Matthew Lee, Embedded Assessment Lead Researcher, Ph.D. Student in Human-Computer Interaction, Carnegie Mellon University
One of the many reasons observations of daily living (ODLs) are interesting is because people don’t normally pay much attention to them as they live their everyday lives. If ODLs easily lent themselves to simple mental accounting, we would not need special sensors or mobile applications to log them. Thus, when we present ODL data back to the user, they often encounter an odd feeling that mixes familiarity (because, after all, the data show their own behaviors) and unfamiliarity (because they have never seen this particular behavior logged and presented this way). This showcasing of the mundane allows people to explore and attempt to make sense of this new kind of data. Is there a process by which people explore their own ODLs?
We had a chance to understand this process in our pilot study with two elders and four months of sensor data about their medication adherence. We gave them each a special pillbox to use that could keep track of when they picked it up and opened the pillbox doors. When we showed them their data, our participants went through a process that involved three basic steps: 1) looking for anomalies, 2) explaining these anomalies, and 3) confirming their explanations with the details in the data. When first presented with their data, our participants wanted to find the anomalous behavior — in this case, instances of days when they did not take their pills. The visualization we used actually made it rather difficult to find these days where no pill-taking event occurred, but nonetheless our users were persistent in finding the days with missing pills. After finding these anomalies, they wanted to find an explanation for why they might have missed their pills. They first tried to find a benign explanation such as being away on travel, but they also looked for explanations that might show their own forgetfulness and lack of structure in their routines. And finally, after coming up with possible explanations for why they might have missed their pills on one occasion, they were able to use the low-level details in the data to confirm their explanations.
We observed one instance with a study participant that illustrates the process involved in making sense of ODL data: As she reviewed the data, the participant noticed that she had missed her evening pills on a Friday evening two weeks prior. She thought back and explained that she visited her nephew that day and had probably taken her evening pills with her and took the pills at her nephew’s home. She then looked at the detailed view of the data for the pillbox for that day and noticed that she had the door open for 20 seconds in the morning — much longer than she thought was normal. She reasoned that she must have been taking her evening pills from that box and filling her travel pillbox in those 20 seconds.
This same sense-making process, which involves identifying anomalies, explaining them and confirming them with other or more detailed streams of data, can also apply to other times when people are exploring their ODLs for the first time. Granted, our example focuses on only one data stream, pillbox activity, but this process may very well be expanded and developed to account for multiple streams of data and different types of ODLs.
For more details on this process of making sense of ODL data, read our CHI 2011 paper.
Learn more about the Embedded Assessment project.
Read more posts from the Embedded Assessment team.
