One last post to wrap up Fitness Week. I realized I let myself get distracted by tax policy and whether I could save money by driving to Missouri (read: me being cheap) that I forgot my last post for Fitness week. That post was supposed to be on modeling fitness data, but first, how did I actually do during fitness week?
FITNESS WEEK STATS
Fitness week ended up being a pretty average week for me, in aggregate above average, as shown below:
A few notes:
- Each weekday was close to its average over time, each day slightly exceeding the Garmin set goal, per my earlier analysis on "targeting."
- Monday and Thursday were my two worst days.
- Saturday and Sunday were both high days, over 30,000 steps. Which, is a record, as I've never had two consecutive days over 30K steps
MODELING THE DATANow an update on modeling fitness data to predict activity levels. I've posted on this a few times before, here's my most recent post for review.
For you nerds here's my model. I don't have a ton of data (about 100 days) so I have to use a fairly limited methodology. I'm using a log-log regression model to predict daily steps using six variables (3 logged "priors" and 3 fixed effects). R-squared is .77, which, is ok, but accounting for variance is fairly easy, because a good amount of the variance is just weekends and travel days being vastly deviant from mean. Here's what the regression looks like, variable definitions in next section.
For you non-nerds here are the simple results:
Most Important Factors
- Weekend: On weekends, all else equal, I take approximately 30% more steps than on weekdays.
- Steps_Prior: Each 1% increase in steps yesterday leads to a 0.4% decrease in steps today. A priori, the more active I am yesterday, the more tired I am today.
- Steps_Three_Prior: Each 1% increase in the steps over the past three days leads to a 0.97% increase in steps today. Essentially: one of the best predictors of steps today, is how I've done recently.
- Hours_Three_Prior: Each 1% increase in sleep in the last three days leads to a 0.37% decrease in steps today. Likely related to cumulative fatigue theory.
- Alone: Days when I'm home alone I tend to take 13% more steps.
- Travel: Days when I have to travel for work or family, I take 11% fewer steps.
CONCLUSIONA couple of things here:
- Fitness week ended up being a fitness tracker success. I exceeded my weekly average, and set a record for consecutive days over 30K steps.
- My fitness model is coming together well. It's fairly predictive, and all the input variables make a lot of sense. It's nice to make ceteris paribus approximations of the impact of certain factors. (e.g. How much more will I move on the weekends? How much does having a big day today impact tomorrow? etc.) I will continue working these models, adding additional factors and potentially acquire other people's data (shameless solicitation?)