Candy bag, 2011

The study followed a bag of candy through
people's own days, with a sensor on the bag.

Its anchor is a thing, with people and places beside it. Its model of activity is how a bag is eaten.

“Instrumented candy bags, supplemented by location information, participant video, and interviews, offered a view of consumption across times and situations.”

McLeish, Modeling Activity, p. 4. The fieldwork ran in May 2011, with in-home interviews in June.

The question here is whether it worked,
and what kind of picture it gives.

• Could a sensor on a bag follow eating through ordinary days?
• Do its readings line up with what people showed on video?
• How far does it reach on three things: one sitting or not, at home or not, poured or picked?

The team's proposal asked when, where and how the candy is eaten, down to the gestures that take it from the package. These three questions are this draft's framing of that.

Each person's kit held a sensor pack,
a GPS receiver and a pocket camera.

The team's kit script and participant field kit, 2011. The kits and packaging carried the client's brand, so their pictures stay out.

The sensor pack read touch on six copper plates
and movement on three axes.

It wrote a row on any change: about eight a second while moving, and one every 10 seconds otherwise.

TJ designed and built the logger, and wrote its firmware.

Fig. 1 · Sensor packs being assembled, May 2011

Firmware of 13 May 2011. Photograph: the team's, from the study's kit folder; the board at right is another version.

There was one pack to a person, bag to bag,
so a bag's life depends on the team's labels.

• Each person stuck the pack on with adhesive dots, moved it to the next bag, and charged it at night.
• The readings do not know which bag they come from, or how much candy is left.
• The team's table labels 63 stories from 16 people, which this draft reads as 40 bags.

The team's kit script and instruction videos; the loggers' files. Reading the labels as bags is this draft's own: a new bag where the size changes, where a story opens one, or after a story that finished one.

The pack's rows and GPS tracks survive,
with the videos and the team's labels on them.

The study's kit folders. No names, ages, addresses, map or pictures of people are in this draft, and everyone appears under a code.

The stories show in the readings,
though the pack was not always on the bag.

So the clocks line up, and stories have a signature. Of those 22 stories, the logger was off for 7.

Readings within 15 minutes of each timed story in its person's study window, against the same clock time on that person's other days. These few people, not a population.

A session is handling, which includes eating
and is not the same thing.

A session is a rule added here: moves and touches with gaps of five minutes or less, holding at least three touches and a move.

The daily check-ins, yes or no only, on days with 12 hours of logger before the question. People also carried the bag and handled the pack to charge it.

The model asks three things of a bag,
and the readings reach two of them in part.

The labels are the team's, by hand, with “Unclear” where it could not say. Home in the GPS codes is the place with the most dwell between midnight and 6 am, a guess.

The readings do not tell pouring from picking;
that label rests on the video.

Only 26 stories have readings near them and a clear label: 21 poured and 5 reached, from 11 people, none of whom has both. So a difference in labels can't be told from a difference in people.

Added here. With whole people held out, no measure beat always saying “poured” by more than one story of the 26.

The model stops where the data stops,
which is part of what the study shows.

“An observation is not an activity model.”

“In each case, the act of instrumentation revealed the limits of what was directly observable, while modeling provided a way to consider what those observations might mean.”

McLeish, Modeling Activity, pp. 3 and 4. The paper is of 2026, and speaks of the Iota studies generally.

These are a few people and bags,
so the numbers do not measure a population.

• Sixteen people with labeled stories, over about ten days.
• The labels describe videos people chose to make.
• Plates could stick on for hours, and 22 of 59 stories have no movement near them.
• GPS has no code for 25 of 59 stories, and home is inferred.

The study's kit folders; this draft's tests. Purchases are not in the readings.

The sources are the team's own files and tools,
with the client's name left out.

• The sensor pack's firmware, May 2011; the team's data viewer and its library, 2011, from which the move rule is ported. The port returns the team's own motion counts for all 16 people it ran.
• The loggers' files and GPS tracks; the team's table of story labels, July 2011; the online study's check-ins, yes or no only; the kit script and field kit; the in-home discussion guide.
• McLeish, Modeling Activity; TJ's answers, 2026.

The study was the Iota team's, for a confections client. TJ designed and built the logger and the data viewer. Ann Kauth moderated the online study.