New round of coding: what do we see?

[alberto]

I see that @Amelia has made significant progress in coding the opencare convo (at the time of writing, 3,498 annotations with 779 codes). The co-occurrences graph has changed quite a bit. Setting the number of co-occurrences to 4, we now get a much richer structure.

1. The community of codes around "mental health" is now connected to the main community via two bridges rather than one: "health and art" and "trauma". 
2. The main community maintains the two central codes "research question" and "community-based-care", but it now has several more hubs.
3. "Legality" seems to have its own community now. It links to the two main hubs, but also to "sustainability", "skill sharing", "regulation", "safety", "existing system failure", "outside existing systems". The latter four are only connected to "legality".
4. I also see repetitions. For example "water" is connected to "water" (4 times?) and "existing system failure" to "existing system failure". @jason_vallet is this a bug?
5. Amazingly, at co-occurrences >= 5, the graph still has structure and is still legible. 

Amelia, @Noemi , do you have an interpretation for this?

[amelia]

I have been seeing a common theme among a lot of posts ---- how existing legal frameworks don’t allow for the kind of innovation required to overcome a problem. Concerns around this usually pull two ways---- first, sharing tactics to get around these legal issues, and second, a discussion or question on whether it is safe to do so. In some cases, the law is there to ensure people’s safety. In others, it impedes it. (This is in part what I wrote on in this piece) It’s a complex issue that people have been discussing across different health and social care problems on the site.

Mental health, too, is a key issue being discussed all over the site. Often it is tied to trauma, which is in turn tied to migration in a lot of cases. But lately I’ve been seeing an expansion of the discussion on mental health— not just focused on refugee issues, but also mental health issues across populations. Youth come up frequently, as do  feelings of hopelessness in the face of the current labour market. There has been a repetitive feeling of disconnect between one’s own dreams and desires and the way cultural and social norms, as well as existing labour and $ structures, require one to live and attribute value to activities.

Most of the solutions, at least as I have observed, are community-based. Modern life is extremely individualising, and for both mental health problems and feeling trapped by a legal system, reaching out to others for care, support, and ideas provides a way out. Alternative ways of living, thinking, and dreaming are put forth by communities, and allow individuals to imagine different futures.

There seem to be technological solutions for this kind of community-building — mainly, platforms for bringing people together and allowing them to share knowledge.

Those are my initial thoughts and interpretations!

Two final notes:

For some reason, duplicate tags frequently get created for the exact same word or phrase (I’m not sure why this happens). I can merge duplicates, but it requires going to the taxonomy manager each time so I usually only do it after I complete a large enough batch. It would be great to know why this happens and get a fix for it.

Research question should be separated out as it isn’t a true code, just a way of aggregating questions that community members are asking.

[amelia]

merged duplicates

I’ve merged the duplicate terms---- @Alberto take another look and let me know if you’re still seeing double

[alberto]

Still seeing double and what to do with “research question”

At least 2: “water” and “existing system failure”. But not sure @Jason_Vallet refreshed his data before or after you merged the duplicates.

About research question.

We are doing data science here. For the software, a code is a code. There is no way to drop codes just because, and probably there should not be. When we do batch processing with Tulip we can of course do this kind of thing, but I am a bit wary of introducing ad-hoc stuff in the actual application.

I think it comes down to good-practice coding. It could be that having “pseudo-codes” is good practice: instead of maintaining a separate file called “questions proposed by informants that the team could work on”, the ethnographer assigns a code like research question. This keeps track of the same thing, but it does it within the OpenEthnographer environment. In this case, we would probably need to add a field to the code itself, or, if that is not supported by Drupal taxonomy terms, to the Annotation: just a “pseudo code/annotation” box to tick. Then we would add a line of code that discards those codes and those annotations for the purpose of building the graph.

If, on the other hand, having pseudo-codes is bad practice, then we should keep track of that information somewhere else, in order to not contaminate the consistency of the data. @Amelia , what do you think?

[alberto]

Check this out!

When @Jason_Vallet regenerated the database, he included some newer annotations that previously had been left out of it. The co-occurrences graph for co-occurrences >= 4 now looks like this:

It reflects your answer very well, @Amelia . mental health has moved yet closer into the center of the graph (in November it was connected to it only via “skill sharing”). legality is still in the same position, as the only hub connecting some in-graph codes like autonomy, sustainability, community-based care and migration ; and some out-graph codes like existing system failure and, yes, safety.

Did you reply based on browsing the graph, or was it just off the top of your head, based on having done the coding?

