The benefits of remoteness – digital mobility data, regional road infrastructure, and COVID-19 infections

1 Introduction

COVID-19 is a pandemic of immense dimension, bringing social and economic disruptions worldwide. An array of research papers emerged in the economics literature within a relatively short amount of time. This literature is mainly concerned with the impact that COVID-19 has on different sectors, regions, and countries in terms of economic and societal implications.

With our study we go another way. We address regional factors that are related to the differentiated spread of the disease in Germany. Specifically, we focus on the regional accessibility of NUTS III regions (Kreise and kreisfreie Staedte) through road infrastructure measuring the average travel time on roads towards the next major urban center (Oberzentrum). Since most regions do not contain a major urban center, the metric assesses both the quality of the network of regional roads as well as the connectivity with the next population hub. We argue that the accessibility measure is an interesting explanatory variable for the spread of COVID-19 as an instrument for mobility as well as in its own right (in reduced form).

In reduced-form regressions we show that there is a “benefit of remoteness”: If roads are bad or absent it takes more time to visit distant relatives and friends, a feature, which naturally leads to more social distancing already without the implementation of lockdown policies. Inferior road infrastructure, which has been shown to be an impediment to regional development in “normal” times (Krenz, 2019a; 2019b), thus turns out to be beneficial in times of a pandemic because it prevents or reduces social interaction, in particular with people from other areas.

Mobility has been regarded as a key variable in the spread of the COVID-19 pandemic. Several German newspapers as well as the Robert Koch Institute, the leading German health institution in the fight against the disease, publish mobility data, gleaned from mobile phone users. The data are used to assess how well the population obeys social distancing policies. Mobility, however, is problematic as an explanatory variable for the spread of the disease because it is certainly endogenous and measured with error. Reverse causality could be an issue when people reduce mobility if the stock of infections and thus the probability to become infected increases. There are also myriads of channels conceivable that may cause omitted variable bias.

For road infrastructure, in contrast, we are convinced that it matters for the spread of a disease only because it alleviates mobility. In particular, it is not the mere presence of roads that allows the virus to travel from place to place but the fact that (potentially infected) people use roads to get in touch with other (potentially infected) people, an activity that we measure as mobility. In our regressions, due to data availability, we consider road infrastructure from 2018, which is certainly exogenous to the spread of COVID-19 but at the same time highly correlated with current infrastructure due to long times to build and low depreciation rates. The fact that road infrastructure changes only very slowly over time explains also our confinement on cross-sectional regressions.

The cross-sectional approach implies that we assess mobility on a particular point of time and here we take Easter Sunday. Easter Sunday, celebrated on April 12th in 2020, happened well after the closing of schools and the ban of big sports events on March 13 and after the shutdown of non-essential shops, hotels, restaurants, and other service providers and the ban of public gatherings of more than two people, on March 22nd in Germany. However, unlike in other European countries, the German federal government never ordered their citizen to stay at home. Some of the federal governments, in particular those with Southern borders (Bavaria, Saarland, Baden-Wuerttemberg, Rhineland-Palatinate) implemented stricter policies of social distancing with the most drastic measures, which came close to a curfew, in Bavaria. Nevertheless many politicians as well as scientists were afraid of compliance in particular around Easter were Germans are used to visit friends and relatives and to go on short-term holidays. Since also the weather stations predicted sunny and pleasant weather around 20 degrees for basically all German regions on Easter 2020, many politicians, among them chancellor Angela Merkel, were alarmed and admonished their citizens in several speeches on Holy Thursday to comply to the social distancing rules (e. g. Der Tagesspiegel 2020).

