General description: the figure shows my daily step counts going back to March 2015, grouped by day of the week. I used Gnuplot 5.0 and its kdensity function (with smoothing) to create the violins. Data were collected using the default health app on whatever phone I had at the time, beginning with a Galaxy S5 and continuing up to today on an iPhone 13.
The x-axis: I chose to group the counts by day of the week, since my job requires quite a bit of walking on weekdays. I shaded the weekends differently, and those are clearly my rest days, lol. I imagine that lots of people in more sedentary jobs would have a distribution just the opposite of this, with fewer steps M-F and many more on Sa-Su.
The y-axis: Miles seemed like something easier to visualize than pure step count, even though that has no effect on the distribution. One limitation is that I used a standard 2200 steps per mile, regardless of the type of steps taking (hiking, stairs, late for the bus, whatever). The data may be less reliable at the low end because of days where I stayed in the house all day, but didn’t carry my phone back and forth all the time.
Box plots: I had first done this with box plots, but a friend reminded me they can mask bimodal distributions so the violin was a better choice. Thanks to S.R. for that.
![[OC] My daily miles walked for the last 9.5 years](https://www.europesays.com/wp-content/uploads/2024/08/12c4syz5exgd1.png)
1 comment
General description: the figure shows my daily step counts going back to March 2015, grouped by day of the week. I used Gnuplot 5.0 and its kdensity function (with smoothing) to create the violins. Data were collected using the default health app on whatever phone I had at the time, beginning with a Galaxy S5 and continuing up to today on an iPhone 13.
The x-axis: I chose to group the counts by day of the week, since my job requires quite a bit of walking on weekdays. I shaded the weekends differently, and those are clearly my rest days, lol. I imagine that lots of people in more sedentary jobs would have a distribution just the opposite of this, with fewer steps M-F and many more on Sa-Su.
The y-axis: Miles seemed like something easier to visualize than pure step count, even though that has no effect on the distribution. One limitation is that I used a standard 2200 steps per mile, regardless of the type of steps taking (hiking, stairs, late for the bus, whatever). The data may be less reliable at the low end because of days where I stayed in the house all day, but didn’t carry my phone back and forth all the time.
Box plots: I had first done this with box plots, but a friend reminded me they can mask bimodal distributions so the violin was a better choice. Thanks to S.R. for that.
Thanks for letting me share!