begriffs

Tracking Joy at Work

March 15, 2015

You can’t improve what you can’t measure, and what better to measure than an activity that consumes your waking life. Work.

Work, that oft-maligned but secretly important part of having meaning in life. Why are some days frustrating and others swept in effortless flow? Are bad days shared among an entire office or suffered individually?

I’ve created a program to help answer this question. Throughout the day at random intervals it privately sends me and my coworkers at Loop/Recur messages on Slack to see how we’re feeling. We respond on a scale of one to five and the results are displayed anonymously on a realtime dashboard.

happiness dashboard

happiness dashboard

Just asking this basic question reveals a lot and has other good side effects. Its simplest effect is to make us reflect on our mood. Quantifying how we feel raises us above inarticulate reaction. Also we can spot bad moods that we all share because they appear on the dashboard. It sparks helpful discussions.

The technique of sampling moods or other subjective reports over time is called the Experience Sampling Method and has been done since the 1970s. It’s been getting easier to conduct this research as communication and data collection become more accessible to computers. Not so long ago pagers were the best way to alert people to respond.

While developing the program I consulted a fascinating book called, “Experience Sampling Method: Measuring the Quality of Everyday Life.” I wanted to be acquainted with the methodology behind these kind of experiments. Whether or not you’re interested in data science I’d recommend you check this book out. It’s full of studies that talk about things that are fundamental to human happiness, from family life to work to gender differences.

The three common types of sampling are event contingent, interval contingent, and signal contingent. The first is where the participant makes reports during a certain type of event. At our software company it might be after committing code to git, or coming back from lunch. The second is when the participant makes recordings at the end of large intervals, like at the end of the day. We chose the last method, signal contingency.

Signal contingent sampling is where participants get random prompts to record data. It helps avoid memory biases and measures all computer-based activities without preference.

To run the system we have a daily scheduler. It picks three random times from a uniform distribution inside each user’s time preference window (the hours during which we each prefer to be solicited.)

The randomness of the sampling is interesting. Sometimes I may get solicited twice within a few minutes, other times more evenly throughout the day. We have considered enforcing a more even solicitation schedule (making it impossible that they occur too close together) but we don’t have a clear justification for the change. For the time being we’re leaving it as a uniform, sometimes lumpy, distribution.

We don’t yet have enough data to make statistically significant claims about how we feel, but it has sure been fun keeping an eye on the dashboard. Many companies have big screens showing web analytics, but we’re the first I’ve seen to track how we feel – and it feels good.