QuantifiedSelf, Sleep, and a Questionable set of Tools

A while back, I went pretty far down the Quantified Self rabbit hole: I was measuring a pile of things, downloading apps to track things, wearing things that tracked other things, and generally creating all kinds of allegedly descriptive data. The fact that it wasn’t prescriptive was no surprise: the data doesn’t and can’t provide answers. Analysis of the data, however, was the big promise.

One of the things I was most interested in was sleep tracking: it’s unquestionably valuable in terms of physical and mental health, and I figured that if I could get better at it, I’d be better off. I started off with SleepCycle, a phone-based app. Then I added a Pebble smartwatch (or rather, Pebble added health tracking to the smartwatch I owned). And I even rather unexpectedly ended up with an IoT internet-connected sensor-filled superbed from SleepNumber called the X10.

Recently, I got curious again: I was still tracking all the sleep data, but wasn’t doing anything with it. I’d gotten bored, complacent, and so the data piled up without much to say for itself. Today, in a fit of rainy day pliability, I tediously copied data into an excel sheet. This was my first gripe: Pebble doesn’t allow csv exports (but then again, hey, they don’t exist anymore). SleepNumber doesn’t allow data exports (they once did, but “retired” the feature because reasons), and their customer service reps are pretty tight lipped on why I can’t access my own data, yet their TOS says they’re allowed to sell it. Only SleepCycle lets you export and study your own data, so thumbs up to them.

SleepCycle and SleepNumber both track total sleep time, and assess a 1–100 score for the night, based on “who-knows-what-exactly”. Pebble only assesses total sleep time, no score. So I built a couple graphs using the past few months of data. The results were dreadful.

Here’s the graph comparing score numbers of SleepCycle and SleepNumber

r-squared is 0.000220147

Yup, those are pretty much randomly correlated.

Then I made a graph comparing the total time-in-bed that each device though was correct. Again, an eye-rolling mess.

so I slept for somewhere between 3 and 10 hours, you know, pretty much in that range somewhere…

So, in conclusion, I’ve got three ostensibly valid ways of tracking sleep data, and their ‘scores’ and even ability to track total sleep time are nothing better than statistically random. Quantified self is nothing if you can’t actually quantify.

So now what? I have no idea. Is there anything that can be rescued out of this? Is there any way to decide if one of these is better than the other? Leave your thoughts in the comments.

For those interested, here’s the data https://docs.google.com/spreadsheets/d/15VbeHVWrPX3yh4gQD_OGTdQXdsT0ehUtRWZgHuqBXx0/edit?usp=sharing

One clap, two clap, three clap, forty?

By clapping more or less, you can signal to us which stories really stand out.