Inferring Mobility Reductions from COVID-19 Disease Spread along the Urban-Rural Gradient

TL;DR

This study addresses how mobility responses to COVID-19 vary across Germany’s urban–rural gradient by decomposing weekly out-of-home duration into a disease-responsive term and calendar-driven factors using anonymized mobile data across 400 districts. It employs a Bayesian hierarchical framework to capture multiplicative interactions: $D_{d}(t) = D_{ ext{base},d} \cdot C_{d}(t) \cdot W_{d}(t) \cdot V_{d}(t) \cdot H_{d}(t)$, where $C_{d}(t)$ reflects incident-driven risk perception with memory and fatigue. The analysis shows disease spread as the strongest driver of mobility reductions, with large cities reducing more than rural areas, and cross-district variation partly explained by population density and unemployment, while political factors gain prominence in the second wave. However, mobility reductions only partly translate into lower peak incidence, indicating additional risk and transmission factors. The work demonstrates that real-time mobility signals can proxy population response and inform locally tailored interventions along the urban–rural spectrum.

Abstract

The COVID-19 pandemic reshaped human mobility through policy interventions and voluntary behavioral changes. Mobility adaptions helped mitigate pandemic spread, however our knowledge which environmental, social, and demographic factors helped mobility reduction and pandemic mitigation is patchy. We introduce a Bayesian hierarchical model to quantify heterogeneity in mobility responses across time and space in Germany's 400 districts using anonymized mobile phone data. Decomposing mobility into a disease-responsive component and disease-independent factors (temperature, school vacations, public holidays) allows us to quantify the impact of each factor. We find significant differences in reaction to disease spread along the urban-rural gradient, with large cities reducing mobility most strongly. Employment sectors further help explain variance in reaction strength during the first wave, while political variables gain significance during the second wave. However, reduced mobility only partially translates to lower peak incidence, indicating the influence of other hidden factors. Our results identify key drivers of mobility reductions and demonstrate that mobility behavior can serve as an operational proxy for population response.

Figures (17)

• Figure 1: Model overview illustrating the composition of the time- and district-dependent out-of-home duration. The out-of-home duration is the product of a baseline out-of-home duration, a disease factor (based on incidence) and three disease-independent factors (temperature, school vacations, public holidays). Arrows indicate if we assume the factor to have an in- and/or decreasing effect on the out-of-home duration.
• Figure 2: Normalized local and national COVID-19 incidence are used to compute the disease factor in a four step process.1. Weighted average of normalized local and national incidence is computed. 2. Case numbers are convolved with a Gamma distribution. 3. Multiplication of convolved case numbers with exponential decay function to integrate pandemic fatigue. 4. Exponential of convoluted and exponentially decayed case numbers is taken.
• Figure 3: Using Bayesian inference we successfully infer the annual course of the out-of-home duration for the three exemplary districts Berlin, Göttingen, and Prignitz. The authors chose to depict their home districts of Berlin and Göttingen, together with Prignitz, the district with the lowest population density. In the middle and the right column, lines represent median factors and the ribbons represent 95% Bayesian credibility intervals. A-C. Normalized local and national incidences. All three exemplary districts experienced a first (spring 2020) and second (winter 2020/2021) COVID-19 wave. D-F. Multiplicative impact of the disease factor $C_{d}(t)$ and the three disease-independent factors $W_{d}(t), V_{d}(t)$, and $H_{d}(t)$. The disease factor $C_{d}(t)$ decreases the out-of-home duration most strongly, followed by smaller effects of $W_{d}(t)$, and only modest reductions due to $V_{d}(t)$ and $H_{d}(t)$. G-I. Observed vs inferred out-of-home duration. The model successfully infers the annual course of the out-of-home duration, specifically the shape of the two major decreases in spring 2020 and winter 2020/2021.
• Figure 4: A. Over time, the decreasing effect of the disease factor relaxes. Distribution of multiplicative impact across districts, using the median multiplicative impact for each district. Across districts, the disease factor emerges as the strongest influence, with pandemic fatigue evident in every district. Due to temperature fluctuations, out-of-home duration is approximately 10% higher in summer than in winter. School vacations and public holidays lead only to modest reductions. B. Gamma distribution delays the effect of the weighted normalized incidence. The weighted norm. incidence is convoluted with a gamma distribution to represent that the out-of-home duration in week $t$ is not only instantaneously influenced by disease spread in week $t$, but also by disease spread in the recent past. C. The disease factor emerges as the strongest influence on out-of-home duration. Distribution of effect sizes, in other words size of impact on out-of-home duration, across districts, using the median effect size for each district. While the effect size of the disease factor decreases from the first to the second wave, it still influences out-of-home duration most strongly. Secondary effects due to temperature, followed by only modest effects due to school vacation and public holiday. D. Local incidence carries only a minority of the weight in explaining mobility behavior. Distribution of weight of normalized local incidence in disease factor $C_{d}(t)$. Across districts, the majority of weight in explaining local mobility behavior is placed on national rather than local incidence, aligning with the introduction of nation-wide NPIs and communication strategies. Stars indicate t-test determining difference in mean across district types to be statistically significant, with significance levels of $^{\star \star \star}$ representing $p~<~0.001$, $^{\star \star}$ representing $p~<~0.01$, and ns representing $p>0.05$.
• Figure 5: Differentiation by district type and spatial depiction strongest reaction strength in cities and parts of the west of Germany. In either panel, incidence represents new weekly cases per 100,000 inhabitants. A. Large cities display the greatest average reaction strength, with reaction strength structurally decreasing along the urban-rural gradient. Stars indicate t-test determining difference in mean across district types to be statistically significant, with significance levels of $^{\star \star \star}$ representing $p~<~0.001$ and ns representing $p>0.05$. B. Depicting the reaction strength spatially highlights great reaction strength in cities (small polygons) and in the west of Germany, contrasting with modest reaction strengths in parts of the former GDR and northern Bavaria.