Teamwork in context of diversity

KEER2022, BARCELONA, SPAIN | SEPTEMBER 6-9 2022 9TH INTERNATIONAL CONFERENCE ON KANSEI ENGINEERING AND EMOTION RESEARCH 2022 TEAMWORK IN CONTEXT OF DIVERSITY Awoniyi STEPHENa a Texas State University, U.S.A., [email protected] ABSTRACT Utilization of teamwork for problem solving is pervasive. Teamwork is employed in classrooms to facilitate learning, but also as preparation for future vocational practice. It is used in research as collaborative ethos. Teamwork features in workplace tasks, leisure time projects, pursuing solutions to intractable trans-situated problems, etc. The space of teamwork is a public space of multidimensional assets. Inherent in it are benefits of collaboration, but it also brings with it challenges that have to be resolved in order for it to work effectively. We pursue the question of effectiveness of team function through team size, given certain identified benefits and disbenefits. We attempt to find out if there is an optimal small team size based on a few characteristics such as burden sharing, transactive knowledge and conflict (e.g. schedule conflict). We employ an agent-based model. For parameterization of variables, we take data from a short questionnaire completed by students and use its results to set values. We create small teams of different sizes and allow the dynamic model to aggregate those values as adopted by agents. We also attempt to see which one of four benefit/disbenefit valuation model specifications might work best. Keywords: knowledge management, perception of team benefit/disbenefit value, social/public space, teamwork 1 INTRODUCTION In day-to-day life, it might not be completely avoidable to have to collaborate with other people in order to complete certain tasks. Collaborating could be motivated by dependency, choice, efficaciousness, protocol, attachment, companionship, or a combination of those or other reasons. Certainly, it takes place in different domains of life: work, leisure, transcendent spaces. Spaces of collaboration or team spaces are public (or social) spaces. One thing that can be expected in them is diversity--of goals, tasks, structure, selves, and so on. Selves are central to the idea of teamwork. They constitute the team. They bring diversity of knowledge, diversity of doi: 10.5821/conference-9788419184849.65 633 KEER 2022 | 9TH INTERNATIONAL CONFERENCE ON KANSEI ENGINEERING AND EMOTION RESEARCH 2022 schedule, diversity of attitude, diversity of personality--including preferences, emotions, temperaments, dispositions, etc. Diversity sketches one profile of the operational space of teamwork. For a team to work well, various forms of difference have to be managed. When teams exist in the context of problem solving, one of the questions that arise is about an ideal size of the team. That is one question to which we turn in the current exploration. Can one identify an optimal size of a small team? Our exploration invariably took us to another question: If goal of small team work is successful completion of a task, is there a way of recognising or earmarking both generative role of benefits and debilitating value of challenges for a theorized optimality? In this exploratory paper, we inquire after those two questions. The former question is examined through creating different team sizes. The latter question is re-constructed through different exploration model designs--in order to see which one might capture benefits better. 2 ASPECTS OF TEAMWORK 2.1 Knowledge generation and management According to Gadgil and Nokes-Malach (2012), beneficial features of constructive collaboration include such things as sharing prior knowledge, collaborative generation of explanations and error correction, the last feature (error correction) happening to a greater degree in the team context than in an individual context. Sharing knowledge is one of the most defensible aspects of teamwork. In organisations, for example, knowledge is created or increased through exchange (of declarative knowledge, procedural knowledge, mental models) among employees (Banks & Millward, 2007; Srivastava et al., 2021). Knowledge sharing enables capitalization on knowledge resources and transfer of expertise (Kim & Choi, 2022; Perotti et al, 2022). Knowledge sharing could be inhibited by issues such as perception of it as a source of individual power (which is to be protected) coupled with perception of procedural justice within the group (Srivastava et al., 2021) or by politics-induced hostility to knowledge sharing (Chen et al., 2022) We refer to knowledge augmentation as another important benefit of teamwork. Augmentation might manifest in the form of a team member correcting error in knowledge of another. It might also present in the form of filling gaps in knowledge. Another way is in assisting a team member to link or connect different fragments of knowledge. Transactive memory system is a further consequential establishment of teamwork. In the transactive memory system, different members of a group bear non-overlapping information. Members of the group are aware of that fact and grant that an individual member bearing certain information