Introduction to Mobility and Covid-19
This article sets out to analyse Mobility and Covid-19 case numbers, or in more detail, the effect of changes in mobility between different sub-regions of the UK on these sub-regions’ Covid-19 case loads.
The Coronavirus pandemic has led to many countries introducing lock-downs and restrictions on movements of citizens to reduce interactions between people, so that the SARS-COV-2 virus can not spread. In the United Kingdom this lock-down started on March 23, 2020. This led to a decline in mobility in many regions. Measuring movements on a large scale like this is impossible for single individuals or small companies.
Luckily on the 11th of April 2020 Google released mobility data for the first time for many countries in the World. This data will be used to look at how changes in mobility affected Coronavirus|Covid-19|SARS–CoV-2 case numbers. The data was split into the constituting countries of the United Kingdom of Great Britain and Northern Ireland: England, Northern Ireland, Scotland and Wales.
While Google has released new data on the 23. April 2020, we will focus on the 11. April 2020 first. The incubation time for Covid-19 is roughly 1 to 14 days with most cases at 5 days. Google Mobility data is as such linked to cases 14 days later when most cases should have developed through differences in mobility within sub-regions.
So we will look if Region 1 has for example a 14% reduction in mobility and Region 2 has a 30% reduction, if there is a difference in recorded coronavirus cases between these regions.
Google’s reports look like this and shows how mobility changed.

Mobility Categories
Mobility is separated into six distinct classes: Retail and Recreation, Grocery and Pharmacy, Parks, Workplaces, Transit stations and Residential Areas.
Retail & Recreation are places like restaurants, cafes, shopping centres, theme parks, museums, libraries, and movie theatres.
Grocery & Pharmacy are places like grocery markets, food warehouses, farmers markets, speciality food shops, drug stores,
and pharmacies.
Parks are defined as national parks, public beaches, marinas, dog parks, plazas, and public gardens.
Transit stations are public transport hubs such as subway (the Underground or Metro) , bus/coach, and train stations.
Workplaces are places were people work.
Residential are places were people live or reside.
Case data for Mobility and Covid-19
Finding case numbers for some of the sub-region chosen by Google is easy as they match local and unitary authorities, for others it is a bit harder, especially as England, Wales, Northern Ireland and Scotland have different reporting schemes. For England data was obtained on the 25th of April 2020 from the Guardian newspaper that lists most regions Google used as their sub units. For Scotland, deaths are easy to find but cases in the local authorities are harder to get. Here the data was taken from this ARCGIS dashboard. Public Health Wales Data was obtained from their Tableau account. Northern Ireland obfuscates its data the most, but case numbers were found here.
Population data
As bigger places, like cities, have obviously more cases, data for the number of people and the population density needed to be found for the Google assigned sub-regions to adjust the case numbers so that population size of a sub-region was not an influencing factor. This data was obtained from the local authorities’ Wikipedia pages.
The case numbers were divided through the population number.
Results for Mobility and Covid-19
Behavioural Timeline
For all of the UK the behavioural time line looks like this.

As one can see there is a weekly pattern to the countries behaviour.
Cross Correlation Function for the different sectors and New Cases for all of the United Kingdom
Retail and Recreation

Grocery and Pharmacy

Parks

Transit

Workplaces

Residential

Cross Correlation Function for the different sectors and New Deaths for all of the United Kingdom
Retail and Recreation

Grocery and Pharmacy

Parks

Transit

Workplaces

Residential

Covid-19 lock-down Mobility differences: Retail and Recreation
From below graphs it appears that when a sub-region in England and Wales adhered more to the lock-down by visiting Retail and Recreation places they had less cases for the size of their region than places that behaved less well.

