Factors influencing peer referencing behavior during an online learning activity

Significance of this study

Answers to the two targeted research questions have importance in several regards. While substantial empirical evidence indicates that people are influenced by peers in various ways (Brechwald & Prinstein, 2011; Hogg & Tindale, 2008; Rubin et al., 2006), they primarily involve un-intervened and regular daily routines. For instance, Rubin and colleagues (2006) studied how children interact socially and develop dyadic relationships and attitudes that impact their behavioral tendencies and peer experiences in informal settings. Very few researchers (e.g., Palonen & Hakkarainen, 2000) dealt with the social attributes notably exerted on learning-related tasks.

Furthermore, current designs for expressing and displaying personal endorsements, referrals, acknowledgments, or other similar intent in social media and online spaces abound (e.g., the ‘like’ function on Facebook and Instagram and recommendations on Amazon). Studies have confirmed that such social endorsements act as a source of peer influence that affects behavioral and neural responses (Sherman et al., 2016, 2018). Nonetheless, thus far, existing studies examining this potential bias have mainly involved simulated situations in lab settings. Investigations of how referencing behavior, which connotes affirmation toward interacting parties, are needed to enrich and extend our understanding of this emerging area in authentic educational activities.

Also, computer-mediated communication in various forms may affect impressions, interactions (Walther, 2011), and even self-perception (Walther et al., 2011). Thus far, existing studies have typically been examined in media communication or interpersonal relationship fields. Studies investigating how one’s affiliations, attributes, or related characteristics may affect interactions in a technology-supported learning environment are few and far between. With networked computers equipped with easy real-identity concealment capabilities, different degrees of social context cues and information can be blocked during an interaction. The findings gleaned from this investigation should be germane in terms of online system design and instructional implementation strategies. Suppose changes in participants’ responses to different identity revelation modes (i.e., real-name mode to anonymity mode) are significant. In that case, it is indicative that peers pay more attention to their interacting parties’ characteristics (e.g., social or academic status in the group) during the learning process. This bias might inadvertently negate the original intended intent or goal — cognitive orientation on the focal task, which includes social learning through inspecting and referencing peer-generated questions in this study, objective evaluation of work, or performance for recommendations in peer-assessment tasks, and so on. If such effects are detected, instructors would be better informed of and alerted to this phenomenon. Preventive measures can, thus, be contemplated and implemented in advance to mitigate the associated effects or obscure contextual cues in online peer-assisted learning spaces. With this investigation, any presence of unintended biased online interaction has a possibility of being revealed, and an egalitarian education ecosystem based mainly on merits could be protected.

Social impact theory and its implications for this study

The social impact theory proposed by Latané and his colleagues identified three variables as the primary determinants of interpersonal power influence that affect personal feelings, attitudes, thoughts, behavior, or decision-making processes: strength, immediacy, and number. Simply put, strength denotes personal attributes that are frequently associated with intellect, wealth, support, perceptions of belongingness, sense of closeness, credibility, social structure, physiological variable, and the status and power of the source of influence (Latané, 1981, 1996). People who are presumably intellectually superior, wealthier, receive more support, have better-established relationships (e.g., close or best friends), are considered trustworthy, affiliate with a particular social group or sex category, and are of higher status or with greater power within the network are believed to exercise more influence over others (Latané, 1996). Immediacy, on the other hand, connotes the physical and temporal distance between the source and target of influence (Latané, 1981). People within close spatial proximity and/or dealing with recently occurring events are considered to have better immediacy and exert more social impact on individuals in the network (Latané, 1996; Oc & Bashshur, 2013). Lastly, the number relates to how many sources there are. When the number increases, the impact on the target also increases (Latané, 1981).

Because online learning activities in class settings mostly involve students simultaneously interacting with one another in the same class, ‘number’ and ‘immediacy’ variables are less of a concern (due to few, if any, variances among the constituents). Conversely, relevant personal factors that make an individual influential and cause that individual to gain more ‘strength’ and high social impact during activity may exert effects; hence, factors related to the ‘strength’ variable are targeted in this work.