possesses within-group de facto expertise on that information set (Rajaram & Pereira-Pasarin, 2010). Expertise might be due to prior experience or it might be delegated specifically to the individual by the team itself. Apart from reassurance that a member of the team can be relied upon in case of need, it frees up individuals to concentrate energy on delimited aspects of the task at hand. 634 KEER 2022 | 9TH INTERNATIONAL CONFERENCE ON KANSEI ENGINEERING AND EMOTION RESEARCH 2022 2.2 Collaborative inhibition and other handicaps Group work is not unconditionally superior. When it is a simple task, such as group recall by a group working collaboratively (e.g. studying items on a list), it is possible for performance of the collaborative group (on aggregated number of unique items recalled) to be lower than performance of a nominal group (a group of pooled individuals, where group members study items on the list individually/alone and number of unique items group members recall is aggregated). The phenomenon described is referred to as collaborative inhibition (Gadgil & Nokes-Malach, 2012 ; Rajaram & Pereira-Pasarin, 2010; Weldon & Bellinger, 1997). According to Weldon and Bellinger (1997), experimental data showed that "collaboration did not optimize individual recall." Individuals in a group recalled less than those who worked alone (p. 1165). Bakir et al. (2020) included lack of communication among group members as an important group work challenge. The authors did not treat "lack of communication" as a unitary construct. According to them, it included "not having enough communication with group members, not having enough interactions, initiating communication at the last minute, conducting low quality discussions, experiencing lack or poor generation and evaluation of ideas, and having conflicts with...peers" with no resolution (p. 80). Roberts and McInnerney (2007) presented seven common challenges that could be faced when group learning is used for instruction in a computer-supported learning environment. Given their list, it is reasonable to consider that all of the listed challenges can also be claimed as present in a person-to-person group learning or problem solving context. Not all of these are relevant, however, to our current project (e.g. membership/constitution matters such as selection of group members and withdrawal of group members). Two relevant ones are identified as follows: • Free ridership: In our study, we took account of this under the idea of social loafing. It featured as reduction of individual effort on our data collection form. • Possible inequalities in ability (and/or knowledge): In our study, we took account of this under the ideas of knowledge diversity and knowledge sharing and featured it under filling in knowledge gap on data collection form. While Roberts and McInnerney's listed idea and ours are not exactly the same, we opted for the positive characterisation (i.e. filling knowledge gap helps group) than the negative (i.e. there is ability/knowledge inequality). In that conceptual space, both perspectives become convergent. There are other concepts and renditions of dynamics of group work. Webber et al. (2019), for instance, noted "team composition elements such as personality, values similarities and differences, and team diversity" (p. 742). For our purposes, we hoped that values similarities and differences would be accounted for under the operational construct of personality difference (conflict or not), which we employed. Diversity could be interpreted as sourced in history-ofbeing which has shaped an individual's unique perspective or as sourced more particularly in the semantic domain (i.e. based on variabilities in concepts, etc. known). We hoped this latter would be covered by the construct umbrella of knowledge difference/variability among team members, which we have explored as a separate construct domain. The functional benefit (we hope) in our 635 KEER 2022 | 9TH INTERNATIONAL CONFERENCE ON KANSEI ENGINEERING AND EMOTION RESEARCH 2022 choice is that it better separates interpreted diversity into its appropriate construct camps as laid out above. 3 METHODS AND FINDINGS Based on review of the literature and personal experience, we identified variables which describe dynamics of team activity. For the current exploration, we selected nine variables which we set into a questionnaire. The nine are these: fill(ing) in knowledge gap (v1), correction of misinformation (v2), disagreement upon ideas (v3), sharing burden of work (v4), conflict of schedules (v5), diverse perspectives (v6), conflict of personalities (v7), social loafing (v8), buddy motivation (buddy system) (v9). After data collection, we dropped v3 due to its respondent misconstrual. A questionnaire with the nine items was created. Response to each item was to be recorded on an eleven-point scale: five levels in the negative direction from "low harmful" to "extremely harmful" and five levels in the opposite direction, indicating "helpful" from low to extreme. Levels were indicated with numbers increasing or decreasing systematically by one, each step. There was a neutral level (scale value