Measuring human mobility in times of trouble: an investigation of the mobility of European populations during COVID-19 using big data

Spatial mobility is a distinctive feature of human history (Bacci 2017, 2018). Many efforts have been made by scientists from different disciplines, from geography to physical statistics, to understand its mechanisms and thus generalise them in the form of a law. The pioneering contributions of Ravenstein (1885, 1889) have since been joined by many other attempts, most notably the contributions of Schläpfer et al. (2021), Alis et al. (2021). However, for a long time, human mobility was studied and approached by considering migration, a type of mobility that is less temporary than non-residential mobility and results in variations in the sizes and structures of the departure and arrival populations (Skeldon 2017). There are several reasons for this increased attention. First, there is a question of data availability; classic migration (typically labour migration) involves flows of people that are widely recorded by official statistics and for which there has always been relatively good data availability. Moreover, these types of migrations, which have characterised the industrial development of the Western world and in particular that of Europe, according to the rural–urban migration pattern, are less complex forms of mobility to measure and thus to interpret and model. Of all the models, it is enough to recall here the famous model of Harris and Todaro (1970), which formalised the relations between push and pull factors in labour migration, and the more general multi-exponential model of Rogers and Castro (1981), which laid the foundations for modelling, in a predictive sense, migration and age structures. In reality, however, as has been well known for many years now, the form of spatial mobility that most affects human activities is not the permanent form but rather that of temporary movements (Bottai and Barsotti 1994). This form of mobility has grown a great deal in recent years and is the form of mobility that, to date, has been observed the least and only partially. There are many reasons for this growth. As theorised by Zelinsky in his seminal contribution on the mobility transition half a century ago (Zelinsky 1971), this form of mobility (i.e. temporary, and increasingly fast and frequent) is characteristic of hyper-advanced modern societies. In these types of societies, in fact, people tend to stay close to each other in space, typically in urban and hyper-urban contexts—it is no coincidence that as of 2007, the world’s population is urban for the first time in history—where there are many interconnections and the infrastructures for both land and area mobility exist. This increase in mobility (in terms of both demand and supply) has been matched, at least until now, by a tendency to decrease the (monetary) costs associated with it. This general pattern has meant that in a relatively small number of years, i.e. during the period of the globalisation of economies, we have witnessed a process of the compacting and shrinking of space that has, in fact, globalised not only the exchange of capital and goods but also of people by drastically increasing spatial mobility on both macro- and micro-scales and changing the relative proportions of its forms (Tjaden 2021). Space has become so narrow and compact that it is no longer easy to understand what is near and what is far and thus to distinguish between long-haul and short-haul mobility. As to the reasons for the lack of observations, a predominant role has been played by data (statistical information), which until now, with respect to these forms of migration, were very few and only related to specific types of non-residential mobility, such as commuting movements recorded by population censuses. The situation has, however, changed rapidly, and this is also thanks to the technological revolution that has made it possible to obtain huge amounts of information on human mobility from new data sources and, specifically, from big data and digital data. This availability of data therefore makes it possible to measure and model these forms of mobility, which, being more elastic than traditional ones, are hyper-sensitive to sudden exogenous shocks such as extreme natural events and economic and health crises. The field of measuring these forms of mobility is quite new and therefore there are many challenges, including methodological ones, that need to be addressed. The implications, however, can be enormous because having reliable methodologies for measuring these forms of mobility makes it possible to prepare real-time-based policies that can make a difference in terms of spatial planning and measuring the effectiveness of certain political actions. Based on these premises and in line with some recent studies on mobility during the COVID-19 pandemic (Li et al. 2021; Lucchini et al. 2021; Hu et al. 2021), this contribution intends to provide empirical evidence on the measurement of the mobility of the European population during COVID-19. In particular, by exploiting Google mobility data, a Mobility Restriction Index (MRI) is formalised based on the different mobility restrictions imposed by different governments due to COVID-19, and it is also able to take into account the behaviours adopted by different citizens with respect to these restrictions. The index, which is sensitive to both the effects of COVID-19 on mobility and the effects exerted by the refugee flows caused by the Russian-Ukrainian War, is based on daily data and thus is theoretically sensitive to mobility restrictions. This is the main difference between this index and the better-known Stringency Index (SI) proposed by Oxford University (Ritchie et al. 2020). The MRI is calculated for Italy for the period from February 2020 to April 2022, as well as for the other 26 EU countries. It is then compared with the SI for both comparative and interpretative purposes. It is important to clearly explain that the aim of the index and therefore of the paper itself is to detect the reduction in human mobility caused by the restrictions adopted by governments in order to limit the spread of the COVID-19 virus. Thus, to a certain extent, we measured the resistance to movement. This implies that our index had low values when people did not comply with the constraints imposed by their governments and measured more accurately by the Oxford Stringency Index. The rest of the paper is structured as follows: Sect. 2 provides a review of the literature that focuses on the transition from migration to digital mobility. Section 3 presents the Google mobility data and introduces the MRI methodology, and Sect. 4 applies the suggested methodology in order to measure the mobility structure in Italy and provide a comparison of the COVID-19 outbreak responses imposed by each European government. Finally, Sect. 5 contains some comments and conclusions.

