Socioeconomic differences and persistent segregation of Italian territories during COVID-19 pandemic

Public health strategies extensively adopted by governments at the outbreak of the COVID-19 pandemic required the implementation of non-pharmaceutical interventions (NPIs) to reduce the transmission rate of the virus^1,2,3,4,5,6. Among NPIs, mobility restrictions proved to play a central role in decreasing social contacts^7,8, effectively contributing to control the diffusion of several pathogens across social systems and motivating their wide adoption for the containment also of the COVID-19^{9,10,11,12,13}.

Specifically, full national lockdowns, i.e. periods of severe mobility restrictions and social distancing for almost all citizens, were widely adopted during the first wave of the COVID-19 pandemic^14,15,16,17, while restrictions targeted to specific geographic areas and for shorter periods constituted the preferred strategy during the second half of 2020^{18,19,20,21,22,23,24,25}.

The spread of the COVID-19 was unevenly distributed among geographic areas and clustered in specific zones during the first wave of contagion^14,19,26. However, the adoption of generalized mobility restrictions across entire national territories severely affected the whole population, causing the disruption of national mobility networks^27,28,29 with relevant consequences for the socioeconomic context. The impact of the contraction of mobility flows on business activities and economic output thus sparked a vivid debate on the direct and indirect consequences of such policy measures^16,30,31,32, stimulating the investigation of the socioeconomic effects of NPIs adoption^26,33,34.

This work aims to assess the impact of NPIs by evaluating how mobility restrictions interacted with socioeconomic characteristics. In so doing, we focus on Italian territories and we rely on the assumption that a reduction in mobility flows represents a first-order proxy for the contraction of economic activities. This is plausible since a wide share of the population relies on mobility to reach workplaces, and a significant portion of productive establishments depends on direct contact with consumers to carry on their activities. For example, mobility restrictions have been shown to be strongly correlated with consumption reduction^35,36 and loss of Gross Domestic Production (GDP)³⁷ during the pandemic.

Official measurements on the economic response of individuals and firms are typically available only several months after the events take place, and often with insufficient geographic detail. Digital traces of individuals, including mobility data, are instead widely diffused, readily available and contain information on the geographic position of users^38,39,40. In this work, we take advantage of a large-scale social network dataset comprising near real-time observations for over 4 million individuals, provided by Facebook through its “Data For Good” program, to measure Italian mobility patterns and assess how restrictions disrupted the corresponding mobility network. Given the severity of the COVID-19 pandemic, we follow the literature on the management of natural disasters and other disruptive events^41,42,43,44 in which digital data have been exploited to evaluate the impact of shocks, even though these sources of information may not represent perfect survey instruments.

Extant literature on the negative impacts of NPIs on economic activities has focused mostly on the lockdown phase^26,34,45,46, although the speed and strength of recovery after restrictions lifting have relevant consequences on the overall impact of the crisis, with potentially long-lasting effects which could modify the socioeconomic structure of a country. On a macroeconomic perspective, the study of the effects of restrictions on economic output has resulted in competing recovery scenarios from the COVID-19 pandemic^{47,48,49,50,51,52}, distinguishing between a long and slow recovery (U-shaped or L-shaped), which signals the existence of long-run negative effects, and a short and fast dynamics (V-shaped), which indicates temporary disruptions.

Against this background, we investigate the existence of persistent effects from mobility restrictions in Italian territories by analyzing their recovery patterns and heterogeneous responses. In fact, some territories might limit the negative consequences of the lockdown, promptly reversing their mobility patterns once restrictions are lifted (a V-shaped recovery). Conversely, other territories might be persistently penalized by restrictions, suggesting long-lasting negative effects (a U-shaped recovery).

To investigate the existence and persistence of these heterogeneous patterns, we analyze the evolution of mobility in Italy from March to October 2020, i.e., covering both the lockdown and the recovery phases, and we construct an indicator measuring the level of mobility disruption. By ranking territories according to their mobility performance, we provide evidence on the existence of two distinct classes characterized by different behaviors. The most affected territories exhibit a slow U-shaped recovery, and at the end of our period of analysis their mobility levels are still nearly 20% lower than their pre-lockdown levels. The least affected territories have a quicker, V-shaped, recovery, with mobility reaching pre-lockdown levels around mid-July, hence only 2 months after mobility restrictions are lifted.

Then, to understand the extent to which pre-existing socioeconomic characteristics of Italian territories are informative to understand such mobility responses, we investigate the presence of commonalities among members of the two classes with both supervised and unsupervised machine learning techniques. We find significant evidence of persistent economic segregation induced by the lockdown, with most affected territories being characterized by lower income per capita and higher income inequality, in line with other recent studies^46,52,53,54. Such findings are confirmed with a quantile regression analysis to assess the individual effect of different socioeconomic dimensions.

Our main variable of interest is income, but we also account for other relevant socioeconomic variables, such as the distribution of revenues per employee by productive sector, the share of employed individuals capable to work from remote, or the North-South geographic divide. Thus, we provide evidence that the effects of mobility restrictions have been unequally distributed across territories, with stronger and persistent effects where income per capita is lower.

These findings are relevant for the decision process of policymakers. The existence of two separate behaviours of recovery among territories signals the urgency of a differentiated policy response, targeting the most affected territories with intervention reducing the persistence of the effects of mobility restrictions. Moreover, our results show the relevance of pre-existing economic and social conditions for the resilience of territories during disrupting events and support governments and policymakers in tailoring effective economic relief schemes. In fact, our analysis does not simply reveal how recovery behaves differently among territories, but it pinpoints which socio-economic characteristics are reliable indicators to assess heterogeneous recovery performances.