Taking our clue from above plot where there are possible relationships a GLM over England and Wales had the following result.
Call:
glm(formula = sqrt(adjCases) ~ Retail, family = gaussian(link = "identity"),
data = subset(mobdf, UKSUB == "England" | UKSUB == "Wales"))
Deviance Residuals:
Min 1Q Median 3Q Max
-0.0245783 -0.0073401 -0.0002243 0.0045432 0.0309703
Coefficients:
Estimate Std. Error t value Pr(>|t|)
(Intercept) 0.1105281 0.0188074 5.877 4.99e-08 ***
Retail 0.0008984 0.0002338 3.842 0.000209 ***
---
Signif. codes: 0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1
(Dispersion parameter for gaussian family taken to be 9.489983e-05)
Null deviance: 0.0113656 on 106 degrees of freedom
Residual deviance: 0.0099645 on 105 degrees of freedom
AIC: -683.47
First the Covid-19 cases were transformed with the R sqrt function to reasonably fit a Gaussian distribution.

Plotting the Analytics (Normal QQ, Residuals vs Leverage, Scale-Location, Residuals vs Fitted) gives the following plots.

In the Retail and Recreation sector one can as such say that counties and unitary authorities in England and Wales that did not adhere as well to the Coronavirus lock-down imposed by the British government than others seem to show more Covid-19 case numbers.
Covid-19 lock-down Mobility differences: Groceries and Pharmacies
From below graphs it appears that when a sub-region in England did adhere more to the lock-down by visiting less Grocery and Pharmacy places that they had fewer Covid-19 cases.

The response variable will behave as in above histogram,
A GLM gives the following result:
Call:
glm(formula = sqrt(adjCases) ~ Grocery, family = gaussian(link = "identity"),
data = subset(mobdf, UKSUB == "England"))
Deviance Residuals:
Min 1Q Median 3Q Max
-0.0217061 -0.0049577 -0.0006011 0.0056409 0.0213646
Coefficients:
Estimate Std. Error t value Pr(>|t|)
(Intercept) 0.0555279 0.0069689 7.968 7.54e-12 ***
Grocery 0.0006080 0.0002139 2.842 0.00564 **
---
Signif. codes: 0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1
(Dispersion parameter for gaussian family taken to be 6.026356e-05)
Null deviance: 0.0054887 on 84 degrees of freedom
Residual deviance: 0.0050019 on 83 degrees of freedom
AIC: -580.73
Number of Fisher Scoring iterations: 2
One can as such say that counties and unitary authorities in England that did not adhere as well to the Coronavirus lock-down imposed by the British government by visiting Groceries and Pharmacies than others showed more Covid-19 case numbers 2 weeks later.
Covid-19 lock-down Mobility differences: Parks
From below graphs it appears that differences in going to Parks or not made not much difference in England, Scotland, Wales and Northern Ireland.

No GLM (General Linear Model) showed a significant influence between people that went to parks or not and the resulting case load in that region two weeks later.
Covid-19 lock-down Mobility differences: Transit
From below graphs it appears that differences in visiting Transit places made not much difference in England, Scotland and had an adverse effect in Wales and Northern Ireland. That means theoretically: If a sub-region in Northern Ireland had more people at transit places they had less Covid-19 cases two weeks later. But in Wales it looks as if there is a non-linear relationship.

A GLM just for Northern Ireland shows a slight effect on the 0.05% probability border but Northern Ireland is not subdivided into many sub-regions by Google, so one should possibly not make to much of this result, as a 0.05% probability means there is a 1 in 20 chance that this a random result.
Call:
glm(formula = sqrt(adjCases) ~ Transit, family = gaussian(link = "identity"),
data = subset(mobdf, UKSUB == "Northern Ireland"))
Deviance Residuals:
Min 1Q Median 3Q Max
-0.0059874 -0.0034193 -0.0001166 0.0032015 0.0066147
Coefficients:
Estimate Std. Error t value Pr(>|t|)
(Intercept) 0.0088540 0.0080323 1.102 0.2989
Transit -0.0004163 0.0001371 -3.036 0.0141 *
---
Signif. codes: 0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1
(Dispersion parameter for gaussian family taken to be 2.275667e-05)
Null deviance: 0.00041462 on 10 degrees of freedom
Residual deviance: 0.00020481 on 9 degrees of freedom
AIC: -82.588
Number of Fisher Scoring iterations: 2
The Residuals vs Leverage Plot shows though that “Belfast” and “Antrim and Newtonabbey” have an over strong influence on this model so this is another reason not to put too much weight on this result.

