Analysing the effects of COVID-19 Governmental responses on mobility in major European countries

The severity of restrictions imposed on people’s daily lives due to the COVID-19 pandemic has varied across different countries, partly depending on the extent of the outbreak but also the political choices of national governments.

In this blog, we examine how mobility patterns of people in major European countries have varied in response to the lockdown policies enforced by their governments.

Data Sources

To assess mobility levels, we used Google’s publicly available COVID-19 Community Mobility dataset (accessed 28th May 2020). To assess restrictions on mobility, we used data from the Blavatnik School of Government, Oxford University, compiled in their COVID-19 Government Response Tracker (OxCGRT).

The Google Mobility dataset measures the percentage change in mobility for different categories of places. Data is collected from the users who have enabled the ‘Location History’ feature in their phones. The changes are measured against a baseline value, which is the median mobility for that day of the week over a 5-week period from 3rd January to 6th February 2020. It is important to note that residential mobility measures the change in duration whereas, for all other categories, mobility shows the change in the total number of visitors. We have chosen parks, residences and workplaces for this analysis. A discussion of the nature of the Google mobility data and comparison with a similar dataset from Apple can be found in an earlier UBDC blog.

The OxCGRT measures the governmental responses to the pandemic in the form of 17 indicators corresponding to various policy measures. The indicators capture closure, containment, economic and health policies. This study uses the Stringency Index, which is a weighted average of a subset of these, relating to policy scores. This enables us to show restrictions on freedom of association not simply as a binary measure (lockdown or not) but on a more continuous or graduated scale. The Oxford index is scaled from zero to 100 (most stringent restrictions). To better visualize the relationship between mobility and stringency, we use the relative freedom of association, which is calculated as:

Freedom of Association Index = (100 – Stringency Index)

Grouping based on Mobility patterns

Other researchers have begun to make comparisons between policy restrictions and mobility levels using these two data sources, notably this paper by researchers at McGill University.

In this blog, we looked at the mobility changes for parks, residences and workplaces along with the Freedom of Association Index for each country. In each case, there is a dramatic change in mobility around the date when freedoms of association are most constrained (‘lockdown date’). The x-axis is divided into weeks for better understanding of the patterns.

In all countries, a few days saw an unusual decrease in workplace mobility and a corresponding increase in residential mobility. These days were found to be public holidays.

For each country, residential mobility increased after lockdown and followed a weekly pattern. The weekends witnessed lower changes than weekdays, which makes sense as people normally spend weekends at home. However, the changes in other categories were not similar in all countries. Three different patterns were noticed.

France, Italy and Spain

As lockdown came in, these countries showed a steep decline in mobility at workplaces but also at parks. They also had the lowest freedom of association after lockdown (highest stringency), which suggests that they took the strictest measures to counter the virus spread and that people complied with these measures. This is not surprising since these three countries have the most serious coronavirus spread in Europe in terms of numbers of confirmed cases and deaths.

In-depth analysis

• Italy started showing mobility changes towards the end of February. This was due to the spread of the virus in the provinces of Lodi and Padua. Subsequently, these areas were put under lockdown from 21st February, with schools and workplaces closed, and public gatherings banned. France and Spain also started implementing restrictions in early March, with a ban on gatherings and closure of schools.
• Italy was put under lockdown from 9th March, as shown by a sharp decline in freedom of association then. Spain followed suit on 14th March and France joined them three days later. People could go out only for essentials, to work (when working from home not possible) and for emergencies. As a result, workplace mobility on weekdays dropped by 60-70% after the lockdown and only slightly increased until the end of April.
• In these countries, parks also underwent a sharp fall in mobility, mainly because of the closure of public parks. Spain and Italy saw a decrease of 70-80% on weekdays, whereas for France it was around 60-70%. The weekends saw further decreases in mobility, pointing to the fact that more people used to visit parks during weekends.
• Lockdown measures have been eased since early May, with parks and workplaces allowed to open. This has resulted in a significant increase in workplace and parks’ mobility patterns, and a decrease in time spent at homes. It is notable, however, that the policy index for France and Spain shows only modest easing of restrictions in the period covered here but that mobility for both workplaces and parks has increased substantially.

Belgium and the UK

Judging by the Freedom of Association Index, these countries saw controls close to those of the first group (scores around 20-22/100 compared with 6-16/100). While these two countries witnessed a large fall in workplace mobility, however, they saw only a small decrease in that of parks.