As mobility measure we source data provided by the Robert Koch Institute and Humboldt University Berlin (Mobility Monitor 2020a), which is gathered from mobile phone data. Figure 1, gleaned from the data in Mobility Monitor (2020b), shows the smoothed average change of mobility in Germany (compared to March 2019). We see that mobility declined by up to 40 percent and that most of the decline happened after enactment of the first set of policy measures on March 13 and before enactment of the second (more drastic) set of policy measures on March 22. Since the end of March, mobility is mildly on the rise again. The fact that the greatest increase of new COVID-19 cases happened from early to mid March 2020, RKI (2020a) suggests that at least part of the mobility reduction is an endogenous response to increasing and/or high infection rates. Maloney and Taskin (2020) show that for the U.S. and other countries, mobility is strongly negatively associated with lagged COVID-19 cases, controlling for policy measures (non-pharmaceutical interventions) and interpret these results as a causal effect of infections on mobility.

Figure 1

Change in Mobility in Germany.

In order to assess the causal effect of mobility on infections, we exploit the fact that there is large variation in mobility reductions across the German regions. Large mobility reductions are found in regions with good road infrastructure because people stayed at home on Easter 2020 and did not travel and move as much as they did in April 2019 (e. g. in Bavaria). On the other hand, mobility reductions are low in regions with bad road infrastructure because travel behavior on Easter 2020 was not much different from traveling in 2019 due to remoteness and the constant low opportunities to leave home and travel smoothly and quickly on road networks (e. g. in East German regions). Especially, a Sunday in a remote area with low quality of road infrastructure is tough: busses might not go on Sundays or have much less frequency to go. In other words: the travel restrictions for Easter 2020 apparently did not change much in remote areas’ population mobility but changed greatly in highly accessible regions.

We use data from the Mobility Monitor (2020a) that provides for each of the 401 German NUTS III regions the number of trips per day, measured relative to the average number of daily trips at the same weekday in the same month of the previous year. Specifically, our data compares mobility on Easter Sunday 2020 with an average Sunday in April 2019. This means that we compare a non-working day (for most of the population) in April 2020 with an average non-working day in April 2019. We expect that, on non-working days, mobility is more of a choice variable than on workdays where it depends to a larger degree on inevitable commuter traffic. This is why we did not use Good Friday 2020 since the average Friday in April 2019 was to 75 % a working day. Compared to other Sundays in April 2020, Easter Sunday was actually less special than one might think. The average mobility reduction in Germany was 34 % on April 5th, 28 % on April 12th (Easter), 30 % on April 19th, and 24 % on April 26th, all values are computed using the same reference value, mobility on an average Sunday in April 2019. Sunday mobility is a more precise measure of voluntary social distancing than commuter traffic. Commuter traffic is externally determined and, if it changes in response to the pandemic, the change likely reflects other responses to the pandemic than voluntary social distancing. For example, if teachers stop commuting after the shutdown of schools, the response is likely to pick up also the impact of school closure on infections. Commuter traffic, however, is included as a potentially important confounder in our regressions of Sunday mobility change on infections.

Our identification strategy assumes that road infrastructure affects the regional spread of COVID-19 infections only through its impact on the mobility of people. We instrument mobility reduction with regional road infrastructure and estimate the impact of (instrumented) mobility reduction on infections.

We are aware of one other study so far, which addresses the regional variation of COVID infections in Germany. Mense and Michelsen (2020) show that the number of newly infected cases significantly depends on the population density per region, on commuter flows and on rain days. Our study differs by focussing on the regional distribution of the stock of infected cases per capita and by using regional road infrastructure as instrument for mobility in order to make causal inferences. Chiou and Tucker (2020) argue that access to high-speed broadband alleviates social distancing and work from home and show that people in U.S. regions with high-speed internet (and with high income) are more likely to comply to social-distancing directives. Kapoor et al. (2020) argue that people are less outgoing on rainy days and use (unexpected) rainfall in U.S. regions to assess the impact of social distancing on COVID-19 infections. Dehning et al. (2020) and Donsimoni et al. (2020) use epidemiological simulations to assess the impact of social distancing interventions in Germany and conclude that the official interventions were effective in curbing the spread of the disease (and, in case of Donsimoni et al.) necessary in stopping the growth of infections. Other economic studies on the COVID-19 pandemic in Germany based on RKI data include Berlemann and Haustein (2020), Weber (2020), Mitze et al. (2020), and Armbruster and Klotzbuecher (2020). Greenstone and Nigam (2020) use an epidemiological model to estimate the (huge) monetary benefit from social distancing that accrues through avoided deaths.

