Année : 2019
Lieu de publication de l'article :

Résumé de l'article

There has been long-standing stereotypes on men and women’scommunication styles, such as men using more assertive or ag-gressive language and women showing more agreeableness andemotions in interactions. In the context of collaborative learning,male learners often believed to be more active participants whilefemale learners are less engaged. To further explore gender dif-ferences in learners communication behavior and whether it haschanged in the context of online synchronous collaboration, weexamined students interactions at a sociocognitive level with amethodology called Group Communication Analysis (GCA). Wefound that there were no significant differences between men andwomen in the degree of participation. However, women exhibitedsignificantly higher average social impact, responsivity and inter-nal cohesion compared to men. We also compared the proportionof learners interaction profiles, and results suggest that womenare more likely to be effective and cohesive communicators. Wediscussed implications of these findings for pedagogical practicesto promote inclusivity and equity in collaborative learning online.CCS CONCEPTS• Applied computing → Distance learning; E-learning; Col-laborative learning;KEYWORDSCommunication behavior, gender differences, collaborative learn-ing, learner characteristics, interaction profileACM Reference Format:Yiwen Lin, Nia Dowell, Andrew Godfrey, Heeryung Choi, and ChristopherBrooks. 2019. Modeling gender dynamics in intra and interpersonal interac-tions during online collaborative learning . In The 9th International LearningPermission to make digital or hard copies of all or part of this work for personal orclassroom use is granted without fee provided that copies are not made or distributedfor profit or commercial advantage and that copies bear this notice and the full citationon the first page. Copyrights for components of this work owned by others than ACMmust be honored. Abstracting with credit is permitted. To copy otherwise, or republish,to post on servers or to redistribute to lists, requires prior specific permission and/or afee. Request permissions from permissions@acm.org.LAK19, March 4–8, 2019, Tempe, AZ, USA© 2019 Association for Computing Machinery.ACM ISBN 978-1-4503-6256-6/19/03. . . $15.00https://doi.org/10.1145/3303772.3303837Analytics & Knowledge Conference (LAK19), March 4–8, 2019, Tempe, AZ, USA.ACM, New York, NY, USA, 5 pages. https://doi.org/10.1145/3303772.33038371 INTRODUCTIONTechnological advances have enabled individuals from differentbackgrounds to work or study collectively regardless of physi-cal boundaries. Computer-mediated communication (CMC) at thesame time, has emerged as a means to enhance students’ explo-ration and understanding of the subject matter [14]. Research hasshown that computer-mediated collaboration may increase thelevel of achievement and knowledge retention, inducing meaning-ful problem-solving that enhances learning [18]. Therefore, it isimportant to explore the dynamics of group interactions in thecontext of online collaboration in order to achieve better learningoutcomes.To capture a better understanding of learners’ interactions pat-terns, it is necessary to discuss how individual characteristics areassociated with types of communication behavior. In the educa-tional context, the knowledge of how learners interact differentlydue to demographic background such as gender, age, and ethnicitycan provide useful information for creating instructional designsthat reduce biases and promote equity in the Internet-based learn-ing environment[2].With the prevalent use of learning technologies, other individualcharacteristics such as computer competency [17], social presence[15], and synchronicity in interactions [3] are taken into account inthe research on Computer Supported Collaborative Learning (CSCL)and Computer-Mediated Communication (CMC) more broadly. Al-though gender differences have been extensively studied in rela-tion to learning performance in traditional classroom settings, thedistinctions have been less well-explored in computer-mediatedlearning. Our research contributes to the field of learning analyticsby providing novel insights into gender differences in learner be-haviors. Specifically, we investigate learner interaction patterns ina small-group, synchronous online collaboration in higher educa-tion, with an analytic method that captures multiple sociocognitivedimensions.431LAK19, March 4–8, 2019, Tempe, AZ, USA Yiwen Lin, Nia Dowell, Andrew Godfrey, Heeryung Choi, and Christopher Brooks2 PREVIOUS WORK2.1 Gender Differences in Online DiscoursePrevious research on gender differences with