The influence of academic prestige on behavior under a social interaction context

Given the social impact theory and the learning context under investigation, firstly and most importantly, ‘academic prestige,’ which has been frequently found to exert social influence over peers in a network (e.g., Adler & Adler, 2003; Gifford-Smith & Brownell, 2003; Kiefer & Ryan, 2008; Sandstrom, 2011) is targeted and measured. For this, in addition to the objective, actual academic achievement, student subjective perceptions of peer academic prestige (i.e., perceived academic status) were collected. In this study, we stress students’ perceptions partly because studies have confirmed a stronger correlation between adolescents’ behavior with their subjective perceptions of peers when compared with the actual conduct of their peers (Prentice & Miller, 1996). Besides, the Ministry of Education, Taiwan (2012) has banned public display and announcement of student academic performance in respective classes and schools for years, so the actual academic achievement status of students may be unknown for one other to exert effects on behavior. Also, in light of the vital role of task knowledge and ability in learning and cognitive development (Flavell, 1979), students’ academic prestige associated with task-specific skills, including both actual performance on the task and inferred ability as endorsed by the teacher, was adopted and measured as well. Questions were examined regarding whether these four indexes of academic prestige influence peers to the extent of affecting their referencing behavior (i.e., student-generated questions being cited more during online learning activities).

The influence of friendship on behavior under a social interaction context

Friendship serves as an essential basis of influence on academic-related performance, attitude, emotion, behavioral conduct, choice behavior related to group affiliation, a sense of belongingness, and identity foundation (Adler & Adler, 2003; Berndt & Keefe, 1995; Brechwald & Prinstein, 2011; Dokuka et al., 2020; Ng-Knight et al., 2019). For instance, Ng-Knight and colleagues (2019) conducted a longitudinal study on pre-adolescent friendship during school transitions. They confirmed the findings of Berndt and Keefe (1995), indicating that high-quality ties of friendship are positively correlated with better grades in school and negatively correlated with emotional or conduct problems (Ng-Knight et al., 2019). Altermatt and Pomerantz’s (2003) empirical and review study on friendship and influence verified a consistent relationship between close friendship ties and students’ self-perceived competence, ability attribution for success, and attainment of academic standards. The study by Takeuchi (2016) found that friendship-oriented groups were more likely to initiate higher levels of assistive learning than individualistic learning behavior.

While existing results confirmed the influence of friendship, most studies used self-reported data to assess such effects. This measure is subject to research validity threats and limitations caused by social desirability effects, respondent introspective ability, response bias, etc. (Brewer, 2000). Further, few studies have dealt with structured learning tasks (e.g., activities designed by the instructor), which are prevalent in educational contexts. Considering all the issues, the question as to whether academic work or artifacts produced by one’s close friends or by those in the same social group (i.e., circle-of-friends) will attract more attention (i.e., be referenced substantially more) is investigated under an educational activity (e.g., student-generated questions).

The influence of gender similarity on behavior under a social interaction context

Gender is a major influence that affects social interactions and behavior (Busching & Krahé, 2020; Gifford-Smith & Brownell, 2003). Specifically, gender similarity (i.e., persons of the same gender) is found to be an influential moderating factor in affecting gender-oriented social interactions and behavior in a variety of social interaction contexts (e.g., social networking services, interactions in the family or school settings, alcohol and substance use among peers, use of physical and relational aggression among friends, etc.) (Andrews et al., 2016; Borsari & Carey, 2001; Brechwald & Prinstein, 2011; Laniado et al., 2016; Shin, 2017; Shrum et al., 1988). For instance, Andrews and her colleagues (2016) found in their empirical research on gender similarity with sixth-grade students that girls who felt similar to their own-gender group engaged in significantly higher relational aggression than boys. In contrast, boys who felt similar to their own-gender group were more elevated in physical aggression than girls. Borsari and Carey’s review study (2001) also found gender-oriented behavior where the drinking on all-male occasions was most extreme, and males were more susceptible to competitive drinking. In a study investigating online relationships of approximately 10 million users, primarily teenagers using a Spanish social networking service, Laniado and his colleagues (2016) observed a remarkable tendency in gender-homogeneous interactions and groupings for female users, especially in younger teenagers. Notably, most existing literature investigating the factor and issue of gender similarity dealt with non-educational contexts. Comparatively, fewer studies were conducted in educational settings. Among these, a great deal focused on gender-segregated play, as being observed more frequently during the elementary-school phase, while cross-gender relationships were typically observed among preschoolers rather than pre-adolescents (Adler & Adler, 2003).