Investigating the non linear relationship between transit movement differences and Covid-19 will not be discussed in this article.
Covid-19 lock-down Mobility differences: Workplaces
There is no evidence that regional behavioural differences in going to Workplaces in the United Kingdom had an impact on Coronavirus cases two weeks later.

Covid-19 lock-down Mobility differences: Residential
There is no evidence that regional behavioural mobility differences in Residential areas in the United Kingdom had an impact on Coronavirus cases two weeks later.

Covid-19 lock-down Mobility differences: Interaction between Retail & Recreation and Grocery and Pharmacy
As the sectors Retail & Recreation and Grocery and Pharmacy were the only places were a difference in behaviour showed an influence on coronavirus cases 2 weeks later, the interaction between these Google mobility sectors was investigated. Visually this is presented in the following graphic. As we are multiplying negative values the resulting value is positive. A high positive value means less mobility.

One can detect with a GLM a significant difference in this interaction in the sub-regions of England.
Call:
glm(formula = sqrt(adjCases) ~ Grocery:Retail, family = gaussian(link = "identity"),
data = subset(mobdf, UKSUB == "England"))
Deviance Residuals:
Min 1Q Median 3Q Max
-0.0220356 -0.0053842 0.0002216 0.0055613 0.0211554
Coefficients:
Estimate Std. Error t value Pr(>|t|)
(Intercept) 5.320e-02 5.720e-03 9.301 1.63e-14 ***
Grocery:Retail -6.612e-06 2.158e-06 -3.064 0.00295 **
---
Signif. codes: 0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1
(Dispersion parameter for gaussian family taken to be 5.941015e-05)
Null deviance: 0.0054887 on 84 degrees of freedom
Residual deviance: 0.0049310 on 83 degrees of freedom
AIC: -581.94
The analytical plots show no major violations but Rutland, Cumbria and the Westmidlands are major influencing regions to above model.
There seems to be as such not only a simple effect in the behavioural differences of the Google mobility categories Retail and Recreation and Groceries and Pharmacies but also how they interact. Regions that didn’t adhere to the lock-down as well as other regions seem to show more Covid-19 cases two weeks later if there was an interaction between above mentioned categories.
Conclusion
The results are as such interesting as that behavioural differences in the Transit and Workplaces category leading up to the 11th of April 2020 did not show much effect on adjusted coronavirus cases two weeks later.
Relevant seem to be only Retail and recreational places and Groceries and Pharmacies visits. This could out side London (which Google together with Manchester just described as Greater London and Greater Manchester) in places without subways be down to the fact that many transit places would be outdoors and probably like Workplaces have a higher likelihood to always meet the same set of people. People would go to work at the same time and meet a similar set of people every day. In non customer facing roles people would also meet the same set of people in Workplaces.
But in Retail and recreational places and in Grocery and Pharmacy visits the probability to meet a different set of people would be higher, hence with meeting more people the probability of meeting someone infected would be higher.
There are also more transit places outside in the fresh air (like bus stations) than inside, outside major metropolises like London that rely on an indoor subway. Obviously on buses there is also a chance to infect others but if the bus is not overloaded there is a better chance of self distancing. The same applies to Parks that are outdoors.
Interesting is that Google detected over all of Britain -81% of difference in movement in Retail and Recreation places, -32% of difference in Groceries and Pharmacies, -37% in Parks , -57% in Workplaces. All these reductions resulted only in a +19% increase in Residential areas.