2 Data

The Robert Koch Institute provides data on COVID-19 infection cases and death tolls down to the level of NUTS III regions which are the 401 district-free cities and districts (Kreise und kreisfreie Staedte) of Germany (RKI 2020b). We extract data on the total number of cases per region and on the number of cases per 1000 inhabitants. The data are from May 5th, 2020, 23 days after Easter Sunday. This ensures that time passed by regarding the incubation time, doctor’s visits, testing and getting test results. Figure 2 displays in Panel A the total number of COVID-19 infections. It shows main hubs of infectious activity, like the region Heinsberg in the very West, several regions in the South-West, South, and in city areas, like Berlin, Munich, Hamburg, and Hannover. Panel B displays infections per capita (multiplied by a thousand). The map shows that many regions in East Germany have low numbers of infections per capita, while Bavaria and Baden-Wuertemberg in the South and South-West have the most infections per capita.

Figure 2

COVID-19 Infections – Regional Distribution in Germany.

Our main explanatory variable is the accessibility of regions by means of road infrastructure (Erreichbarkeit von Oberzentren) which is the average travel time by car (in minutes) from all communities of a NUTS III region to the next major urban center. These data are obtained from the German Federal Institute for Research on Building, Urban Affairs and Spatial Development (BBSR 2020) through its INKAR database (Indicators and Maps for Spatial and Urban Development). The accessibility calculations for motorized individual traffic are based on route searches in a road network model that take into account speed profiles of the elements of the road network, from fast federal highways to slow city streets, as well as structural and topographical conditions. Notice that, by construction, the accessibility indicator is lowest (namely zero = best accessibility) for cities that are classified as major urban centers and the indicator increases with remoteness of the region.^[1]

Accessibility indicators measure the level of service of an infrastructure network, regardless of utilization (Baradaran and Ramjerdi 2001). They are thought of as a measure of potential opportunities of interaction (Hansen 1959). Travel time is the simplest of accessibility measures. It is straightforward to compute and to understand but it provides no information on the quality of travel services provided and on behavioral aspects of their travelers and their heterogeneity. Accessibility measured by travel time is a central parameter in the spatial planning and assessment of transport infrastructure in Germany (BBSR 2018). This is understandable since actual travel decisions are likely more influenced by travel time than by spatial distance. It has been argued that travel time on roads measures actual accessibility relatively precisely compared to travel time by public transit. For public transit, waiting time, frequency of service, out-of-vehicle travel, and other factors additionally affect the ease at which travelers reach their destination and thus actual accessibility (Kim and Lee 2019).

Major urban centers (Oberzentren) are agglomerations of the highest level of centrality. They are classified by functionality. In contrast to smaller agglomeration centers (Mittel- and Unterzentren) they provide services and infrastructure that satisfy non-essential and non-periodic needs such as theaters, museums, universities, special clinics, special shopping centers, and administration centers. Usually, larger cities are classified as major urban center. Germany consists of 401 NUTS III regions and by definition of the BBSR of 85 major urban centers (Oberzentren). Figure 3 displays the accessibility of the German NUTS III regions. It becomes apparent that especially in the East German regions, travel times are higher. Travel times on roads are lower in the South, especially in Bavaria.

Figure 3

Accessibility by Road Infrastructure.

Figure 4

Change in Mobility.