respect to communica-tion has suggested that men tend to be more dominant and directivewhile women are more expressive and cooperative in face-to-faceinteractions [1]. Herring suggested [9] interaction dynamics of thisnature reflect traditional gender expectations that have carried overto the digital space. Indeed, there is evidence to suggest that menare more likely to use authoritative languages while women tendto make more personal statements in CMC [8]. However, recentfindings on communication behavior point to patterns that con-tradict gender stereotypes. In Tsai’s study [21], women are foundto be generally more active participants than men in online dis-cussions. Moreover, Sullivan et al. [19] observed male’s employedstereotypically “feminine” discourse style, while female’s resembled“masculine” discourse. Evidence as such provides a different perspec-tive on understanding interpersonal dynamics during collaborativeCMC.2.2 Research MotivationThere are some notable concerns, and areas of improvement inprevious research efforts that should be addressed in order to moveforward as a field towards our common goal of fostering under-standing and creating gender equity in the digital learning com-munity. First, many methodological approaches employed havebeen surface level, wherein a large proportion of research devotedto examining gender differences in the patterns of interaction hasrelied on measuring frequencies of message and length of texts [16].In recent years, there has been an increased amount of attentiongiven to content and automated discourse analysis on interactions,such as the positive and negative valence of messages, agreementlevel, and emotional expressiveness [13]. Typically, these analysesare applied at the student level (i.e. individual posts or totality ofthem per person). Aggregating the text of the individual or a threadoffers a cumulative account of female and male learners’ discourse,but it provides only coarse-level granularity, and disregards thesociocognitive processes that reside in the interaction betweenlearners’ discourse contributions. Unlike aggregated text analysismeasurements, semantic analysis captures a deeper sociocognitiveprocesses in communication[5], and shows greater potential tobring forward novel discoveries in understanding gender differ-ences in collaboration.Second, the language and discourse that can be captured andexplored in the context of observed female and male differenceshas been restricted. For instance, previous studies on gender dis-crepancies show an emphasis on assessing intrapersonal interac-tions properties, including the style of speech, individual attitudes,perceived satisfaction for interaction[4]. This focus on intraper-sonal properties, has resulted in the field having limited knowledgeabout more complex mechanisms at play during collaborative learn-ing, and CMC more broadly. For example, one may ask whethera learner’s contribution is likely to be responded to by anotherlearner (i.e.social impact), and how this might differ between malesand females, or how social regulatory communication processes,such as monitoring and responding to peers presents a differentialpattern between genders. Marks et al. [12] argue that multidirec-tional interactivity should be investigated in order to determineappropriate pedagogical methods and class activities. Therefore, afine-grained instrument which captures multiresponsive dynamicsfrom a sociocognitive perspective could provide valuable insightsin facilitating small group collaborations online. In this paper, weaddress these limitations by using a new, robust methodology called,Group Communication Analysis (GCA) (described in the followingsection).Lastly, the importance of learner roles in collaborative learninghas been recognized, particularly in relation to their significantinfluence on group processes and outcomes. Hoadley [10] acknowl-edges the ability of roles in reflecting group and individual leveldynamics in moment-to-moment interactions. The concept of roleshighlights individual differences and the emergence of group mem-bership based on learner interaction patterns. However, there is ascarcity of research which has combined, and investigated thesetwo important interaction features, namely i) Gender, and ii) learnerinteraction profiles. In this paper, we aim to investigate how genderis associated with emergent roles in small group interactions, to pro-vide stronger evidence for the effect of gender on learners’ commu-nication behavior. To avoid confusion, we will use the term“learnerinteraction profile” instead of “roles” in the subsequent discussion.3 CURRENTWORK3.1 GCA MeasureOriginally developed by Dowell et al. [6], Group CommunicationAnalysis (GCA) is a quantitative methodology which can be used tounderstand the dynamics of group communication. This approachapplies computational linguistic techniques to detect multiple so-ciocognitive dimensions present in group interaction (Table 1). GCAused latent semantic analysis (LSA) to calculate the relatedness oftext data, so as to infer the semantic relationships of individual con-tributions in group interactions [6]. More comprehensive detailsare described in the original article of Dowell et al.GCA provides a framework to investigate micro intra- and in-terpersonal processes during online multi-party interactions. Thisallows us to specifically focus on different gender patterns of inter-actions (i.e. active or passive, leading or following), contributioncharacteristics (i.e. providing new information or echoing givenmaterial), and social orientation (i.e. individual or group). In pre-vious work, GCA has also categorized learner profiles based ondifferent combination of interaction characteristics, using clusteranalysis. Each cluster were then labeled with six different roles(Figure 1 2), after comparing the cluster patterns to learner rolesthat are presented in previous literature [6]. Figure 1 and 2 providefurther information on how each profile was characterized acrossthe six GCA properties. Since the sample in the current research isthe same as that of the original study, we are able to explore genderdifferences using existing clustered data and identified profiles.In this paper, we aim to extend previous research by using theGCA measure to capture the semantic relationship between learn-ers’ comments over time. In particular, we contribute measures ofthe qualities of conversations, relatedness of learners contributions,and relate these to learner gender identities. Some unique proper-ties such as responsivity and social impact, will allow us to examine432Modeling Gender in Intra and Interpersonal Dynamics LAK19, March 4–8, 2019, Tempe, AZ, USATable 1: Descriptions of GCA measuresGCA Dimensions DescriptionParticipation Mean participation of an individual relative to the expected average of the group of its sizeInternal cohesion How consistent an individual is with their own recent contributionsResponsivity The tendency of an individual to respond, or not, to the contributions of their collaborative peersSocial Impact The tendency of a participant to evoke corresponding responses from their collaborative peersNewness Whether one is likely to provide new information or to echo existing informationCommunication density The extent to which participants convey information in a concise mannerthe degree of interpersonal influence correlated to gender. Lastly,whether we could find association of gender with learner profilesalso provides stronger evidence of gender divergences in commu-nication patterns in online collaboration. Using this approach, wewere able to address the following research questions:(1) What are the gender differences in the amount and type oflearner contributions?(2) Is gender associated with learner interaction profiles in smallgroup collaboration?Overall, we expect that the findings would reflect sociocognitiveaspects of interactive behavior, that would be beneficial for rais-ing the awareness of instructors and learners on individual andgroup differences in CMC. We hope the findings would serve usefulinformation for facilitating better discussions, designing adaptivelearning strategies that improve learning outcomes, and promotinggender equity in online collaborative learning.4 METHODS4.1 ParticipantsThe sample used in this paper comes from a previous study [6].The participants were enrolled in an introductory-level psychologycourse taught at a university in Southwest U.S.. The sample in theoriginal study consisted of 840 students. We excluded responsessuch as prefer not to answer and N/A on gender. This yields toa total of 824 responses in our final sample. Overall, 510 femalestudents made up 62% of the sample.4.2 Course details and procedureStudents were told that their assignment was to log onto an onlineeducational platform specific to the University at a specified time.Students were also instructed that, prior to logging onto the edu-cational platform, they should read certain assigned material onpersonality disorders. After logging onto the system, students tooka ten-item, multiple choice pretest. After completing the quiz, theywere randomly assigned to a chatroom of two to five people, alsochosen at random (average group size was 4.59), and instructed toengage in a discussion of the assigned material. The group chatlasted for exactly 20 minutes, when the chat window closed auto-matically. Then students took