8
correlation of date versus steps r = -0.517, p = 1.20 x 10
-107
), whereas step counts only slightly
dropped the same period for the adult (Figure 1B, negative correlation of date versus steps r = -
0.066, p = 0.005). On average, the adult walked significantly more steps than the teen over the
5 years (adult mean 7757.2, standard deviation 2850.7; teen mean 6568.3, standard deviation
3685.9, t-test p < 2.2 x 10
-16
). Interestingly, both individuals walked significantly less on
weekend days (Sundays and Saturdays) than on weekdays, but the difference was more
pronounced in the teen (Figure 1C and 1D, teen t-test p < 2.2 x 10
-16
, adult t-test p = 0.006).
[Figure 2 inserted here]
We further investigated the teen’s step counts. Using six years of school calendars, we
determined the specific dates of five yearly recurring holidays (one week mid-winter break in
February, one week spring break in March or April, ten week summer break from June through
August, Thanksgiving one week break in November, and two week winter break from between
December and January). Approximately 30% of the available step count measurements could
be classified as occurring during one of these holidays (teen: 472 of 1566 measurements; adult:
571 of 1823 measurements). The teen walked significantly more on school days compared to
non-school days (combining holiday and weekend days, Figure 2A, 2105.9 more steps on
average, t-test p = 8.17 x 10
-31
). Interestingly, the adult showed no significant difference in step
counts between school days and non-school days. The teen made fewest steps in February
and March compared to August through October (Figure 2B), and there was significant
variability in the steps across the months (Chi-square test p = 3.71 x 10
-67
). To test the effect of
seasonality, we fit a linear regression model on the teen’s step counts, with parameters
representing whether a day was a weekend, in one of the five holiday periods, the year of the
study (first through fifth), and the month of the year. All of these parameters were highly
significant in the fit regression model (Table 2).
. CC-BY-NC-ND 4.0 International licenseIt is made available under a
is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review)
The copyright holder for this preprint this version posted October 26, 2020. ; https://doi.org/10.1101/2020.10.21.20215491doi: medRxiv preprint