Access and Discussion of Google mobility data feature

[payoto]

I was very interested to read your latest report using Mobility data from Google to arrive at a better and more reactive estimate of Rt, as we were exploring similar ideas and were in the process of correlating simulation results with changes of mobility to gain an initial idea of what to expect.

Are you planning to release the code which uses the mobility data for public use?

Questions/discussion on data periodicity

As we were also thinking of using this data we had a few thoughts about what different features of the data represented, notably the weekly periodicity. I'd be very interested in getting your thoughts and viewpoint on the following comments:

It is my impression that the strong weekly periodicity of the mobility data is due at least in part to the periodicity of the baseline data, which is a daily baseline. My hunch is that this periodicity does not accurately reflect behaviours and cannot be used to inform the Rt. The mobility data could probably be averaged over a 7 day window to get a more realistic estimate of transmission patterns.

If we consider the google mobility data for work place frequentation in Italy and France, there is a clear relative increase in workplace attendance (compared to other days in the week) on Saturdays and Sundays. In countries where most things are usually closed the interpretation of people sneaking to work on Sundays during a pandemic is a little far-fetched. It is much more likely that the periodicity is mostly in the baseline data, on a day where 90% of people are normally not at work, the essential workers still at work represent a larger proportion therefore appearing as an uptick in work place frequentation.

My suggestion is that the strong weekly periodicity does not reflect the real world transmission pattern as the google data are normalised per day rather than against a weekly average.

Extract from Google's About this data

Changes for each day are compared to a baseline value for that day of the week:

• The baseline is the median value, for the corresponding day of the week, during the 5-week period Jan 3–Feb 6, 2020.
• The datasets show trends over several months with the most recent data representing approximately 2-3 days ago—this is how long it takes to produce the datasets.

[s-mishra]

Hi @payoto, bit busy with few releases but the short answer for availability is yes code should be up by tomorrow noon (cleaning up uncecessary bits).

I will answer your other concerns in sometime as I get more time to write a detailed reply.

[payoto]

Thanks a lot for the hard work!

[s-mishra]

Hi @payoto , please find the release here de6f3e5

I will reply for other part tomorrow, just busy and do not want to not engage when I can't concentrate. Hope that is fine

[JeanClaudeR]

Thanks @s-mishra for such hard work and openness !

[payoto]

Hi @s-mishra , just a friendly reminder if you have time to address the points I made about the weekly temporality of the data.