With a stay at home policy one would expect all the people that usually go to the other places to stay at home. Google does not seem to totally explain what their mobility means exactly. Are these people just moving (like dog-walking) in a residential area or are these people that simply reside aka stay there but do not physically move.
These results are of course no 100% proof. Google states that using this data to compare urban and rural settings might be fraught with difficulties. Especially in London GPS would be less precise than on open country roads. This data is of course also based on a subset of Google users that don’t care enough about their privacy to switch these settings off on their Android smartphones.
Supermarkets in the UK serve up to 25000 customers per week. That is 100000 per month. With the findings by Japanese Scientists
and this paper that in closed rooms infectious micro droplets can stay for a quite long time in the air it is believable that the results of this study have some validity. Above video is also just for droplets in the air. Droplets that fall onto surfaces can stay longer.
“Studies have shown that the COVID-19 virus can survive for up to 72 hours on plastic and stainless steel, less than 4 hours on copper and less than 24 hours on cardboard.”
from the WHO FAQ.
And if 100000 people pass through a place that is a higher risk of infection than in an open bus-shelter or a Workplace with not much people throughput.
The most important change to behaviour seems to be to avoid these shopping places.
Hence shopping is highly likely more dangerous than going to work.
Regions
The areas that are in Googles report are:
Aberdeen City |
Aberdeenshire |
Angus |
Antrim And Newtownabbey |
Ards And North Down |
Argyll and Bute |
Armagh City and Banbridge And Craigavon |
Bath and North East Somerset |
Bedford |
Belfast |
Blackburn with Darwen |
Blackpool |
Blaenau Gwent |
Borough of Halton |
Bracknell Forest |
Bridgend County |
Brighton and Hove |
Buckinghamshire |
Caerphilly County |
Cambridgeshire |
Cardiff |
Carmarthenshire |
Causeway Coast and Glens |
Central Bedfordshire |
Ceredigion |
Cheshire East |
Cheshire West and Chester |
City of Bristol |
Clackmannanshire |
Conwy |
Cornwall |
County Durham |
Cumbria |
Darlington |
Denbighshire |
Derby |
Derbyshire |
Derry And Strabane |
Devon |
Dorset |
Dumfries and Galloway |
Dundee |
East Ayrshire |
East Dunbartonshire |
East Lothian Council |
East Renfrewshire |
East Riding of Yorkshire |
East Sussex |
Edinburgh |
Essex |
Falkirk |
Fermanagh And Omagh |
Fife |
Flintshire |
Glasgow City |
Gloucestershire |
Greater London |
Greater Manchester |
Gwynedd |
Hampshire |
Hartlepool |
Hertfordshire |
Highland |
Inverclyde |
Isle of Anglesey |
Isle of Wight |
Kent |
Kingston upon Hull |
Lancashire |
Leicester |
Leicestershire |
Lincolnshire |
Lisburn and Castlereagh |
Luton |
Medway |
Merseyside |
Merthyr Tydfil County Borough |
Mid And East Antrim |
Mid Ulster |
Middlesbrough |
Midlothian |
Milton Keynes |
Monmouthshire |
Moray |
Na h-Eileanan an Iar |
Neath Port Talbot |
Newport |
Newry, Mourne And Down |
Norfolk |
North Ayrshire |
North East Lincolnshire |
North Lanarkshire |
North Lincolnshire |
North Somerset |
North Yorkshire |
Northamptonshire |
Northumberland |
Nottingham |
Nottinghamshire |
Orkney |
Oxfordshire |
Pembrokeshire |
Perth and Kinross |
Peterborough |
Plymouth |
Portsmouth |
Powys |
Reading |
Redcar and Cleveland |
Renfrewshire |
Rhondda Cynon Taff |
Rutland |
Scottish Borders |
Shetland Islands |
Shropshire |
Slough |
Somerset |
South Ayrshire |
South Gloucestershire |
South Lanarkshire |
South Yorkshire |
Southampton |
Southend-on-Sea |
Staffordshire |
Stirling |
Stockton-on-Tees |
Stoke-on-Trent |
Suffolk |
Surrey |
Swansea |
Swindon |
Thurrock |
Torbay |
Torfaen |
Tyne and Wear |
Vale of Glamorgan |
Warrington |
Warwickshire |
West Berkshire |
West Dunbartonshire |
West Lothian |
West Midlands |
West Sussex |
West Yorkshire |
Wiltshire |
Windsor and Maidenhead |
Wokingham |
Worcestershire |
Wrexham |
York |