As for learning behavior per se, Palonen and Hakkarainen’s (2000) work provided a vivid picture of gender-biased interaction among fifth and sixth graders engaging in a computer-supported collaborative learning environment. In their study, high-achieving female students dominated the conversation and interacted more intensely with average and above-average females, revealing an apparent tendency toward gender similarity interaction during collaborative learning. In light of this, it is speculated that questions generated by elementary school students of the same gender (i.e., the operational definition of gender similarity) may be referenced more due to their sense of belonging to the same categorical group. In other words, gender-related bias behavior that has been frequently detected in other studies (e.g., Ryan, 2001) would be observed in students at the pre-adolescent stage.

Research method and hypotheses

A mixed methods research (specifically, a correlational embedded design model) was adopted in this study. RQ#1 was examined using a correlational study, and a survey questionnaire was developed based on insights gained from non-participant observation and informal interviews to collect data for RQ#2 (Figure 1).OPEN IN VIEWER

Because the variables examined in RQ#1 (i.e., academic prestige, friendship ties, and gender similarity) involve matrix and continuous data types, which demand differential data analysis procedures, three hypotheses (H1-H3) were formulated:

(1) Perceived academic standing, friendship, and gender similarity can positively predict the number of student-generated questions being referenced by peers.

Participants, the study context, and experimental procedures

Seventy-five fifth-graders (n_males = 37, M_age = 11.08, SD _age = 0.28) from three intact classes at a primary school in Tainan City, Taiwan, participated in this study for 12 weeks after securing proper parental consent forms. The three fifth-grade classes taught by the same computer literacy and science teachers were purposively selected. Fifth-graders were targeted in this study mainly because they have moved beyond the concrete operational stage and entered the formal operational stage at this age, according to Piaget’s theory of cognitive development (Ginsburg & Opper, 2016). Thus, they can benefit more from a structured learning activity (i.e., online SQG and STC in this study). Also, with a computer course being part of the required courses since the third grade (as regulated by the Ministry of Education in Taiwan), the participants had the fundamental computer competencies and operational skills necessary to engage meaningfully in the focal activity. Further, the integrated online student-generated questions activity was designed to help realize the objectives envisioned in the white paper on information technology education in primary and secondary schools (Ministry of Education, 2008) and the digital learning promotion plan, which values the integrated approach to using computer technology to support teaching and learning, cultivate key core competencies, and actualize learner-centered education (Ministry of Education, 2016).

This study was conducted in the regular weekly 40-minute computer classes by a female, certified elementary school teacher with more than 3 years of teaching experience. The integrated online learning activities were introduced to support the teaching and learning of science, which were allocated three 40-minute instructional sessions weekly. True/false and multiple-choice questions were chosen as the question types for the online activities considering their prevalence at the primary school level.

Mainly, the study consisted of two stages: induction (6 weeks) and intervention (6 weeks) (Figure 2). During the induction stage, the focus was on familiarizing and equipping the participants with essential knowledge and skills in question generation, test construction, and the operating procedures of the adopted system. On a weekly basis, the participants individually generated three true/false and multiple-choice question items on the system (Figure 3) on the science content covered in the current week after a brief whole-class feedback session on their performance on the previous task. Noting the pragmatic value of whole-class feedback (Hattie & Clarke, 2018), the implementing instructor selected three to five student-generated questions from the respective classes so that an adequate diversity of models could be exposed and used as exemplars for participants. Such practice could help students understand and appreciate assessment criteria and academic standards (Hendry et al., 2011; To et al., 2021). Meanwhile, their knowledge, skills, and capabilities at the task can be expanded (Hattie & Clarke, 2018). Subsequently, after the last instructional session on the current science unit was concluded, the participants were directed to construct a 10-item test on the entire unit, with reference to both self- and peer-generated questions. Essentially, the students selected any question(s) to be included in the self-constructed tests by dragging them from the questions window to the test window (top of Figure 4) after reviewing and editing (if deemed necessary) any question item shown at the bottom. Clicking on the icon at the bottom-left corner of the questions window switched between the self and peer-generated questions spaces. Each self-generated question was represented as a blue box, and each peer-generated question was shown in yellow in the questions window.OPEN IN VIEWER

OPEN IN VIEWER

Figure 3.

