Visualizing Uncertainty

Tara Kola

DREU 2015

The Passage of Time

Week 1

I spent my first week familiarizing myself with the problem space I would be working in. My mentor, Jessica, gave me a list of papers to read to learn more about the field of visualization and the specific problem of visualizing uncertainty. My favorite paper was "Displayed Uncertainty Improves Driving Experience and Behavior: The Case of Range Anxiety in an Electric Car" by Jung et al. It describes the effects of displaying a gradient on the range meter instead of an exact value for remaining number of miles. The paper illustrated for me the importance of showing users uncertainty, so that the trust a user places in a device’s prediction isn’t misused. Uncertainty is a part of the future, and predictions attempt to create a clearer picture of the future. Thus, not showing the uncertainty in a prediction is a fundamental flaw of the many systems that simply show a point prediction.

Apart from reading papers, I also worked on setting up my dev environment, and started looking at how we can collect historical data using the OneBusAway API in order to create a model that predicts a probability distribution for the arrival of buses in Seattle.

Week 2

This week I worked on setting up a script to get the historical data, and collected information along with my team on when buses actually arrive at bus stops to check how reality compares to the data we get form the OneBusAway API.

My team and I also established the schedule for the rest of the summer. Our goal is to run a survey to get information on how OneBusAway is used. We will meet for a few design brainstorming sessions. We will then shortlist our designs based on our survey results, and test our designs in the lab. We will then further refine our designs and test them on Mechanical Turk. Based on these results, I will code up the designs in Android. This alternative version of OneBusAway will then be further tested, refined, and deployed in September.

I spent the rest of the week setting up the database to store the historical data, and thinking about the probabilistic model that we need to develop for the bus arrival data. I also started testing my Android development knowledge, which has been dormant for the last year, so that when it is time to code up the designs, I can do so efficiently.

I continued reading relevant papers, and jotting down design ideas for our design meeting next week.

Week 3

This was an exciting week. I spent the first half of the week working on the final designs for the probabilistic model and the database. On Wednesday the team met up for our first design brainstorming session. I then spent Thursday analyzing our designs and categorizing them according to information representation systems and features. We identified areas where we need to work on further designs. I will now be working on the second round of design drafts based on our discussion.

Apart from designing databases, algorithms, and visualizations, I also worked on creating a survey for OneBusAway users to help us identify user goals so that we are designing with our users’ interests in mind.

Week 4

The results from our user survey came back. We learnt more about what users are looking for when they use OneBusAway, which will help us inform our designs. I was able to create a design matrix to evaluate our interfaces against how conducive it is to the user behaviors we discovered in our survey. We continued to work on the model to refine it and improve it. We began running a scraping script to get the data we need to run our predictive model.

Week 5

This week, we picked a final set of designs. I used these designs to produce digital mockups of our interfaces - including multiple types of layouts and representations of uncertainty. I also began working on the app, and thinking about the experimental design for the upcoming studies.

Week 6

We used this wake to decide on a final set of layouts. We then began discussing various types of uncertainty representations. I began testing these, and developing mockups based on real data in R. We now have a set of designs that we can test on real users.

Week 7

I've started walking around and showing our designs at bus stops. I have been doing informal tests to see how people are responding to the designs. We have found that there were some small considerations we hadn't accounted for previously. I then began adjusting the designs according to what I heard in the bus stops. I also began designing a survey to test perceptual understanding of the various representations of uncertainty.

Week 8

We began evaluating the statistical inferences various types of visualizations allow for. This allowed us to refine our set of designs we will test. We have been going into the details of the study design - fixing the wording, thinking about how to organize the study. In addition, I have picked up working on the app that we will test in the fiedl study.

Week 9

I have learnt quite a bit about how difficult it is to get a study just right! We are continuing to work on the study. I'm amazed by how much time it is taking to get all the artifacts in place, but it is exhilirating to know that this study will soon be live, and we will be able to answer tangible questions about perceptual understanding of uncertainty.

Week 10

I am extending my DREU by 2 weeks so that I will have time to finish up the study before I leave. At present, we are continuing to work daily on the study and the app. It has been wonderful getting to do research at UW with this group. I enjoyed the research experience, and it has confirmed my interest in doing research in the future. Thanks CRA!