The recent expansion of the discussions around mental health is, I think, tied to the onboarding workshop we organised in Galway in December 2016. Apparently the West of Ireland has an endemic suicide problem, especially among men. This has raised public awareness of issues like depression. So a lot of the people doing open care there are focused on mental care.

Finally, there is an interesting regularity. Try “stressing” the co-occurrences graph, by imposing that the number of co-occurrences sustaining each edge is higher than 4: at 5, 6 and even 7 the graph seems to be telling me the same story, although it gets more and more barebones-simple as you increase the minimum required edge strength. At co-occurrences >= 7 you get this:

What are we talking about here? Well, if you ignore research question, as we should, community-based care becomes the center of the graph. It connects to legality and to resource strain, itself connected with crisis and migration. The latter is then connected to two codes representing solutions, or at least concepts underpinning solutions: building relationships and story sharing. mental health and open processes are each at the center of its own component: they are very important concepts (edge strength is at least 7!), but with a weaker connection to community-based care than codes in the giant component. I don’t know if this is an artefact of Amelia’s approach to coding or the signature of a coherent conversation, coded well. What do you think? Are we looking at Amelia thinking, or are we looking at the community thinking? Amelia? @Federico_Monaco ?

Maybe the way to present these (secondary) data is to start from a very high edge strength (5-7) to give ourselves a high-level description, and only then go on to lower our requirements on edge strength to discover new codes and new connections. It would be like looking at volcaninc islands emerging from the ocean: as they are pushed up, what looked initially as a set of disconnected islands reveals itself as the peaks of the same mountain range.

[amelia]

Quick reply

I didn’t look at the graph at all when formulating my original response, so this is interesting!

I like the last thing you said---- the  volcano metaphor is apt, and I think the approach makes sense given how many codes we are working with at this point. There’s always the tension between presenting a single problem very robustly and trying to see what’s going on/what connections are made/what is common across different problems presented across the site.

If I’m doing my job right we are looking more at community thinking rather than me thinking. It’s helpful to have one ethnographer (it’s going to be interesting to see what happens when we try and expand to do more collective ethnography) because I have a standardised vocabulary across codes. (e.g. ‘migration’ rather than all the other synonyms one could use instead). Part of cleaning up the data will be streamlining codes that mean the same thing, while being careful not to over-correct (subtle differences are crucial and informative, like the difference between ‘resilence’ and ‘sustainability’). Also important to note that mental health is higher up in a hierarchy that includes codes like ‘depression’ and ‘suicide’ nested underneath it. Same with ‘therapy,’ which has ‘group therapy’ and ‘online therapy’ nested underneath it, for example. So once I do another pass at organising hierarchies we could use that to enrich our analysis.

[alberto]

Very cool

Great. So it works like this: you code away, and build a story in your head, that eventually will become the ethno report. Every now and then, you check the graph and see if it tells the same story, and adjust as needed. Eventually the whole thing is finished, and the graph tells the same story as as the report. In fact, it is a visualization of that story.

At the next consortium meeting I would like to do some user testing. Bring qualitative researchers in, show them different graphs (as we developed in the latest Masters of Networks) and see how they process them, which one is most explanatory.

[federico_monaco]

the fish and the net

Thanks for mentioning me.

All this is very interesting and if you remember in Milan @bpinaud and I were looking for as highest as possibile edge strenght visualizations in order to understand how these “twin peaks” were behaving and what could they mean. Also the fact that co-occurences happen among comments only were something to work on. In answer to your question @Alberto i tried to check some literature and i found something about transitivity , even though @Amelia 's answer is clear on the method she is following and how she is considering good biases in the emerging of meanings by coding.

[alberto]

Something there…

Applying the idea of (semantic) transitivity to the co-occurrence graph is intriguing, but I did not understand the paper by Tanaka-Ishii and Iwasaki. @melancon , what is a “branch” in graphs? How do people know that a branch is “anchor”?

[jason_vallet]

Well, not a bug exactly but still some unwanted by-product

As Amelia has said, some duplicated terms seem to appear from time to time and if some update is performed before the duplicate terms are merged, we end up with a duplicate in the graph database as well. This seemed to have happened at least twice as you have pinpointed (e.g., tag water with id 3002 and the same duplicate tag with id 1209).

At first, I thought this was due to different tags existing with the same label, some previously created by Inga and some by Amelia, but it doesn’t seem to be the case as far as I know and the “clone” tags just mirror the results of the original one, so something is not quite right here.

I will perform a full flush of the graph database and rebuild everything from scratch next week-end to give you a clean database for the DSI Fair. Some tests to check for duplicates when importing the tags will also be add to avoid the situation in the future.

[amelia]

Thanks, Jason