True/false (left) and multiple-choice question-generation in the system (right).

OPEN IN VIEWER

Figure 4.

Viewing and including self-generated questions (left) and referencing of peer-generated questions in the system (right).

The intervention stage basically followed the same implementation procedures as those used in the induction stage. Nevertheless, for the purposes of this study, the name of the question-author was shown at the intervention stage when the participants examined and decided which peer-generated question(s) would be included as part of their self-constructed tests. Three online test-construction activities were undertaken at this stage and, thus, were used in the data analyses.

Instruments

Several instruments and data sources were used in this study. First, prior to the study, two questions were disseminated to each of the participants to collect data on perceived academic prestige and friendship states — Please name the five students you think demonstrate the best academic performance in your class; Please name the top five students in your class with whom you befriend.

Second, all questions included in the participants’ final submitted tests during the intervention stage (i.e., three in total) were analyzed for the number of references received from their peers. Third, the participants’ science performance on the post-test was collected to measure actual academic achievement. Fourth, to assess the students’ actual task-specific abilities, in reference to the literature on question-generation and test-construction (e.g., Gronlund, 1993), six indexes were devised for the question-generation performance assessment (i.e., fluency, flexibility, elaboration, originality, cognitive level, and importance), and three were designed for test-construction (i.e., arrangement, comprehensiveness, and distribution of low and high cognitive levels). Each of the indexes was operationally defined to ensure an objective assessment. Furthermore, 30% of the last student-constructed tests were randomly selected from each of the three participating classes and evaluated by another independent rater, yielding satisfactory inter-rater reliability of the SQG score (r = 0.851, p < .01). Fifth, model questions selected by the teacher and shown during the weekly whole-class feedback sessions were counted and used as the students’ inferred task-specific abilities assessment.

Finally, with insights gained from non-participant observation and informal interviews during the study, one question was administered to tap into students’ perceptions regarding important factors that affect their online referencing behavior—What factor(s) affect your decision as to which peer-generated question(s) to refer in your constructed tests? (Choose all that apply.) Questions that are: (a) interesting to me, (b) of high quality, (c) from good friends of yours, (d) of high difficulty level, (e) selected by the instructor as models during the whole-class feedback session, (f) original and unique, (g) authored by the academically excelled, (h) not covered by you, yet targeting important concepts of the learning material, (i) easy, and (j) referenced by many students.

Data analysis

To test H1, a multiple regression quadratic assignment procedure (MR-QAP) was adopted. The dependent variable was the referencing matrix, and the independent variables were the peer influence matrices, including perceived academic prestige, friendship ties, and gender similarity. As the data on perceived academic prestige and friendship are interdependent, QAP was adopted to analyze the connections between dyadic data sets (Kilduff & Krackhardt, 1994). The data was transformed by collapsing the perceived academic prestige and friendship matrices into an adjacency matrix. First, the ties in the perceived academic prestige matrix were collapsed to a binary variable: coded 1, if listed among the top five in the prior-to-study survey by the participants; coded 0, if otherwise. Similarly, the friendship matrix entailing the top five selections from each participant was transformed into a binary variable. Based on these criteria, the perceived academic prestige and friendship network information were then translated into an adjacency matrix of links for each participating student, where 1 represents existing endorsement and 0 denotes the absence of endorsement. Gender similarity attribute was also converted into a matrix format, whereas the same-gender ties are coded 1, and 0, if otherwise. The dependent variable derived from the log files of the referencers (i.e., the participants who adopted other pupils’ generated questions in their final submitted STC) and referenced (i.e., the participants whose work was used by others) were converted into a one-mode matrix using RStudio 1.0.143.

For statistical testing of H2, a simple regression was adopted with actual academic prestige as the independent variable and the participant’s item referencing centrality as the dependent variable. Eigenvector centrality was embraced to better capture the effect of a participant’s influence in the referencing network by considering the centrality of those with whom one is tied (McCulloh et al., 2013).

H3 was tested using a two-step hierarchical linear regression (HLR) after variance inflation factor tests were satisfied. The number of student-generated questions selected by the teacher as models was entered in step 1, and the SQG score was entered in step 2.

