Chapter 7 • Visualizing Coordination and MLA Usage
7.3 Visualizations of coordination in CatchBob!
7.3.5 Commented visualizations
As a first comparison, let us examine the influence of the MLA tool on groups who made a low number of mistakes when drawing their partners paths (MM+: accurate Mutual Model) and who had a planning phase before playing the game. We compare a group with MLA (MLA: experiment 1, group 6) and another without (Control: experiment 1, group 15).
The first thing one notices on the visualization shown in Figure 65 is that this Control group took more time to complete the CatchBob! game, especially in the last phase. The reason why it is slower comes from the last phase. They seem to have trouble forming the triangle: the curves depict 2-3 cycles of moving away and getting closer. Nevertheless, the player represented above the horizontal line representing the caller seems to have more trouble than the one below this line (her curve has more peaks). We can however see that one player performed used the refresh tool a lot: the caller, since she was waiting for the others and trying to track where they were to join her. Since the Control group did not have an MLA tool, there is no refresh represented on the screen and we cannot compare the two groups on this subject. Regarding the messages, the Control group sent more messages: there are more questions and more occurrences of dialogues. The main difference also lies in the third phase in which Control players tried to compensate for the lack of mutual location-awareness by writing more messages. In
the two first phases, the content and the pragmatic status of the messages are quite similar and then in the third phases the situations are different.
Figure 65. Confrontation of two visualizations of groups who had and accurate mutual model and a planning phase: (a) with a MLA, (b) without MLA (Control group).
One of the most striking features is the fact that Control players needed to communicate in this last phase; this was not the case of the players who had the MLA, we only see two messages sent by two players. Control players all had to share map annotations because even though they were close they did not manage to coordinate easily in completing the game. What is remarkable in Phase 3 is the link between the type of coordination devices exchanged and the difficulty in making the triangle. Players from the Control group indeed sent lots of messages about signal strength; till the end of the game, they had to send messages that indicated their proximity to Bob, which was definitely not the case for players with MLA. Looking at the moment during which these players were sending annotations is also intriguing: Control players seem to send
messages only when they were close to each other (there are indeed no messages on peaks); this might be accounted for by the fact that the peaks correspond to moments when that they attempted triangle configuration, which did not fit and then immediately tried to refine it. Compared to the MLA group, the exchange of these map annotations was not as efficient as having the automatic MLA; this is also confirmed by the fact that they even had to set a face-to-face meeting (seen at the middle of phase 3) to refine their plans.
As we have seen in the first CatchBob! experiment, this visualization depicts the influence of the MLA tool: the inhibition of communication, fewer questions, fewer dialogues or a shorter third phase (during which the MLA tool made more sense).
Figure 65 concerned groups with or without MLA who made a low number of mistakes when drawing their partners’ paths (MM+). Now let us have a look at players with the MLA and a good versus a bad modeling of their partners paths. We thus selected two groups with MLA that match these criteria: a group that made a low number of mistakes when drawing the partners’ path (MM+: experiment 1, group 6) and a group that made a lot of mistakes (MM-: experiment 2, group 14). The resulting visualizations are represented in Figure 66. As we can see, the two groups performed the task in approximately the same amount of time, the only difference is that the MM+ group completed phase 2 more rapidly. The number of messages is very low in the two groups (they have the MLA tool, which explained this communication inhibition). However, the MM+ shows two occurrences of dialogue and acknowledgement whereas we do not see that for the MM-. We can also notice the communication asymmetry: there is indeed a silent player in the MM- group (the one above the horizontal line representing the caller) who never sent any message. Besides, most of the messages have been sent in the second phase, the one, which, lasted the longest. As for the use of the MLA, it is very interesting to see that MM+ players did almost no refresh in the last phase: their mutual modeling might be accurate enough so that they did not need to know where they were located. Most of the refreshes were also done by the caller player in the second phase, which makes sense since he called his two partners and waited for them to join him in forming the triangle. Finally, though the spatial behavior is similar in phase 1 and 3, MM- players seem to have trouble in phase 2: they move away from each other, get closer and one player moved away again. This shows that players from this MM- group had trouble joining each other before forming the triangle. We can consider this to be a miscoordination indicator.
Overall, based on the experiments results we discussed in the previous chapter, what this visualization seems to indicate is that an efficient exchange of coordination devices, with dialogues and acknowledgements, led to better mutual modeling within players of a group. When this does not occur, players may have trouble coordinating spatially, for instance, when trying to meet each other in the second phase.
Figure 66. Confrontation of two visualizations of groups with MLA and a planning phase: (a) with a good mutual model, (b) with a bad mutual model.
Finally, we are also interested in what happened when we suppressed the plan. We compared groups that gave a MLA and accurate mutual modeling. We chose two groups: one with a planning phase (Plan: experiment 1, group 6) and one without it (NoPlan: experiment 2, group 5) as shown on Figure 67. The main difference concerns the duration of the game: it lasted longer for the NoPlan groups, especially in the first phase. In terms of the messages exchanged, the NoPlan players sent a lot more messages, not only because the first phase lasted longer but overall because they compensated for the lack of planning phase. There are more dialogues and use of coordination devices about Bob’s position, direction and also more messages not conveyed by the symbols we proposed in the palette. This is logical since the palette has been designed with the coordination devices used by players who had the planning phase. Regarding the spatial behavior, the differences are also very prominent in the
first phase: players have trouble finding a proper strategy to localize the object: they move away from each others, get closer, and move away again and this happens 2 or 3 times. The second and the third phase are more or less similar in the two visualizations in terms of messages, duration and spatial behavior. This last comparison attests to the importance of the plan, especially regarding what happens in the first phase of the game. From all the comparisons we performed, the suppression of the plan is the most detrimental to collaboration and eventually led to worse performance. Nonetheless, it shows that the inhibitive effect the MLA tool had on communication in our second study is not visible here because the suppression of the plan made the exchange of communication devices absolutely necessary as this compensated for the lack of plan.
Figure 67. Confrontation of two visualizations of groups with MLA and a good mutual model: (a) with a planning phase, (b) without a planning phase.
7.4 Discussion
This chapter has described how we combined the different data sources from the CatchBob! experiment so that we can visualize coordination in a mobile context. To do so, we chose to go beyond a map representation and to integrate both logged and interpreted data to represent the exchange of coordination devices. This led us to define a data structure and a visual grammar that enabled us to generate these visualizations based on manual coding. The last section showed how we used these visualizations to depict mobile coordination and draw comparisons between groups from the different experimental conditions. They offered an interesting way to graphically illustrate the main conclusions of the two CatchBob! studies.
Visualizations are of interest not only to researchers and analysts but also to the users either after the game during interviews or during the activity itself. We can indeed think about real-time visualizations of the users’ activities that would enable what Jermann et al. (2001) called a “virtual mirror” of the group to both evaluate users’ contributions and to give the group a representation of itself over time. Of course, this does not mean that we should give those previous visualizations as such to the player; this might indeed be disruptive and complex. A simplified version of these visualizations should be designed with the idea of using the visual grammar and the data structure to give users an awareness of the group coordination. It could indeed allow users to see when they’re close to each other and whether communication within the group is efficient or if there is left over player. This simplified visualization would eventually take the form of an awareness tool that would go beyond MLA because the awareness it can convey would be more integrated, structured and less easily replaced by communication. For example, each Catchbob! player could have the size of his or her avatar modified depending on the number of coordination devices that had been sent. A timeline interface could also depict a player’s dispersion to the partners as well as the accumulation of coordination devices already sent.