a second set of ten multiple-choicequestion post-test quiz.4.3 Data AnalysisThe discourse data used for analysis were linear messages loggedduring the non-directed chat room discussion. We calculated themeans for each GCA measures for male and female and performedANOVA test to examine whether there were significant differences.Subsequently, we conducted Tukey’s post hoc analysis to detectwhich variables contributed to significant differences. To answerour second research question, we generated descriptive data for thedistribution of six GCA profiles among female and male students(Figure 2). We recorded the entire sample based on separate GCAprofiles. For example, we assigned a new value to each participantas “chatterer” versus “non-chatterer”. Then we used chi-square testto determine whether there were differences in the proportion ofthis learner profile between two genders. The same procedure wasrepeated for “driver” and “non-driver”, and the rest of the learnerinteraction profiles.5 RESULT AND DISCUSSIONFigure 3 depicts the mean and standard error of male and femalestudents across the six dimension of GCA. No significant differencewas found in mean participation between male and female. UsingTukey’s post hoc test (q = .189), we found that themean female socialimpact (M = .31, SE = .04), responsivity (M = .27, SE = .04 ), and inter-nal cohesion were significantly higher than that for males, withqs =.349,qs = .244 and ,qs = .314, respectively. The significance level sug-gested robust differences between male and female students acrossthese three independent variables. Male and female students weresimilar in newness, and the result was not significant. Althoughmale students appeared to have slightly higher communicationdensity, the result was not significant. In the previous research, thesix GCAmeasures were used to identify distinct interaction profilesthrough a cluster analysis [6]. Figure 1 and 2 provide further infor-mation on how each profile was characterized across the six GCAproperties. Figure 4 shows the distribution of GCA profiles amongmale and female learners. We observed more drivers and influentialactors among all females. The two profiles are both characterizedby high social impact and responsivity, while differentiated by thedegree of participation. The results from chi-square test show thatcompared to male students, female students are more likely tobe drivers χ2(2,N = 824) = 6.01,p = .014 and influential actorsχ2(2,N = 824) = 14.42,p = .00. On the other hand, it appearedthat the proportion of chatterer and lurker was most evident amongmale students. Both profiles are similar in having a low average onall GCA dimensions, except that chatterer is characterized by high433LAK19, March 4–8, 2019, Tempe, AZ, USA Yiwen Lin, Nia Dowell, Andrew Godfrey, Heeryung Choi, and Christopher BrooksFigure 1: Learner profiles based on GCA measuresFigure 2: Learner profiles based on GCA measuresFigure 3: Gender differences across six GCA dimensionsparticipation. Compared to female students, male students are morelikely to be identified as chatterer χ2(2,N = 824) = 4.29,p = .38and lurker χ2(2,N = 824) = 13.80,p = .00. No significant dif-ferences were found between the percentage of male and femalestudents occupying the follower, or socially detached profile.Our results suggest, the differences between males and femalesis not in their degree of participation, but the extent to which theircontributions are semantically related to the subsequent contri-butions from other group members, their responsiveness to otherlearners, and consistency with their own contributions over thecourse of interaction. It is interesting to note, that the distinctionFigure 4: Distribution of GCA profiles among male and fe-male learnersbetween the more active female and male profiles, driver and chat-ter, resides in the intra and interpersonal features captured throughsocial impact, internal cohesion and responsivity. A similar patternemerges when we view the less active roles, that are frequentlyoccupied by males and females, namely lurker and influential actors.This comparison in the proportion of learner interaction profilesprovides stronger evidence for the significant differences betweengenders. We also observed that male students are relatively weakerin interpersonal communication, which is represented by socialimpact and responsivity. One possible explanation is that since mentend to use more assertive and authoritative communication, theircontributions may seem more distant to their peers. On the otherhand, one may argue that female’s high responsivity is due to theirtendency to explicitly agree and support others [8]. Although thismay make a reasonable claim, it is not