of zero). The questionnaire was completed by 31 undergraduate students. We ultimately dropped two respondents due to arbitrary responses. (Sample item: "Research shows that when people work as a team, sometimes there is reduction of individual effort, compared to when working alone. Sometimes, that is due to the fact that those working less are depending on others to do the task. How much will that help or harm us?") A model was created in Netlogo (Netlogo, 2020). Seven levels of team were created (from a one-person "team" to a seven-person team). Teams were named after number of members. Variables identified in the questionnaire were inserted in the model as unique agents. At initiation of each model run, they were deployed randomly. (Each run represented length of a team task/project.) During model run, agents of each team moved about and, on encounter with a variable agent, recorded occurrence of that variable as an accompanying score for the team. The team with highest positive (or least negative) net score at end of a run represented the most optimal case. We ran each protocol (see below) 3,000 times. In protocols with weighting of variables, each variable value was specified as median score of that variable from our field data. Figure 1 is a succinct graphical chronicle of runs (narrative of transactions, scores and run log). We allowed the model to determine length of each run so as to provide for a range of possibilities. We allowed for team projects that last between one day and two weeks (fourteen days). We took each time step of the model to represent one minute. The model was allowed to choose a duration between 1,440 and 20,160 minutes. It was hoped that our large number of runs would yield a reasonable distribution of project duration. 636 KEER 2022 | 9TH INTERNATIONAL CONFERENCE ON KANSEI ENGINEERING AND EMOTION RESEARCH 2022 Figure 1. Biography of runs indicating dynamic outcome scoring per team 3.1 Protocols One of our goals is to compare different model designs (we call them protocols) in order to see if the effect of team size, if any, might remain consistent over different conditions of valuation of team dynamics. In other words, does the factor, protocol, have an effect? We compared four. PROTOCOL I: Incident valuation as [unweighted] unit/fragment of Fm (where summed value of all incidents per run = Fm = median value of observed scores [i.e. respondent ratings] on factor) We use the word, incident, to describe each instance (i.e. each occurrence) of one of the variables (e.g. an instance of personality conflict among team members) within a team during a model run. The incident is recorded as a model quantitative value by the model agent which has induced its occurrence. Group represents a hypothesised team (sizes vary from two to seven). Task represents a project completed by a team. This protocol was the first created (Stephen, in press). In the current exploration, we compare it with other models. This protocol looks at each group + task constitution (i.e. a group operative through one model run) as a closed system, where value assigned to a variable by survey participants (those who completed our questionnaire) denotes an absolute state: as if the group + task event (group in model run) was imagined as totality of a one-and-only and finite universe. In this protocol, the group + task event is conceptualised as a singular event, an isolated episode that manifests and is self-complete. It is a perfectly self-contained or self-sufficient occurrence indicated by the optimal value assigned to each variable. That optimal value, in this case, is the median. During model run, each instantiation of one of the constituting variables (e.g. fill knowledge gap) is only a share of the optimal value, Fm (think: total energy), of that dimension (i.e. each incident = portion of Fm). That instantiation generates what we denoted earlier as incident-asfragment (see name of this protocol/subtitle of this protocol). So, for instance, one occurrence of "fill knowledge gap" recorded by a moving model agent is only one "chunk" of total of all possible "fill knowledge gap" instances within the current closed system. Another agent within 637 KEER 2022 | 9TH INTERNATIONAL CONFERENCE ON KANSEI ENGINEERING AND EMOTION RESEARCH 2022 the group could record another instantiation during the current model run. Summation of all possible "fill knowledge gap" instances (Fm) is capped at a value--which, in our case, is the median. Universally, within this protocol, every encounter between any team agent and any factor agent, across all factors, records as an unweighted unit (as opposed to recording as a weighted unit based on questionnaire-derived weight of factor encountered). Signs are maintained: beneficial factors deliver positive values, unbeneficial factors, negative values. PROTOCOL II: In this protocol, every encounter between any team agent and any variable agent, across all variables, is also recorded as an unweighted unit (as opposed to being recorded as a weighted unit based on variable encountered). Signs are also maintained, beneficial variables releasing positive values, unbeneficial variables