it manifests at all. Thus, we present the following:
Hypothesis 1a-c: Women, but not men, use more gender-typical language
(a) references to emotion, (b) apologies, and (c) tentative languagewhen
the gender of their avatar matches their true gender than when it mismatches.
We also tested the effects of a CMC partners gendered avatar because it too can
play an influential role in computerized interactions. In most circumstances, gender
differences are more likely in intergroup (i.e., mixed-sex) than intragroup (i.e., same-sex)
interactions. For example, women referenced emotion more than men when gender
Downloaded from jls.sagepub.com at UNIV CALIFORNIA DAVIS on March 18, 2013
10 Journal of Language and Social Psychology 29(1)
was salient but chiefly in mixed-sex e-mail exchanges (Palomares, 2008); likewise,
gender differences in tentative language were present in intergroup but not intragroup
CMC (Palomares, 2009). Self-categorization theory accounts for such effects by
arguing that mixed-sex interactions render an intergender distinction more pertinent
than same-sex settings do, such that assimilation to the prototype of gender salience
becomes more likely (Hogg & Turner, 1987). We, therefore, might expect a partners
gendered avatar to affect gender-based language as well, which is analogous to other
research revealing partner-avatar effects for nongender groups. People in a virtual
environment, for example, maintained greater distance when encountering an avatar
of an ethnic minority than an avatar of an in-group member, especially if they held
implicit prejudice toward the minority out-group (Dotsch & Wigboldus, 2008).
Precisely predicting how anothers gendered avatar might interact with a gendered
graphical self-representation, however, is difficult because what constitutes mixed
sex is muddled when gendered avatars are introduced in CMC to represent anonymous
interactants. That is, whether people compare their true or virtual gender with
their partners gendered avatar can alter their inter-/intragroup determination. For
example, a woman who is represented by a masculine avatar when interacting with a
partner using a feminine avatar might consider the interaction to be intergroup if she
contrasts her and her partners avatars; whereas if she compares her partners avatar
with her actual gender, then she might conclude that the interaction is intragroup. In
fact, Lee (2007b) found that dyadic team members felt stronger group identification
when their avatars belonged to the same gender category (rather than different categories).
Such results suggest that perceptually salient, albeit explicitly arbitrary,
avatars can serve as a formative basis for an intra-/intergroup distinction. Nonetheless,
if and how self-other avatar comparisons have effects beyond fostering group
cohesion remains unclear in Lees study; that is, even when participants thought My
partner and I are similar, by virtue of the similar avatars, they might not have fully
embraced the specific identity represented in the avatars (We are both masculine),
especially considering that their avatars gender always mismatched their true gender
in the study. By examining social perceivers linguistic behavior as a function of their
own and their partners gendered avatars, the present study extends past work. Yet
given the difficulty gendered avatars present for ascertaining the inter-/intragroup
nature of an interaction in anonymous CMC, we ask this research question:
Research Question 1: Does the gender of a partners avatar influence (via either
main or interaction effects) gender-based language use?
Method
Participants and Design
Participants were 157 undergraduates (74 men, 83 women) enrolled in communication
classes at a large, West Coast university. A 2 (participants gender: men vs. women)
Downloaded from jls.sagepub.com at UNIV CALIFORNIA DAVIS on March 18, 2013
Palomares and Lee 11
2 (gender matching of participants avatar: match vs. mismatch participants true
gender) 2 (gender of partners avatar: male vs. female) between-subjects design was
employed wherein participants completed a trivia game with an ostensible partner
both of whom were represented via gendered avatars in synchronous text-based CMC.
Avatar Manipulations
Two masculine and two feminine avatars manipulated self and partner representations.
An additional 50 undergraduate students (66% women) participated in a pretest to confirm
an effective manipulation of avatar gender. Participants first saw one of the four
cartoon characters and then indicated how feminine and masculine the character was on
10-point scales (1 = not at all masculine/feminine, 10 = very much masculine/feminine).
