Uncertainty Visualization Study Group Notes – 11/19/2012

Bill / Miriah Morning Recap

• 2.5 hour ‘intense conversation’
• Now agree about what we disagree about.
• Few days: write up few sentences about key position
• Actual will be somewhere between the two positions – there seems to be an interesting question

Lapinski Paper

• 11 step process: considerations
  • sequential checklist
  • laundry-list in middle

Important points made with regards to outlined process:

• (1) Identify Uncertainty Vis Task
  • Task-Context is Important
  • Difference between “problem statement” & “mission statement”
    • our goal: technique – closer to mission statement
    • fundemental problem is not create visualization, but to convey the information – assumption that visualization may help solve the problem
• (2) Understand the Data
  • [Bill] criticisms:
    • step has to be done at some point but more has to be done, can’t just propogate concept from this ste like they do in this paper
    • open question: does knowing the provenance of the uncertainty help in decision making? Does complexity just make it more confusing if we’re having trouble understanding it in the first place?
      • visualizing for system improvement –> possible; doesn’t seems to make less sense for end user in decision making
      • do certain contexts cause people to use it more rationally? if “optimal” statistical estimate, why not lie to person?
        • [Heidi] not all decisions are going to be Bayseian or fit into this neat format; real life not as compartmentalized
      • [Ross] agree for diff reasons: at some point you have to be willing to lump things together – can’t create a completely new visualization every time the uncertainty source changes slightly
• (6/7) Determine Specific Causes and Causal Categories
  • [Bill] not necessarily ‘wrong’, but asserted without support
  • the question ‘Does presenting context change people’s decision?’ is different from the question ‘Does presenting context help people make better decisions?’
  • creates categories from ‘measurement’, ignores modeling error sources in simulation
  • these particular categories may be valid encapsulation of measured data uncertainty sources
• (9) Prepare the Uncertainty for Visualization
  • difference: reasoning with uncertainty vs. reasoning about uncertainty
  • will need to ensure that distinction is clear within our project(s)
  • will care about different characteristics of uncertainty b/c problems are very different
  • [Bill] takeaway – if talking about causality have to be vary careful to distinguish who cares / benefits and why
  • possibility that uncertainty is actually getting folded into the task
  • can often solve the problem without uncertainty – may not be optimal, but can solve
  • if task is: ‘Do we need more data?’, do we represent uncertainty vs. represent consequence of uncertainty
    • boat identity example – uncertainty is data – “here’s bad data, fix it” is data quality issue, not really a meaningful uncertainty issue in the way we’ve been discussing
• (10) Try Different Uncertainty Visualization Techniques
  • need to make the reasons for choices (with regards to encoding) explicit
    • just decided on categorical representation for what is underlying quantitative data, and while that may have been the right choice based on context, that decision should be explicit