All statistical testing described above was conducted for each of the three participating classes, respectively, since referencing from other classes was not supported in this study due to the focus of this investigation. Finally, descriptive statistics were used to calculate the students’ selections of factors affecting their online referencing decisions for the survey.

Discussion for RQ#1 — if and how academic prestige, friendship ties, and gender similarity impact students’ referencing of peer-generated questions during an online STC activity

In consideration of the focal context (i.e., online structured learning tasks in classroom settings) and in light of the ‘strength’ influence of the social impact theory (Latané, 1981, 1996), three factors were identified and expected to exert social power over peers in terms of impacting peers in referencing others’ work: academic prestige, friendship, and gender similarity. For a fuller understanding of the concept of academic prestige, four measurement indexes were included to address academic achievement in the applied content and the level of task-specific ability in both actual and perceived/inferred dimensions. The major findings are summarized and discussed below.

First and foremost, as reflected by the results of the MR-QAP, when there were friendship ties between the participants, there was significantly more referencing in self-constructed tests. This finding confirmed what would be expected from the social impact theory. Essentially, having better established personal relationships is believed to exercise more impact over others within a network (Latané, 1996). Conceptually, our finding on the relationship between friends and online referencing behavior resonates with the results of studies concerning the influences of friendship on learners’ academic-related attitudes (e.g., beliefs in ability attribution for success) (Altermatt & Pomerantz, 2003; Chen et al., 2009) and social interactions (e.g., assistive learning) (Takeuchi, 2016). The fact that participants’ referencing behavior was more pronounced in their friends also corroborates with previous studies (e.g., Linden-Andersen et al., 2009; Shin, 2017). For instance, in Linden-Andersen and colleagues’ study (2009), adolescents in reciprocated friendship were found to value friendship and strive to resemble their friends compared to those in non-reciprocated friendship. This part may help shed some light on why the participants in our study exhibited significantly more referencing behavior for questions generated by their friends. We theorize that participants might have a better chance of reaching a state of resemblance in terms of the quality of produced work or style of presentation to those of their friends.

Second, the significant finding regarding the influence of model questions highlighted by the teacher on peers’ online referencing behavior resounded with that of the post-session survey. More than two-thirds of the participants (68.92%) regarded this as an important factor to consider during referencing. It also reflected what French and Raven (1959) identified as ‘legitimate power’ — having the position of authority to make demands for others to comply in a formal or informal social structure. Because teachers are commonly regarded as authority figures who may significantly impact students, they act as a powerful social force in directing students’ actions and decisions in educational settings (Hendrickx et al., 2017; Stefanou et al., 2004). Therefore, the fact that questions hand-picked by teachers as models affected the participants’ online referencing behavior significantly can be explained and understood.

Besides the two significant findings, the other non-significant ones also warrant noting. First, of the four academic prestige indexes, actual academic achievement, perceived academic status, and actual performance in the online task were not significant predictors of the number of times student-generated questions were referenced by their peers. In other words, in this study, students with better academic achievement, higher perceived academic status, and superior task-specific performance did not impart a significant social influence over their peers. These findings, nevertheless, highlighted that ‘intellectual superiority,’ a term coined by French and Raven (1959), alone did not hold enough power to affect students’ referencing of peer-generated questions in online spaces. Additionally, the fact that online referencing behavior was found to be impacted by friendship but not by academic prestige, in essence, corroborated with Kwon and Lease’s (2014) study. They found that closer friends influenced students’ behavior (in terms of conforming to their close friends’ level of academic effort) more than ‘reputational influence’ (i.e., high-status peers).

