- keep doing the analysis: sequential analysis + nasa tlx + overlaps/backchannel/dispersion/covariate with bob's position + regression/multilevel modelling...- when looking at means differences, we have to be careful about the differences. For instance, a difference of 1 units (i.e. between 4 and 5) even significant must be stated carefully. The amplitude is not that big. - there might be an inertia caused by knowing where are the partners. - the performance measure in this task is the group path length. HOWEVER we did not really design the game to have a clear performance measure (this was just meant to find winners), the REAL performance if the accuracy/quality of participants' model of their partners' (intents, goals, activity). - the (planned) strategy, if well established imply less coordination on the field. AND coordination implies awareness of others OR messages. Now there might be a relation between the (planned) strategy and awareness/messages. Awareness...Messages... Errors BUT there are also link between awareness and strategy, awareness and message, awareness and errors. - That is why we need to discriminate the different strategies groups had (what I sketched on Pierre's whiteboard: three different group path + the fact that some groups stick to their strategies and others reshape it + when people converge to Bob, some go directly and others still wander around). - Discriminating people's strategies lead us to find descriptors: if we wanted to replay the games, which paramaters would we need? A simulation of this would imply to play with these parameters so that the agents use the same strategies as the ones encountered (emergence?): dispersion at the beginning + a certain speed to converge to bob. - we want to access to groups' strategies! Which indexes: angles made by the 3 persons? dispersions index (then have a graph: y=mean distance between the 3 persons, x=time) + we can define the EPFL as a graph/network amde up of different PLACES connected (like CO -> Archi, Elec, Unil, Esplanade....). So: 1) represent the EPFL as a graph 2) Draw players' paths (groups) thanks to 2a) path logs 2b) autoconfrontations, 3) Define categories (2/3) like for instance if they explore this graph in width or height: that may define strategy descriptors. - FIND literature about how can we describe people's trajectors: look at rat experiments in psychology. How can we analysed this?
- does the presence of the awareness tool trigger other/different strategies? What I saw is that people with AT stick to their strategy and the others modify it, reshaping it by annotating the maps.
- a mobile representation: strategy depends on: A's position, A's signal, B's position, B's signal, C's position, C's signal. In the condition with awareness tool, people have everybody's positions + potentially access to everybody's signal. In the condition without the tool, people have just their position + potentially access to everybody's signal. Agents use those inputs to build their strategies. We'll have to use them to simulate the task