# 27.11.19 Meeting with Luke

## Interactions in diagrams
- Not supporting dragging people from on column to the other (changing underlying data)
- Shuffling graph, keeping one object in mind and see where it ends up next
- Group by gender, colour by ID (so using colour scheme dynamically)
---
- Pick out specific individuals, real life individuals, similarities between neighbours
- Are the majority of people who think like me older than me or younger than me? Would be an interesting question
---
- Control of visualisation should be through data
---
- More interested in clicking on single data points, not whole groups
- Or maybe groups of people who are similar
- -> which would resolve the problem with aggregates, means and so on
- Build visualisation from smallest elements of data, and then build aggregates from that, not other way around
---
- Keeping the real people in mind (we need means to look at large data sets and understand them, but the people are the true core)
---
- usually more categorical than quantitative or correlative
- example asking question about crowdsourcing, more answers one year later does not say that population has changed, we need to encourage this type of thinking
---
- Maximum similarity between people in one group
- Building a space of interactions and possibility, playing around with data, starting with one use case through, and then adding another
---- 
- selecting people, reshuffling
- Tell me what this person said, provenance chain
- How did we get here? Decision making process
- Show me data points that are similar
- Show me people’s interactions across a season (we can track phone numbers, but sometimes phones are shared, language shifts, recalibrate and re-ask for people’s demographic info again)


## Taxonomy
- Which parts are specific to certain visualisation and which are generally useful

## Questions
- How can we select similar groups of people?

- Advice:
- Paint up diagrams with mark up pens, come up with ideas for diagrams that could be helpful, send them over to Luke
- (He thought we would have more scribbled paper hanging around with visualisations, not just scratches for coding
---
- Our questions:
- Validating explanation drawings

## Stakeholders and goals
- Key stakeholder: citizens who are answering radio shows 
---
- process automation would be useful
- Building diagrams that support building a curiosity about the data
- Eliminate premature limitations, do not narrow thinking about the world
---
- Politicians: trying to build connections to people, empathise with people they represent,
- Are not comfortable with data analysis, but are comfortable with getting to know people
- Tool for building empathy with population, then develop that to do decision making
- Make our tool feel right, or at least not feel wrong
- Doubt anything that isn’t something that somebody said
- Grouping has to explain itself or show that it is up for questioning
- Being professionally competent at going from people’s goal to decision making
---
- Two groups: “technocrate” and “politician”, two mindsets/capabilities
- Numbers needed to defend empathy, should support arguments and intuition, not build it (doesn’t help building it)
---
- Top arch on new system diagram is more interesting
- AV database: is a gigantic python object :D
- True data is only the structured data, not the database
- Eliminate database, because this is only representation again
- But structured data object is too big to put into browser, so we’ll need a querying process
- User interaction could be programming, not just user interaction
- Different users might have different methods and user interactions, separate diagram into user groups exploring phase /showcase mode
---
- Looking through AV reports for used diagrams
- Weird diagrams may be bubble charts with number of people correlating with bubble size
- Coming up with brand-new visualisation is difficult, but maybe looking at diagrams in reports and add new user interaction is easier

## Workflow from data to visualisation
- Radio show, answers, follow up for further information
- Tool going through messages, doing categorical 
- People defining themes, tool again puts messages into these themes (more water, more policy)
- Python generate two tables: what did they answer by demographic, and?

- -> Talk through process for one example
- What are standard things, what are not standard things, but we’re not interested in standard things