Second, gender similarity was not found to affect students’ online referencing behavior in this study. The finding indicates that affiliations based on gender similarity did not have enough social power to affect peers’ referencing behavior in the network. The currently non-significant finding, on the one hand, reflects the point accentuated by Brechwald and Prinstein (2011) that gender may interact with other variables, thus rendering mixed results in gender-related studies. On the other hand, while studies have suggested that gender similarity is a factor influencing peer behavior (e.g., Palonen & Hakkarainen, 2000), further analysis of the literature revealed that of those studies yielding supportive results, many focused on non-academic behavior (e.g., alcohol consumption, eating problems, body image concerns, aggression, health-risks, and illegal behavior) (Borsari & Carey, 2001; Brechwald & Prinstein, 2011; Jones, 2001; Shin, 2017), or dealt with different age groups. For example, Jones (2001) found clear evidence of gender similarities in adolescents’ social comparison and its relationship with boys’ and girls’ body image concerns. Specifically, girls reported elevated levels of social comparison with their same-gender peers on weight; for boys, similar positive body image evaluation ratings were found in muscular build with other boys. The participants also endorsed different attributes for body dissatisfaction with their same-gender peers. The findings, taken as a whole, implied that even though gender similarities may play a salient role in forming gender-oriented norms and perceptions, they might not exert the same effects on the targeted behavior due to the different cognitive, emotional, and social development of the participants (Rubin et al., 2006; Sterry et al., 2010). Our result may also reflect what has been suggested by Adler and Adler (2003) and Shrum and colleagues (1988) that different behavioral tendencies in terms of gender-crossing and segregation would be exhibited under different contexts.

Discussion for RQ#2 — what factors students subjectively perceive as dominating their online referencing of peer-generated questions

As reflected by the post-study survey, most participants weighed their decisions on referencing peer-generated questions on a mixture of factors while engaged in online learning tasks. In particular, factors related to the content of questions as recognized by most of the participants (more than 50%) include questions that are ‘original and unique’ (86.49%), ‘of high quality’ (71.62%), ‘interesting to me’ (70.27%), ‘targeting important concepts of the learning material but yet missed by the student’ (58.11%), and ‘of high difficulty level’ (56.76%). In contrast, factors not directly related to the content of the questions, including questions ‘authored by the academically excelled’ and ‘referenced by many students,’ were pinpointed by comparatively fewer participants (less than 50%) as influential in dictating their decisions when it came to referencing peer-generated questions for self-constructed tests in the online learning space.

Many of the factors identified and obtained via the post-study survey echoed those found in one study examining assessors’ targeting behavior during peer-assessment learning activities (i.e., deciding which item to assess). Explicitly, ‘interestingness,’ ‘difficulty level,’ and ‘relating to the key ideas of the study material’ emerged as salient features when students scanned for target items to assess (Yu & Sung, 2016). Even though the focal learning task in this study — referencing peer-generated questions, which deal with selecting which item to be included in one’s work, differed from other research, it essentially entails both the act and intent of one’s targeting behavior. Thus, the results in this study corroborate and substantiate the previous work.

In addition, the finding that questions that are ‘original and unique’ were chosen by most of the students (ranked #1) as influencing their referencing decisions reflects the idiosyncratic feature of student-generated questions as accentuated by Leung (1997). As specified in Keller’s motivational theory (1987), uniqueness is one method that can be used to attract and sustain learner attention. It is thus not surprising that most potential users consider questions regarded as original and unique during referencing.

Finally, the finding that students ‘attend to peer-generated questions targeting important concepts that are not yet covered by the questions generated by themselves’ accentuates a potential affordance of online spaces for learning. By enabling an observational learning space, referencing others’ work provides learners with additional learning opportunities (e.g., alerting students to the major points they missed), through which existing skills and knowledge can be enhanced, as suggested by Bandura’s observational learning theory (1986).

The contributions of this study and suggestions for instruction and system designs

The social impact theory quantifies social influence by highlighting strength, immediacy, and number variables and translating them to denote one’s impact on the interacting group (Latané, 1981, 1996). The theory has been applied to various disciplines and contexts, including political science (e.g., Seltzer et al., 2013), communication (e.g., Latané, 1996), marketing and organizational behavior (e.g., Perez-Vega et al., 2016), social media (e.g., Chang et al., 2018), and educational settings (e.g., Smirnov & Thurner, 2017), among others. Generally speaking, there has been a consensus that interpersonal power structures and social forces impact one’s attitudes, behavior (Sandstrom, 2011; Seltzer et al., 2013), and decision-making processes during either virtual or physical interactions (Kiefer & Ryan, 2008; Smirnov & Thurner, 2017). Nevertheless, rarely has the theory been applied to formal, structured learning activities. Given that existing studies on peer influence mostly deal with non-educational matters (e.g., un-intervened daily routines or concerns), studies exploring if and how group affiliations may significantly affect learning-related interactions to the extent of inadvertently biasing behavior are warranted. More examinations and discussions of the applications of social impact theory in teaching and learning could contribute to our understanding of if and how different individual and group characteristics affect thoughts, feelings, behavior, and interactions. More importantly, such a gained understanding can help create an equitable online interaction space that is suitable and optimal for learning. This topic under examination appears pressing, especially in the face of the growing demand for online learning for the next generation comprising the current student population.