discharging negative values. In this protocol, however, aggregated incident values are not capped at Fm (median of observed, questionnaire value of variable). Team agent and variable agent encounters can keep occurring--and incidentsas-fragment (values) keep being recorded--throughout duration of the current run. PROTOCOL III: The protocol allows as many encounters as possible between team agents and variable agents. Each score on encounter is weighted by median value of the variable. In this condition, no ceiling is set for summated fragment scores (summated scores = outcome score). (Compare protocol IV below.) The protocol respects possibility of recurrence of a condition--i.e. the named variable (e.g. in a team of four, there could be more than four instances of correcting erroneous information; e.g. in a team of seven, social loafing might occur seven times or even more). It is a strategy which respects the fact that we are not able to precisely predict how often a phenomenon might occur. PROTOCOL IV: This protocol has same structure as protocol III above, but there is a ceiling restriction: that total summated fragment scores is not higher than weight of variable transformed by team size (i.e. median/weight of variable x size of team). The "size of team" stipulation allows that every team member gets a chance once on each variable (i.e. is represented by one encounter--even if, for example, one member encounters twice and another encounters zero number of times). 3.2 Outcomes Summated fragments of data (i.e. outcome score) were recorded for each team at end of each model run. It was noticed that there was a superfluity of zero scores presenting as outcome. That meant that either it was commonplace for incidents not to occur or they occurred often enough, but positive and negative fragments cancelled out one another. The effect of so many zeros was very severe, overwhelming rest of the data and resulting in steeply peaked distributions. Kurtosis coefficient was high in every case. (It is also efficient to note here that skewness coefficient was acceptable for protocols I and II, but was high--1.5 or higher--for protocols III and IV.) We sampled 100 runs (i.e. 600 columns of data over six teams). Only 3.34% of zeros represented incidents that occurred, had data recorded and then had positive and negative scores cancel one another out to result in zero. Eventually we chose to remove zeros from the data. 638 KEER 2022 | 9TH INTERNATIONAL CONFERENCE ON KANSEI ENGINEERING AND EMOTION RESEARCH 2022 After removal of zeros, all skewness coefficients fell below ± 1 (range of -0.22 to .73) and kurtosis coefficients ranged from -0.61 to 1.18. A later Shapiro-Wilk test presented nonsignificance in only three out of our 24 groups. Certainly, the Shapiro-Wilk test is a stronger indicator, but we held onto the skewness and kurtosis coefficient statistics (as well as reasonable q-q plots) as plausible arguments for assuming some degree of normality. We proceeded to remove outliers from our first data examined. Subsequent Shapiro-Wilk test still indicated only three groups (though different groups--not all same groups as above) with non-significance. Since there was no effective difference as result of removing outliers, we decided to retain outliers in the data. Levene's test for equality of variance indicated significance. So, our data did not meet the homogeneity assumption. In one more bid to explore possibility of re-shaping our distribution to meet testing assumptions, we carried out three forms of data transformation: square root transformation, log (10) transformation and data reciprocal transformation. None of them resulted in effective reconstruction. Our two independent variables (team size and protocol) induced interest in interaction. We largely failed assumptions for a two-factor analysis of variance. There is yet not a definitive nonparametric alternative (though Friedman test is sometimes used for dependent data) (Scheff, 2016). Given that our data are independent, we decided to apply Kruskal-Wallis test to each separate protocol in order to determine if a team emerges as most optimal per protocol. Results are presented below. Within protocol I, Kruskal-Wallis chi-square was significant at .05 level: Χ2 (5) = 11.43; p = .04. Epsilon squared < .01 (trivial effect, given Cohen's classification). Pairwise comparisons showed difference between team 7 and 6, team 7 and 5, team 7 and 3, team 7 and 2. Bonferroni correction removed all significance. Table 1. Test results of team comparisons per protocol Protocol n Test Statistic df p Adjusted sig. notes Effect size (E-squared) I 4,342 11.43 5 .04 n.s. † II 4,453 4.43 5 .49 n.s. † III 4,691 997.86 5 <.01 All sig. except teams 5 x 6 .21 (medium) IV 4,672 1,031.44 5 <.01 All sig. except teams 6 x 7 .22 (medium) Notes: Test = Kruskal-Wallis; † : correlation used for effect size is trivial (<0.1, Cohen's convention); E2 = epsilon-squared Within protocol II, Kruskal-Wallis chi-square test was not significant at .05 level: Χ2 (5) = 4.43; p = .49. Epsilon squared < .01 (trivial effect, given Cohen's classification). 