Historically, scholars have long attempted to find the causes and functional mechanism of migration. One of the earliest attempts that had a strong impact on the literature can be traced back to the work of the German geographer Ravenstein, who formulated the so-called ‘migration laws’ (Ravenstein 1876, 1885, 1889).

Ravenstein’s seminal work and his many insights into functional mechanisms that can explain and, to some extent, predict migration (like, for example, the gravitational approach) have significantly influenced theoretical approaches and interpretative models on internal and international migration. In this regard, one can consider the works of Lee (1966), Tobler (1995), Simini et al. (2012), Piché and Dutreuilh (2013) and Stillwell and Thomas (2016).

Based on Ravenstein’s migration laws, theories and interpretative approaches have proliferated over time (King 2012). Despite the strong expansion of new interpretative paradigms, including the idea of living in the so-called ‘age of migration’ (Castles et al. 2014), the study of migration still does not have its own migration theory. Rather, scholars basically agree that there is a fairly heterogeneous set of theories that are ‘valid’—that is, verifiable—for particular geographical contexts and at defined historical moments. In short, migration theory is very rich (for an exhaustive review, see Massey et al. (1993)), but at the same time it is weak (Arango 2017; Carling et al. 2020).

The interpretative and explanatory limits of migration theories depend, among other things, on the complex and mutable nature of the concept of ‘migration’ itself. Migration (in its traditional form) implies a non-temporary movement that, on the one hand, presupposes a migratory project of the migrant subject and, on the other, causes changes in the sizes and structures of the origin and destination populations. In fact, migration causes a decrease in the size of the population of the place of origin and, automatically, causes an increase in the size of the destination population. Since those who migrate usually have a young average-age structure, one of the effects of migration is also the transfer of potential new births and a rejuvenation in the age structure of the populations of the destination countries (Plane 1993; Rogers et al. 2002). However, ‘classical’ migration (often called labour migration) is only one form of mobility, and, moreover, it is progressively less relevant (in numerical terms) in the contemporary world, where spaces are highly compacted (Miller 2004) and the underlying causes of migration flows are no longer only economic (think, for example, of migration induced by climate change). Even with respect to mobility, however, interpretative schemes, theories, and of course modalities have changed a great deal in the recent past. According to classical theory, everyday mobility (typically labour mobility) was a form of systematic spatial interaction closely connected with migration and linked to the overall cost function (home and commuting costs) determined by the worker’s need to cover the distance between home and work given a certain wage Baccaïni (1997) and Bottai and Barsotti (2006). However, as can be guessed, many years ago this form of everyday mobility was only a part (albeit a significant one) of general mobility (which clearly included all, and even non-systematic, trips, such as those made for shopping, parental care, cultural consumption, etc.). The study of these forms of mobility, or the use of space, was of course hypothesised but almost never carried out, given that the only source of data for measuring the daily mobility was the population census, which, as is well known, historically surveyed mobility for work purposes and only more recently has been used to collect data for study purposes. Apart from scattered cases of ad hoc sample research (Bottai and Benassi 2016), all other forms of mobility were undetected, even though they existed and affected the lives of individuals and the definition of living spaces.

The distinction between migration and (non-definitive) mobility has been increasing over time; this is also an effect of the social transformations induced by technological diffusion and the compaction of space. Migration obviously remains a form of mobility, but there are wide and growing differences between the two that do not just concern the temporariness of spatial movement. Another fundamental difference is the speed at which the movement is realised. If it is true that migration, compared to other demographic events such as births and deaths, can be considered a ‘fast’ event Billari (2022), it is equally true that compared to mobility, it is still a slow process. We refer in particular to sudden (i.e. non-systematic) and very random forms of mobility such as, for example, those induced by sudden exogenous shocks (war, environmental disasters, pandemic crises, etc.). However, it is also true that the areas of overlap between the different forms of mobility are wide, and to some extent, the distinctions between one form of mobility and another are often very blurred. This is also because contemporary migration does not generally imply a lasting project and a definitive change; one thinks, for instance, of all those forms of migration linked to de facto temporary movements for work, education, adventure, etc.

From this perspective, the possibilities offered by the digital world and the nascent science of digital and computational demography are manifold, and in recent years, there have been many attempts to quantify human mobility (and related aspects) using digital data (from Google, Facebook, LinkedIn, etc.). Particular attention has been paid to the quantification of migrant stocks and, in recent times, to the quantification of mobility as a response to sudden and impactful exogenous shocks, especially because of the considerable implications these processes have for the preparation of intervention plans and policy programming.