In-depth analysis

• Belgium started taking preventive measures from 12th March by banning public gatherings and closing schools, cafes, etc. However, people were still allowed to go out for exercise. The UK government issued advice against non-essential travel and socializing on 16th March, and ordered the closure of schools, pubs and restaurants from 20th March, but also permitted continued daily exercise outside the home. These measures led to a significant decline in mobility at workplaces.
• Belgium went into total lockdown from 17th March, with all non-essential businesses ordered to close and citizens told not to leave homes unless necessary. The UK also imposed similar restrictions from 23rd March. It also passed the ‘Coronavirus Act 2020’, which gave the government emergency powers to handle the pandemic. As a result of these measures, workplace mobility decreased by more than 60%.
• Both the UK and Belgium did not close parks and allowed people to go out to exercise, as long as social distancing protocols were followed. The mobility for parks shows a lot of variation, ranging from a decrease of 60% to a slight increase over baseline on a few days. These could be either due to weather conditions or a lack of data, as discussed in one of our previous blogs.
• Starting 4th May, Belgium eased the lockdown by restarting economic and industrial activities and allowing people to visit relatives. As before, the scale of movement in the Freedom of Association Index appears very modest compared to the rise in activity in workplaces and parks. It is also evident that both kinds of activity were on the rise before any changes in restrictions.
• The UK also started easing the lockdown on 10th May by allowing unlimited exercise and return to work for people who could not work from home. This has resulted in a slight increase in workplace mobility and a rise in park visits.

Germany, Switzerland and the Netherlands

These countries saw similar levels of restrictions on freedom of association according to the policy index as Belgium and the UK, but smaller reductions in workplace mobility, and a general increase in park visits.

In-depth analysis

• Germany and Switzerland started preventive measures towards the end of February, with the German city of Heinsberg closing schools, libraries and public events, and the Swiss Federal Council cancelling all public events and gatherings of more than 1,000 people. The Netherlands imposed restrictions in the province of North Brabant from 9th March, cancelling events and advising people to work from home. This resulted in workplace mobility starting to go down.
• Germany went into lockdown in stages, with the states of Bavaria and Saarland the first to restrict people’s movements from 20th March. The measures were extended to the whole country on 22nd March, with social distancing mandatory and gatherings of more than 2 people banned. In Switzerland, educational establishments were closed from 13th March, and bar, pubs and non-essential shops closed on 16th March. No lockdown was enforced, although gatherings of more than 5 people were prohibited. The Netherlands also imposed strict social distancing protocols from 23rd March. We could see a clear correlation between the freedom of association index and workplace mobility for these countries. However, unlike the other countries, the drop in workplace mobility remained less than 50% (except on public holidays) and started decreasing gradually.
• The major point of difference for these countries was mobility in parks. As these countries did not restrict access to parks, there was a general increase in the number of visitors. It should be noted that the baseline values are from January when the weather is generally cold. Nevertheless, there was an upward trend after the lockdown, which suggests more people going to parks. In Switzerland and Germany, we observe a dip in mobility in late March and mid-May. On checking the weather for that period (in Zurich and Berlin), we find that cloudy weather and a fall in temperature were factors in some cases.
• These countries have also started easing restrictions since late April. This has been reflected in the mobility patterns, with workplace mobility going up and residential going down.

Summary

This initial exploration suggests three areas for further analysis. First, we see that workplace and residential mobility show a strong correlation with the Stringency/Freedom of Association Index, but this is less true with parks, where countries show large variations in responses. What these underline is that the overall Stringency Index is a composite of policy measures concerning several different areas of activity: work, education, leisure and so on. When understanding mobility responses, it would be useful to use the component parts of the index in future rather than the overall score. That would enable a better comparison of the effectiveness of policy measures.

Second, the relationship between the two sets of measures also shows a variation over time. In other words, there are several countries where the index suggests policy remained stable (and highly restrictive) but mobility levels crept up over time. There has been much speculation about how long governments can maintain compliance with lockdown measures. In the UK, this was one of the factors used to justify delaying the introduction of lockdown. It would be useful to model this compliance decay more formally.

Third, the data show substantial variations in mobility even under similar levels of restrictions. It would be interesting to compare the effects of these different governmental approaches on rates of infection. This could help understand whether limiting access to parks was a factor in flattening the curve, for example. There would, of course, be substantial challenges here due to both lag effects in the infection spreading, varying levels of community infection at the onset of lockdown and varying testing regimes.

There are a few issues that need to be discussed regarding the mobility data as a measure. Firstly, as discussed in our blog ‘Apples and pears? Comparing Google and Apple mobility data’, we are not sure of the underlying methodology used in getting this data, and whether any modelling techniques were used beforehand. Secondly, this method of data collection is not suitable for third-world countries, where ownership of location-enabled smartphones is probably significantly lower. It could produce biased results which may be entirely different from the real picture. Finally, as the data collection relies on the ‘Location History’ feature of the user being enabled, the data might not be representative of the actual mobility patterns. Thus, we should be wary of making policy decisions while using these methods without additional corroboration.

The code used for creating the above plots is available on GitHub.

Authors: Mohd Sarim (student, MSc in Urban Analytics), Dr Qunshan Zhao and Professor Nick Bailey.