sufficient to explain highsocial impact among females, which suggests their ability to evokemore corresponding responses from their peers. It was interestingto see that the two main GCA roles found among men, were chat-terer and lurker, which are distinctively different in the degree ofparticipation. Likewise, women were seen more likely to be identi-fied as drivers and influential actors, the former of which exhibitthe highest participation rate while the other the lowest. This mayexplain why we were not able to observe significant differences interms of the degree of participation between men and women. Thisevidence also shows that the dynamics of learners interaction aremore complex than mere participation, and examining the intricaterelationships in group collaboration requires more specific and elab-orated measurements. Our findings provide a contradicting viewto stereotypical gender expectations in group interactions. Eventhough male students traditionally were thought to be more activeparticipants in group interactions [14], our result suggests not onlydoes this not appear to be the case, but also that the contributionsfrom male partners did not exhibit effective interpersonal engage-ment, including a lack of monitoring and social regulatory skillsneeded for appropriate levels of responsivity and social impact. Onthe other hand, while women are often considered to more submis-sive and agreeing, we found that female students are more likelyto trigger relevant discourse that move the conversation forward.It is important to take into account these gender differences inlearners interaction patterns to reduce potential biases in course434Modeling Gender in Intra and Interpersonal Dynamics LAK19, March 4–8, 2019, Tempe, AZ, USAdesign to promote inclusive pedagogy and gender equity in onlinecollaboration.6 CONCLUSIONOverall, the results supported previous research showing that maleand female students interact differently in online collaboration.Specifically, our findings suggest that female students are signifi-cantly more likely to engage in productive discourse and a strongertendency to take on the roles of "driver" and "influential actor" incollaborative CMC.This paper has several limitations. First, our sample is based onintroductory psychology class at a U.S. university, whichmay not berepresentative of all online learners, thus our future research effortswill focus on the generalizability of these findings across other, andmore diverse populations of learners (e.g. MOOCs). Second, Savickiet al. [16] points out that gender composition in groups influencesgroup interaction dynamics. In this study, women comprised 62%of the population, and since participants were enrolled in the sameclass, it is likely that they may have unintentionally changed theway they interact with their peers based on their perceived gen-der distribution in class. Another confounding factor is that realanonymity is practically impossible to achieve since certain linguis-tic features imply gender information [11]. Although random IDswere assigned to participants, we can not confidently determine towhat extent students were interpreting gender inferences duringthese non-directed CMC, and whether they adjusted their commu-nication behavior according to that. Future studies may eliminatethis potential effect by imposing more rigorous control throughexperimental design.Despite the limitations, our research suggests that more atten-tion ought to be given to scaffolding effective communication incollaborative learning. This study brings up several possible chan-nels we may consider to improve learner interactions. For instance,information on learners’ interactive styles and profiles capturedby GCA dimensions may provide more fine-grained, constructivefeedback to instructors on student performance. In turn, this wouldallow instructors to adjust instructional design and incorporatestrategies to facilitate cohesive dialogue and engage students inmeaningful discourse. Evidence has shown that effective communi-cation is crucial to successful collaborations [20]. Therefore, tailoredadvice on learners’ communication skills could not only benefitindividuals’ learning experience but ultimately enhance overallgroup collaborative outcomes. Monitoring students’ GCA perfor-mance would provide instructors with have specific informationon intra and interpersonal dynamics at individual and group levelto achieve that goal. On the other hand, according to social com-parison theories [7], giving learners access to data may incentivizetheir motivation to learn. For instance, we may consider building areal-time feedback tool that allow students to see their GCA per-formance compared to the average