As regional control variables we source from the INKAR data base information on outward job commuters in percent of the number of employees, the number of general practitioners (medical doctors) per 10,000 population, and regional GDP, area size, and population size. Therewith we implicitly control for GDP per capita and population density. We always use, for each explanatory variable, data from the latest available year, which, for example, is from the year 2018 for accessibility. Since these control variables are persistent characteristics of regions, values from the recent past are good proxies for the present. We corroborate this claim by computing the correlations between the latest available data and previous years’ data in Table A.4. These correlations are very high, between 92 and 99 percent, strongly suggesting that past values are reliable proxies of a region’s current performance. A description of the variables and sources can be found in Table A.1. Descriptive statistics are shown in Table A.2. Correlations between variables are shown in Tables A.3.A, A.3.B.

To measure mobility we use a novel dataset collected by the Robert Koch Institute and Humboldt University Berlin (Mobility Monitor 2020a). From the Mobility Monitor website we extracted data on mobility profiles of individuals from the German NUTS III regions for Easter Sunday 2020 across the 401 NUTS III regions. The Mobility Monitor website displays in an online monitor ‘how much more or less are people on the go’, that is the change in trip frequency as compared to the average trip frequency on the same weekday of the same month in the year 2019. The mobility data are constructed from mobile phone data and make use of the information on the number of trips within and between areas (NUTS III regions) but do not trace single individuals and their movements. This is important, as every institution in the process of the data generation and aggregation process guarantees full anonymity. According to Mobility Monitor (2020a), “a movement is registered by the mobile phone provider when an individual switches cell tower areas, and ends when the person becomes stationary again. The start- and end-tower can be the same.” The data they use comes from the German Telekom and from Telefonica and is provided by the firms Teralytics and Motionlogic.

Figure 4 displays the change in mobility for Easter Sunday 2020 as compared to an average Sunday in April 2019. As can be seen, the highest reductions in mobility as compared to the previous year (light green colours) took place in the South German regions, especially in Bavaria. The smallest reductions or even small increases in mobility took place in the East German regions.

In Table A.6 we collect the top 10 of the German regions regarding i) the highest number of total COVID-19 infections, ii) the highest numbers of infections per capita, iii) the largest decline of mobility on Easter 2020 compared to an average Sunday of April 2019, and iv) regions with high accessibility (few minutes of travel time). It becomes apparent that the regions with the highest numbers of infection cases per capita are located in the South, in Bavaria (BY) especially. Mobility reductions are also largest (more negative value) in various regions of Bavaria. In Table A.7 we present the bottom 10 regions according to the same criteria, but with lowest mobility reductions and the worst road accessibility. The ranking shows that the number of cases per capita is especially low for some East German regions, located in the federal states of Sachsen-Anhalt (SA), Brandenburg (BB), and Mecklenburg-Vorpommern (MV), as well as for regions in Northern Germany, located in Schleswig-Holstein (SH) and Niedersachsen (N). Likewise, the lowest values for mobility reductions and the worst accessibility by road infrastructure are also found in the East and the North of Germany.

3 Empirical analysis

4 Conclusion

In this paper, we analyzed the regional distribution of COVID-19 infections in Germany and the regional factors that explain its distribution. We made use of a novel, innovative digital technology dataset which traces mobility profiles of the inhabitants of the 401 German regions. Our analysis showed that COVID-19 infections are spread unevenly across regions. There exists a distinct divide between the East and North German regions – which show lower infection rates – and the West and South German regions with higher infection rates.

We showed that in times of a pandemic there exists a benefit of remoteness. Controlling for potential confounders, regions that are far away from major urban centers (by means of travel time on roads) display less infections. In order to make inferences on the causal effect of mobility reductions on infections, we use road infrastructure as an instrument for the change of mobility (on Easter Sunday 2020 compared to an average Sunday in April 2019). Our results show that not being very mobile is a benevolent factor for reducing COVID-19 infection rates. According to the IV regression results, 1 percent less mobility reduction, which means a one percent lower mobility level at the outbreak of the disease, explains a decline of infections by 3.4 percent. Reaching other people, urban centers or meeting points contributes to an increase in infection rates. Social distancing, here conceptualized as bad accessibility and mobility, appears to be a main factor to hold infection rates down.