The femininity rating was reverse coded and then combined with the masculinity rating
to form a femininitymasculinity index (r = -.86, p < .001; range: 2-20). A 2 (participant gender) 2 (avatar gender) analysis of variance (ANOVA) established that male characters were considered to be more masculine (M = 14.24; SD = 3.71) than female characters (M = 5.16; SD = 1.95), F(1, 46) = 113.91, p < .001, hp 2 = .71. Furthermore, one-sample t tests revealed that the attribution of masculinity to male characters was significantly greater than the scale midpoint (11.00), t(24) = 4.37, p < .001, whereas female characters were perceived as significantly less masculine (or more feminine) than the scale midpoint, t(24) = -14.97, p < .001. There was no interaction between participants and avatars gender, indicating that both men and women perceived the avatars gender as intended, F < 1. The four avatars served to randomly manipulate participants avatar gender and the partners avatar gender. A participants avatar was never identical to his or her ostensible partners avatar in the main experiment. Procedure Participants played a computerized trivia game with someone whom they believed to be another study participant. To reduce participants suspicion about the purpose of the experiment, they were first asked to choose a letter on the computer screen, ranging from A to E, to determine the avatar (i.e., cartoon character) that would represent them during the interaction. Unbeknownst to the participants, however, the characters gender was randomly predetermined to be either male or female regardless of their true gender and the chosen letter. Once the participants avatar and their ostensible partners avatar appeared on the computer screen, participants selected a number, ranging from 1 to 10, to determine a set of questions to be asked during the game. Regardless of the number chosen, however, the computer presented a fixed set of fastfood trivia. For each multiple-choice question, participants indicated their initial answer and confidence level and typed a comment to their partner. After participants typed a comment, the participants and the partners characters showed their initial responses, as illustrated in Figure 1. The partners responses were preprogrammed and held constant across conditions, and their comments contained no apologies, tentative Downloaded from jls.sagepub.com at UNIV CALIFORNIA DAVIS on March 18, 2013 12 Journal of Language and Social Psychology 29(1) language features, or references to emotion (e.g., I have no clue, D seems too obvious). At this point, participants submitted their final answer and confidence level, after which the computer presented the next question without revealing the correct answer or the partners final response to the previous question. This procedure was repeated for 12 unique questions that were held constant across all conditions. Finally, participants were debriefed. Language Coding The comments that participants wrote to their ostensible partner during the trivia game served as the source of gender-based language use. All comments formed a transcript booklet with only a unique number identifying each participants transcript. Two research assistants, who were blind to the design and hypotheses, underwent training sessions where they learned definitions for, saw several examples of, and practiced coding each language feature. Once well-trained and pretested for sufficient reliability, the assistants individually coded all language features one at a time and then settled disagreements via postcoding discussions. Across all language features the coders agreed at a rate of at least 87% (Krippendorffs as > .90).
The operationalizations of the three language variables were modeled after past
language and gender research (Palomares, 2008; S. A. Reid et al., 2003; Thomson &
Murachver, 2001). References to emotion were any mention of an emotion (e.g.,
happy, that should thrill you, mad, excited). Apologies were defined as a statement of
being sorry (e.g., Im sorry, forgive me, I was wrong and wont let it happen again).
Tentative language was defined as the combination of three unique language features
that indicate uncertainty and low confidence: hedges (e.g., might, pretty much, sort of,
Figure 1. Sample screen snapshot of avatars with responses: Participant with a male avatar
and a female partner character
Downloaded from jls.sagepub.com at UNIV CALIFORNIA DAVIS on March 18, 2013
Palomares and Lee 13
maybe, probably), disclaimers (e.g., dont trust me, but Im not sure, I may be wrong,
who knows though) and tag questions (e.g., dont you think? isnt it? right?).
Results
Pretest
To ensure that the experimental task did not overtly favor one gender, we used fastfood
trivia whose gender neutrality was confirmed in previous studies (Lee, 2005).
