Multi-level analysis of distributed learning, uptake

Dan Suthers, A unified framework for multi-level analysis of distributed learning LAK11: http://blip.tv/solaresearch/8_dan_suthers-5692149

Dan works mostly in computer supported collaborative learning. This work came out of doing analysis of interactions in small groups, face to face, online, multiple media and put this together somehow? He was also supported an online community of teachers (TappedIn. Representation of data and analysis enabled the scaling up.

Many theories about how learning happens in social settings:
1. the individual is stimulated by the social setting-social as stimulus to social entity as learning agent
2.Castell's networked individualism
3. Jeremy Rochelle's maintaining a joint conception of the problem
4. how much coupling is needed between learners
diffusion of innovations
knowledge building
they all have in common--contact between people must occur for learning to happen
interaction-not necessarily conversation--maybe Uptake better concept

UPTAKE-the act of the actor taking something relevant from what someone has done before as being relevant for your own ongoing activity (downloading files, etc.)
evidenced by what we can see directly-people's actions in their environment

Analytic Challenges
interaction between the individual, small group and collective agency
requires multiple level analysis
practical challenge-- studyuing learners working in whiteboard and a threaded discussion--interaction is spread or distruted over multiple media--put it back together
traces of activity are fragmented-how to create the whole in the analysis?

ties--not at all like log data-want to unpack what the ties mean
want to put together this trace of interaction

system: that collects things together in and puts things together in a single
annotated artifact and analyses it.

using db or log files (http) want to figure out what is happening
1. understand what entities we want to see and the relationship between them
ie threaded discussion, has some features to illustrate
2.the granularity at which things are recorded at may be different than the granularity of what you want to look at
message may have 3 logs for it, threading relationship
person 2 has written msg 2 and written msg 3 snd then posts something--now more abstract version as a transcript rather than log files-- wikis etc -unify the record of different media.

3. what about interaction? adjacency carries-assumption-
4. construct contingency graphs-- identify empirical relationships between events that collectively evidence uptake
called contingencies after Garfinkel's "contingently achieved accomplishments" how actors draw on the evolving context.
not truly proof of interactions, when people do things they draw on the resources of their environments

5. did by hand, now figuring out how to automate:
contingencies: media dependency

6. contingencies: Lexical or Semantic Overlap: also contingencies between the read events and writing of messages --events that are near each other may be related, events by the same actor,
for example, reuse of noun phrases ( contingency graph showing contextual action mode) graph of entity-relations of discussion 1 and all actors.

7. end up with: Contingency Graph as Contextualized Action Model
•analytically relevant manifest relationships between the actor's actions and other events that have been recorded
•Next:raise the analytic level of description to latent relationships and higher order structures.

8. be selective in what contingencies you put in--could become pretty complex
all above are data structures that are manipulated computationally.

•interpret collections or subgraphs of contingencies as corroborating evidence for uptake
•supports sequential analysis of interaction

Uptake Graph--an Interaction Model
possible automated way to find the uptake in discussions-- way to find the potential for learning- a structure you can look for in the data structures
micro analysis of transactions- manually, now trying to automate-especially highly interactive discussion

Next Layer--abstracting away from the sequentiality of the events-
affiliations of people through media-- Accociograms
directed affiliation network of actors and artifacts
mediation model: how actors' associations are mediated Latour
this largely factors out time--looks at mediated interactions--finds round trip between actors-- wouldn't see this in threaded discussion structure-round trips are important-dialogue is how learning happens in groups

Relationships
patterns of mediated associations reveal relationships
dialogue pattern-round trip
consumer pattern P3 reads everything P2 produces

Multi-media associations
characterize pairwise relationships in terms of distribution across media
compare roles of various media in supporting associations (suthers and Chus, networked learning, 2010)

cluster analysis
compare roles of media bridging between groups
transitive closure,
Ties--
straightforward to collapse into sociogram by transitive closure or similar computations
mediated associations
SNA methods can be applied to the sociograms

this framework allows potential automation of representation of data to do analyses on-- interpretation of analyses
multi-level analysis

Prior research
contingency graphs are used for:
microanalysis of process through which learners achieved an insight
semi-automated analyses of graph manipulations to find pivotal moments

currently applying this to TappedIn, longest running network of educators

latours idea-following the actors
Advantages of this framework:
as a data representation
integration of distributed data: uncloak distributed interaction
common format for reuse of algorithms
as an analytic framework
multi-level multi-theoretical analysis possible
multiple ontologies allow for mapping between interaction, mediated affiliation and tie levels of analysis