The obtained data help make it possible to provide explicit suggestions for instructional implementation involving structured online learning tasks and online system designs involving referral, acknowledgment, and endorsement behavior. For instructional implementation suggestions, while the majority of the participants appeared to weigh their online referencing decisions based on a mixture of factors, with the content of questions considered more extensively, still, more than one-third of the participants considered questions ‘referenced by many students’ and ‘authored by the academically excelled’ during their referencing decisions. In addition, friendship ties appear to be a significant predictor of referencing patterns. Also, prior studies found that adolescents in reciprocated ties of friendship tend to conform to or resemble their peers’ behavior for the sake of social acceptance (Crockett et al., 1984; Erikson, 1968; Gifford-Smith & Brownell, 2003; Linden-Andersen et al., 2009; Slavin, 2015). In light of all these, it is suggested that instructors demand that their students provide explicit reasons for any online referrals, acknowledgments, or endorsements (e.g., referencing in the case of the current study). This piece of suggestion is particularly relevant, noting the cultural values, beliefs, and social norms of non-Western contexts, such as Chinese societies (e.g., emphasis on sympathy, harmony, high academic performance, academic success, etc.) (Hau & Ho, 2010; Yang et al., 2006) and South Korea (e.g., more conservative friendship dynamics) (Shin, 2017), which may lend participants to indistinguishably concur with those excelling academically or with whom they choose to befriend. This measure, to say the least, would help avoid the bandwagon effect, which involves conformity or responses to peer pressure based on social categorization or socialization, which is prevalent in group situations (Sandstrom, 2011). For a greater effect, it would not only help promote meaningful and reflective learning on the part of the learners but also help address the ethical requirements of avoiding plagiarism in early schooling for all elements of submissions and the factor of endorsement within the microcosm of a school.

Similarly, with ‘questions selected by the teacher as models’ found to dictate students’ online referencing behavior (as confirmed in the HLR and reflected in the post-study survey), requesting a rationale for endorsing any piece of information on the part of a student should help mitigate the effects of the legitimate power of those in a position of authority within a social structure (French & Raven, 1959). This online measure is crucial because whole-class feedback highlighting the model work of students has been suggested to be an effective and pragmatic instructional practice (Hattie & Clarke, 2018; Yelon, 1996).

Some designs are suggested to be embedded in related systems to support the aforementioned instructional measures and unbiased online interaction. Explicitly, a designated field in the system that the referencers can use to legitimize the referencing of any peer-generated question (or the approval of any peer’s work for recommendation or peer-assessment) is demanded. In addition, rather than adhering to a fixed item sequencing method, versatile multiple sequencing methods are suggested to be built in systems entailing endorsement or referencing features so that student-created work can be accessed. Some common sequencing approaches, including submission time, number of times the posting has been referenced, shared, or liked, the author names, the content or keywords of the question, and so forth, can be integrated into the system to ensure that the work can be examined effectively, efficiently, and flexibly.

Before proceeding, the authors would like to point to the implication of the currently obtained finding on friendship as a significant predictor of referencing patterns for cooperative or collaborative learning (CL) practice. Explicitly, while studies have found that friendship grouping was preferred by Asia students (Pham & Gillies, 2010), and CL grouping based on affection and personal relationships was practiced (Zajac & Hartup, 1997), allowing students to choose their own group members may tend toward homogeneity, which is against the heterogeneous group composition guideline as commonly recommended by CL experts (Jacobs & Seow, 2015). Hence, instructors are advised to carefully guard against any undesirable signs of conformity within CL groups by requesting elaborated explanations for consensus or group decisions reached, especially when forming CL groups initiated by students is practiced. Through such a designed mechanism, the essential elements of CL (e.g., equal opportunity to participate, maximum peer interactions, promotive interaction, etc.) (Jacobs & Seow, 2015; Johnson & Johnson, 2009) and its beneficial effects on enhanced cognitive and affective outcomes would be more likely to be realized.