Specifically, when asked to indicate how interested they were in the fast-food questions
(1 = not at all interested, 10 = very much interested), men (M = 2.89; SD = 2.42) and
women (M = 2.20; SD = 2.00) did not significantly differ, t(110) = 1.66, p = .10 (Lee,
2005, Study 1). In addition, participants directly rated how gender biased they thought
the questions were (1 = not at all gender biased, 10 = very much gender biased), and
the mean (M = 3.81; SD = 1.88) was significantly lower than the scale midpoint (5.5),
t(75) = -7.80, p < .001 (Lee, 2005, Study 3). Hypothesis Tests A series of 2 (participant gender) 2 (participant avatar) 2 (partner avatar) ANOVAs was computed for (a) references to emotion, (b) apologies, and (c) tentative language. One-tailed a priori contrasts tested any hypothesized differences (as indicated), whereas two-tailed tests compared conditions when a difference was not expected or when a possible difference was not explicitly hypothesized (Tabachnick & Fidell, 2007). Figure 2 displays the pertinent results. References to emotion. A significant interaction emerged between participants gender and self-representation avatar, F(1, 149) = 5.64, p = .02, hp 2 = .03. No other effects were statistically significant, all Fs < 1. Participants avatar had a greater impact for women than men, which is consistent with Hypothesis 1a: Women used more references to emotion when the character correctly represented their gender (M = .84; SD = .99) than when it did not (M = .44; SD = .82), t(153) = 1.95, one-tailed p = .03, hp 2 = .02; yet mens references to emotion did not significantly vary across the male (M = .54; SD = .82) and female (M = .87; SD = 1.10) avatars, t(153) = 1.50, p = .14. When the interaction was decomposed within the self-representation conditions, gender differences were more pronounced in the mismatched than matched avatar condition. If participants characters gender mismatched their true gender, then men used more emotional references than did women, t(153) = 2.04, p = .04, hp 2 = .03; when the avatar correctly represented their gender, women tended to reference emotions more frequently than men, but this difference was not statistically significant, t(153) = 1.39, p = .17. Apologies. We found a significant interaction between participants gender and avatar for apologies, F(1, 149) = 4.11, p = .04, hp 2 = .03. No other effects were statistically significant, all Fs < 1. Hypothesis 1b received tentative support: Women were Downloaded from jls.sagepub.com at UNIV CALIFORNIA DAVIS on March 18, 2013 14 Journal of Language and Social Psychology 29(1) more apologetic when the avatar matched their gender (M = .16; SD = .37) than when it mismatched (M = .05; SD = .22), t(153) = 1.56, one-tailed p = .06, hp 2 = .02; whereas mens apologies did not statistically significantly differ across the two conditions (match: M = .03, SD = .17; mismatch: M = .13, SD = .41), t(153) = 1.36, p = .18. Within the matched self-representation condition, women used more apologies than men, although this effect did not reach statistical significance, t(153) = 1.83, p = .07, hp 2 = .02. The same gender difference with mismatched avatars was not statistically significant, t(153) = 1.08, p = .28. Tentative language. There were no significant main or interaction effects on tentative language use, all Fs < 1.84. Even though the interaction between gender and self-representation failed to reach statistical significance, F(1, 149) = 1.75, p = .18, we still tested Hypothesis 1c because omnibus tests are dispensable when specific predictions exist (Rosenthal, Rosnow, & Rubin, 2000; Wilkinson & Task Force on Statistical Inference, 1999). Supporting Hypothesis 1c, women were more tentative when a female avatar matched their true gender (M = 1.20; SD = 1.25) relative to a mismatched male character (M = .69; SD = 1.08), t(153) = 2.04, one-tailed p = .02, hp 2 = .03. In contrast, mens tentative language use was identical when the character either correctly (M = 1.00; SD = .97) or incorrectly (M = 1.00; SD = 1.21) represented their true gender. Comparing men and women within each self-representation condition, however, yielded no significant effects, both ts < 1.20. Figure 2. Effects of matched versus mismatched gendered avatars on gender-based language use for men and women 0.54 0.84 0.03 0.16 1.00 1.20 0.87 0.44