The chart is subject to noticeable fluctuations - on weekends the readings are higher, and on weekdays - noticeably lower, now the difference exceeds 4 percentage points.
I wondered what would happen if we removed this noise and whether it would be possible to see something interesting if we clear the data from weekly fluctuations.
I downloaded the file from Google and calculated the moving average. I threw out the results for February 29, how to align, did not come up with, and it seems that this does not affect anything.
Here is the result:
Here is hi-res .
Interesting observations:
- on the schedule for 2020, the moment of the beginning of mass quarantines is clearly visible - the third week of March;
- the first week of May is accompanied by a surge by a couple of percentage points, apparently, not only in Russia it is customary not to work at this time.
- the nature of the previous spike, which occurred in the third week of April in 2017, in the fourth week of March in 2016 and in 2018, and in the fourth week of April in 2019, is unclear. I think this is some kind of holiday associated with the lunar calendar, but what exactly - I don't know?
Orthodox Easter? Some kind of national holiday in India? I will be glad for ideas.
- the late November spike is likely related to Thanksgiving in the United States.
- after bursts at the end of August, there is usually a month and a half of stagnation or even a rollback, the further, the more noticeable. This effect disappears by mid-October. I believe this is due to the start of the school year, university campuses do not support IPv6 enough. Then other forces compensate for this fall.
- and of course the end of the year is the biggest spike.
Quarantines around the world continue, so we probably won't see the effect of the cancellation - the fall will be spread over months.
What other non-obvious things have you noticed?