Exploring Google's COVID-19 Community Mobility Reports
Today, Google released a series of PDF reports that use their tracked location data to "chart movement trends over time by geography [and] across different categories of places".
Report contents
Reports can be downloaded for a specific country/region (currently 131 are supported). Within the report, time series charts display mobility trends (% compared to baseline) across 6 location categories. The date range available for the first round of reports are Feb 16 2020 - March 29 2020.
Most of the countries I looked at had a country-level overview at the beginning and then region specific charts. However, I noticed that South Korea only had the country-level overview. Within the report, it's noted that a region will be left out if there aren't statistically significant levels of data available.
Also, one country that does not have a report at all is China. I can understand why this is - my guess is that there is no data being collected at all in China (and will not have data in the future). In an alternate world though, it would be interesting to compare the trends in China to other countries.
Findings
I looked at a few countries that I was curious about. I also highly encourage you to explore the reports yourself and look at relevant countries to you.
People in the US and Canada reduced their visits to retail & recreation, transit stations, and workplace locations starting the week of March 8. For context, it was during this time that state of emergencies were declared and social distancing was encouraged.
On the other hand, grocery & pharmacy saw an increase in visits (panic shopping) the week of March 8 before decreasing in the middle of March.
While it's difficult to read exact dates on these charts, you can see slight timing differences between regions. For example, in the US, the District of Columbia only had a short peak in grocery & pharmacy visits.
Report contents
Reports can be downloaded for a specific country/region (currently 131 are supported). Within the report, time series charts display mobility trends (% compared to baseline) across 6 location categories. The date range available for the first round of reports are Feb 16 2020 - March 29 2020.
 |
| Screenshot of report contents. Image from Google |
Also, one country that does not have a report at all is China. I can understand why this is - my guess is that there is no data being collected at all in China (and will not have data in the future). In an alternate world though, it would be interesting to compare the trends in China to other countries.
Findings
I looked at a few countries that I was curious about. I also highly encourage you to explore the reports yourself and look at relevant countries to you.
People in the US and Canada reduced their visits to retail & recreation, transit stations, and workplace locations starting the week of March 8. For context, it was during this time that state of emergencies were declared and social distancing was encouraged.
On the other hand, grocery & pharmacy saw an increase in visits (panic shopping) the week of March 8 before decreasing in the middle of March.
 |
| Mobility changes across 6 location categories in Ontario, Canada |
While it's difficult to read exact dates on these charts, you can see slight timing differences between regions. For example, in the US, the District of Columbia only had a short peak in grocery & pharmacy visits.
 |
| Mobility changes to grocery & pharmacy stores in the District of Columbia, United States |
In Japan, one interesting trend is that workplace and residential traffic remained close to baseline apart from two days. These correspond with public holidays that occurred on February 24 2020 (天皇誕生日 / Emperor's birthday) and March 20 2020 (春分の日 / Spring Equinox Day).
Caveat
An important caveat is that the report says "Location accuracy and the understanding of categorized places varies from region to region, so we don’t recommend using this data to compare changes between countries, or between regions with different characteristics (e.g. rural versus urban areas). " I think this should be emphasized more clearly because I initially also wanted to compare the % across countries. For example, Italy has a much greater mobility drop than Japan or South Korea. Unfortunately, we can't draw any conclusions because we don't know the sample size and location accuracy of the data.
Similarly, we can't conclude that DC had less panic buying than other regions in the US. Perhaps location accuracy is worse in DC or they might've had a less representative sample. However, these reports are still useful in generating quick questions that can be investigated through other avenues. For example, you could try to back up the panic buying hypothesis using sales data from grocery stores.
Wishlist
Here's my wishlist of additional data that would be useful for analysis. Whether it's possible or easy to include is another matter (and I would be pleasantly surprised if Google released more information than what's currently provided).
 |
| Mobility trends for workplaces and residential areas in Japan |
Caveat
Similarly, we can't conclude that DC had less panic buying than other regions in the US. Perhaps location accuracy is worse in DC or they might've had a less representative sample. However, these reports are still useful in generating quick questions that can be investigated through other avenues. For example, you could try to back up the panic buying hypothesis using sales data from grocery stores.
Wishlist
Here's my wishlist of additional data that would be useful for analysis. Whether it's possible or easy to include is another matter (and I would be pleasantly surprised if Google released more information than what's currently provided).
- sample sizes (even rough, rounded numbers would be helpful)
- split between Android / Apple devices - I'm guessing that the majority of data comes from Android devices but it would be helpful in better understanding the sample sets
- access to anonymized data so we could build our own visualizations
- indication of location accuracy for a country
Comments
Post a Comment