639 KEER 2022 | 9TH INTERNATIONAL CONFERENCE ON KANSEI ENGINEERING AND EMOTION RESEARCH 2022 Within protocol III, Kruskal-Wallis test produced significance at .01 level: Χ2 (5) = 997.86; p < .01. Epsilon squared = .21 (medium effect, given Cohen's classification). Pairwise comparisons showed difference between all pair combinations. Bonferroni correction retained all significance except for teams 5 and 6 pairwise comparison. Within protocol IV, Kruskal-Wallis chi-square test was significant at .01 level: Χ2 (5) = 1,031.44; p < .01. Epsilon squared = .22 (medium effect, given Cohen's classification). Pairwise comparisons showed difference between team 7 and 6, team 7 and 5, team 7 and 3, team 7 and 2. Bonferroni correction retained all significance except for teams 6 and 7 pairwise comparison. Regarding potential for factor interaction (protocol x team), we provide figure 2. The matrix explores variability in team outcome score as conditioned by protocol. Due to weighting applied in protocols III and IV, we expected summated fragment scores (i.e. outcome scores) to be much larger than in protocols I and II. In order to facilitate some comparison, we applied a sensible stabilizing transformation by subtracting each score from group mean and dividing by standard deviation. (Essentially, this is a z-score transformation, but we use the handle of stabilizing transformation.) While resulting scores of protocols III and IV were still higher, values were significantly more amenable to comparison with those of protocols I and II. In matrix below, we depict lowest scores as root scores (minima) and highest scores as potential ceiling of scope of effect (maxima) of protocol factor. Figure 2. Potential effect of protocol on team: Outcome score minima and maxima Outcome score (min., max.) 640 KEER 2022 | 9TH INTERNATIONAL CONFERENCE ON KANSEI ENGINEERING AND EMOTION RESEARCH 2022 Observing columns, protocols I and II (1st column) and protocols III and IV (6th column) seem to depict similarity of configuration between each pair. That suggests that these protocol pairs are similar in terms of structure they have capacity to generate. When protocol I is brought against III and IV (2nd and 3rd columns), however, and when protocol II is brought against III and IV (4th and 5th columns), there is visible difference in paired graph form/behavior. That suggests possible interaction. There is appearance that as team size increases, efficacy increases as well. This is, as yet, descriptive from the data, and not confirmatory. By description shown in figure 2, it appears as if teams 6 and 7 perform at the most optimal. A very close look might suggest a very slight edge in advantage for team 7, but possibly a matter for further interrogation as these data stand. While these data suggest that team size might not matter at times (see similar pair, I & II, in table 1 above), difference in protocol design (particularly, when weighting is introduced) tends to suggest a different story (description around figure 2 above). 4 CONCLUSION In this paper, we examined two questions: Is it possible to identify an optimal size of a small team? Our exploration invariably took us to another question: If goal of small team work is successful completion of a task, is there a preferred system of assigning value to team benefits and challenges? Difference between unweighted encounters (protocols I and II) and weighted encounters (protocols III and IV) suggests that a combination of things might be of interest: 1. It appears as if there is potential, latent cognitive/emotional effect on outcomes through how data are perceived by respondents (i.e. potential team members). Degree of distress/trauma or assurance/inspiration by which team members perceive (i.e. associate with, indicated by differential scoring/weighting of variables in survey responses) might be a matter worth interrogating. Weight attached to a variable, when translated into a theoretical proxy and incorporated into model design, results in team-size difference--as seen in protocols III and IV. 2. Although examination of our data indicated that number of positiveand negative-bearing variables does not seem to matter (as summated run outcome scores were a mixture of both positive and negative values despite the fact that there were more positive than negative variables in our field data) it might matter that there was a seeming bias by respondents in weighting positive and negative variables: Participants tended to have given factors which favoured team success stronger scores in the positive direction than disfavouring factors were scored in the negative direction. This might have accelerated positive score gains, thereby enhancing growth of scores of larger teams with a larger weighting factor (as seen in rapid escalation of protocols III and IV). Although our results here are explorative, the question is still worth pursuing--of finding out whether there is an optimal small group size for some problem solving tasks (or whether task success is really dependent on other factors besides team size). It is hoped that future studies will be facilitative in addressing this core question within framework of attendant conditions. 641