Matthew Cornell

November 5, 2010

Identifying patterns is crucial in experimentation because patterns can indicate useful correlations. After all, the whole point of experimenting on ourselves and collecting data is to find ways to make changes that help us to be happier, and patterns tell us where there are points of leverage. Patterns should make us curious, and we should pay attention to them.

When analyzing our results both during and after a self-tracking effort, how do we find patterns? Unfortunately the answer is not straightforward. On the one hand our brains have evolved marvelous pattern recognition facilities that connect the dots and make meaning out of what we see in nature. In fact, human vision and verbal language are still far beyond what computer researchers can write programs to do. And our system is programmable. Just try this trick: Look around where you’re sitting and quickly scan the room. Now think about the color red and re-examine the room. Notice how everything red jumps out. Similarly, we’re easily able to recognize a face or voice in a crowd when we filter for it.

However, our ability to identify patterns is fraught with errors. This is because evolution has conservatively favored finding patterns when they don’t exist, rather than the opposite, which would hamper survival. So although we can easily find patterns, it is very possible that the two things we think are connected are not. Finding the face of the Virgin Mary on a grilled cheese sandwich is a fine example, as is falsely hearing the phone ring in the shower, hearing hidden words in a song played backwards, and seeing shapes in ink spots.

So where does this leave us when it comes to making sense of our personal data? I can think of two approaches. The first one (innate human reasoning) is readily available and generates intuitively satisfying results, but is prone to inaccuracy. The second approach (statistics) is far more accurate but more complex to apply. In fact, the field of statistics was created exactly because our human ability is so notoriously poor.

However, I think it’s still possible to find useful patterns even if we’re not good at it, and even if we don’t have the knowledge to apply sophisticated statistical models (I certainly don’t). So how do we reconcile that with our limitations? What I’d love to have is a handbook of strategies for discovering patterns in our self-tracked data. Though I’m ill-prepared for creating one, let me naively toss out some general ideas and see what you think.

How to look for patterns

As I described above, we must first be clear with ourselves that any connections we tentatively find within our data are quite possibly invalid. This means we need to question them, ask whether we have enough of the right kind of data, and continue to test them if necessary. In other words, we need to be good scientists.

Generally our goal is to find connections in the data and to look for cause and effect. To do this we look at each variable we measured and ask how it might relate to others. For personal experiments, two general kinds of causality are temporal (something happens followed consistently by something else happening, such as “My mood goes south the day after I drink alcohol”) and spatial (something that consistently takes place at a certain location or set of circumstances, such as “I feel happy when I’m riding my bike.”).

Sometimes changing the data around makes things more evident, so it can be helpful to use a visualization tool. Even the charts in your spreadsheet program might be useful. At the highest level we might reorder the data (say by quantity or magnitude), look for sequences (sorting by time), or look for repetitions or groupings. You can get more specific with your particular data. Injecting a big dose of creativity is key.

Factors to consider

Following is a Socratic-inspired set of questions that a friend might ask if you brought your data to her.

Space/environment:

• Where were you?
• Where were you going?
• What was going on around you?

Activity:

• What were you doing?
• What events were taking place?
• What did you do before you got here?
• What changed right before? Right after?

People/relationships:

• Who were you with?
• What interactions were you having?

Personal:

• What was your mental/psychological state?
• What were you thinking?
• What was your physical state?

If you can’t find useful patterns

If you are experiencing a lack of patterns, then there are two possibilities. Either 1) the pattern is there but you’re not seeing it, or 2) the data you’ve collected doesn’t manifest any patterns. To address the first problem you should bring a fresh pair of eyes to the data set. Often we become too close to the data and get stuck looking at it one way; others might see it differently.

For the case where both you and your collaborators are convinced there are no patterns evident, you need to go back to the experiment’s design and change the data you measure. We’ll save for another time the details on how to create a design, including selecting variables, but beware of throwing out the entire set of variables and starting over from scratch. Consider whether there is one more piece of information that might give you some insight, or one small change you could make in what you’re already tracking. Alternatively you might decide that you’ve tapped the experiment for all it’s worth, and move on to more fertile ground.

The joy of patterns

Finally, not only are they potentially helpful, patterns can also give us joy. After all, art gives us pleasure due to the beautiful patterns artists create. Glowing colors in a painting, rhythm in music, and the tactile pleasure of a fine weave. By being on the lookout for lovely patterns in your world you can bring yourself into the moment and appreciate the life around you.

What do you think?

I would love to hear your thoughts on this. What strategies do you use when finding patterns? How successful have they been? Is there a general set of heuristics for finding patterns in arbitrary data?

[Image from Joana Roja]

(Matt is a terminally curious ex-NASA engineer and avid self-experimenter. His projects include developing the Think, Try, Learn philosophy, creating the Edison experimenter’s journal, and writing at his blog, The Experiment-Driven Life. Give him a holler at matt@matthewcornell.org)