of others in the group. Or givestudents choices to receive a report on their own communicationpatterns. Another potential intervention could be using prompts toencourage less engaged students, who are identified based on thetracked history of their interaction behavior. GCA opens up manypossibilities in future study designs. Nevertheless, it would requirefurther investigations to identify the impact of these applicationsin the context of collaborative learning.We intend to conduct future research to examine whether ethnic-ity affects intra- and interpersonal dynamics in learner interactions.Additionally, how interaction dynamics is related to overall learningoutcomes should also be examined. With a broadened scope of re-search focusing on gender and ethnic differences in communication,we believe that it will serve as a meaningful channel for learninganalytics to promote inclusion and success in online learning.7 ACKNOWLEDGMENTSThis work was supported in part by funding from the Office ofAcademic Innovation, University of Michigan.REFERENCES[1] Elizabeth Aries. 1987. Gender and communication. Sage Publications, Inc.[2] Rachel Baker, Thomas Dee, Brent Evans, and June John. 2015. Bias in onlineclasses: Evidence from a field experiment. In SREE Spring 2015 Conference, Learn-ing Curves: Creating and Sustaining Gains from Early Childhood through Adulthood.5–7.[3] Yoany Beldarrain. 2006. Distance Education Trends: Integrating new technologiesto foster student interaction and collaboration. Distance Education 27, 2 (Aug.2006), 139–153.[4] Lee J Cronbach and Richard E Snow. 1977. Aptitudes and instructional methods:A handbook for research on interactions. Irvington.[5] Nia M Dowell, O Poquet, and Christopher Brooks. 2018. Applying group commu-nication analysis to educational discourse interactions at scale. Proceedings of theof the 13th International Conference on the Learning Sciences (2018), 1815âĂŞ1822.[6] Nia MMDowell, Tristan M Nixon, and Arthur C Graesser. 2018. Group communi-cation analysis: A computational linguistics approach for detecting sociocognitiveroles in multiparty interactions. Behav. Res. Methods (Aug. 2018).[7] Leon Festinger. 1954. A theory of social comparison processes. Human relations7, 2 (1954), 117–140.[8] Jane Guiller and Alan Durndell. 2007. Students’ linguistic behaviour in onlinediscussion groups: Does gender matter? Comput. Human Behav. 23, 5 (2007),2240–2255.[9] Susan C Herring. 2000. Gender differences in CMC: Findings and implications.Computer Professionals for Social Responsibility Journal 18, 1 (2000), 0.[10] Christopher Hoadley. 2010. Roles, design, and the nature of CSCL. Comput.Human Behav. 26, 4 (July 2010), 551–555.[11] Eun-Ju Lee. 2007. Wired for Gender: Experientiality and Gender-Stereotyping inComputer-Mediated Communication. Media Psychol. 10, 2 (June 2007), 182–210.[12] Ronald B Marks, Stanley D Sibley, and J B Arbaugh. 2005. A Structural Equa-tion Model of Predictors for Effective Online Learning. Journal of ManagementEducation 29, 4 (Aug. 2005), 531–563.[13] James W Pennebaker, Matthias R Mehl, and Kate G Niederhoffer. 2003. Psycho-logical aspects of natural language use: Our words, our selves. Annual review ofpsychology 54, 1 (2003), 547–577.[14] Fleur Ruth Prinsen, Monique L L Volman, and Jan Terwel. 2007. Gender-relateddifferences in computer-mediated communication and computer-supported col-laborative learning. Journal of Computer Assisted Learning 23, 5 (2007), 393–409.[15] Jennifer Richardson and Karen Swan. 2003. Examing social presence in onlinecourses in relation to students’ perceived learning and satisfaction. (2003).[16] Victor Savicki, Merle Kelley, and Dawn Lingenfelter. 1996. Gender and groupcomposition in small task groups using computer-mediated communication.Comput. Human Behav. 12, 2 (June 1996), 209–224.[17] Hassan M Selim. 2007. Critical success factors for e-learning acceptance: Confir-matory factor models. Comput. Educ. 49, 2 (Sept. 2007), 396–413.[18] Elizabeth Stacey. 2007. Collaborative Learning in an Online Environment. In-ternational Journal of E-Learning & Distance Education / Revue internationale due-learning et la formation à distance 14, 2 (Aug. 2007), 14–33.[19] Florence R Sullivan, Manu Kapur, Sandra Madden, and Stefanie Shipe. 2015.Exploring the Role of ‘Gendered’ Discourse Styles in Online Science Discussions.Int. J. Sci. Educ. 37, 3 (Feb. 2015), 484–504.[20] Esther Suter, Julia Arndt, Nancy Arthur, John Parboosingh, Elizabeth Taylor, andSiegrid Deutschlander. 2009. Role understanding and effective communicationas core competencies for collaborative practice. Journal of interprofessional care23, 1 (2009), 41–5

Mots-clés

Embedded system,Agent-based model,Mathematical model,