The first strand includes the works of Napierała et al. (2022) and Rowe (2021) on using the Facebook platform to monitor migrant stocks, the contribution of Rampazzo et al. (2021) on defining a framework for estimating migrant stocks by combining digital footprints and survey data, and also the seminal contribution of

Alexander et al. (2022) on the combined use of social media and survey data to nowcast migrant stocks in the United States. Regarding mobility due to exogenous shocks and its impacts at the territorial level, it is worth mentioning the work of Alexander et al. (2019) on using Facebook data to determine the impact of Hurricane Maria on emigration from Puerto Rico. More recent contributions analysed the effects of COVID-19 on the temporary population of Rome using Facebook data (Brollo and Celata 2022) and the link between mobility and mortality in England and Wales using Google mobility data (Basellini et al. 2021). Finally, Rowe et al. (2023) analysed human mobility in Britain during the COVID-19 pandemic using meta-Facebook data.

All these contributions have shown that, even with some limitations related to the nature of the data sources used, indicators and measures built on digital data can be used as an early warning system for changes in migration and mobility (Barker and Bijak 2022; Napierała et al. 2022).

Our contribution fits into this strand of studies; we investigate a particular aspect of mobility or, rather, resistance to mobility. On the basis of data obtained from Google and referring to the daily mobility of individuals in selected European countries, we quantified the degree to which individuals adhered to the restrictions on mobility imposed by different governments to curb the spread of COVID-19. The idea is to understand, using an ad hoc constructed index and on the basis of the Google data, where and to what extent the mobility of individuals adapted to the restrictions imposed and, thus, to evaluate, in essence, the effectiveness of different governments’ policies in restricting the mobility of individuals. This is an entirely original theme compared to the most recent studies that used digital sources to study migration and mobility processes.

The importance of using non-traditional sources (and approaches) to study human population changes due to mortality, fertility, migration, and mobility has been recently underlined by a book published by the European Union on ‘Data innovation in demography, migration and human mobility’ (Bosco et al. 2022). New opportunities arising from innovative data sources also pose new challenges related to new measurement tools and analysis methods. It is these challenges that must be met in order to be able to provide governments with timely and accurate data on which to base territorially targeted and thus more effective policies. In this general context, mobility studies play a key role. Indeed, mobility is a phenomenon pertaining to what is known as fast demography (Billari 2022). In particular, as we have been able to verify empirically with this study, temporary mobility, i.e. mobility that does not necessarily result in a change of residence, is extremely sensitive to exogenous factors such as, for example, pandemic crises and the consequent mobility intervention policies. The results showed that the MRI precisely follows the path of the mobility restrictions imposed by governments. However, this trend has not remained constant over time; after the first wave, people probably started to consider the risk of being infected as a part of their ‘normal’ life, so they started to comply less with rules and voluntary restrictions on mobility and social interactions. This first result is, in our view, quite interesting. In fact, it shows that mobility in space is an intrinsic and structural characteristic of human nature, and therefore it can only be partially restricted for short periods. If we transpose this exercise to international mobility flows (between nations), we can clearly see how the flow-containment policies implemented by the destination countries are doomed to fail in the medium term. The comparison between the two indexes (the MRI and SI) shows that the SI is greater than the MRI on average. The highest difference between the two indicators is that recorded by Italy, a paradigmatic case in the study of the COVID pandemic. It is not the case, in fact, that the highest level of restriction out of the full sample of countries was adopted by Germany, Italy, France, and Greece. However, overall, the results show different patterns between countries without a clear model. It seems therefore that, although to a lesser extent, the mobility patterns during COVID seem to follow—after the first phase of restrictions—a sort of random process. In this context it seems that, after the end of the first wave in May 2020, the SI overestimates the level of restriction for Italy, extending the regional level of restriction to the whole national territory. We can therefore infer that the MRI works quite well for measuring and detecting the changes in the pattern of daily mobility. This aspect allows us to formulate a set of considerations regarding the usefulness of the MRI instrument and the potential of what has been observed in prospective terms. Having an index capable of capturing changes in mobility patterns and levels due to exogenous shocks can in fact lay the groundwork for simulating future mobility scenarios in which exogenous shocks may differ in terms of both the nature of the event and its intensity. What would happen, in fact, if governments decided to set policies to curb mobility due to the energy crisis or to contain ecological impacts? These results, suitably developed and modelled, can serve as a basis, and the MRI can serve as a cognitive tool. Of course, we are reasoning in the short term because in the medium and long term, we have seen how the individual’s propensity for mobility seems to take over, and this is in line with human history. The mobility of the human population has always contracted in times of war, disease, and economic crisis and then resumed vigorously. However, it is precisely in the short term that certain types of policies have to be adopted; think, for example, of the effects on mobility of exogenous shocks associated with extreme natural events. Finally, what we call fast demography, i.e. mobility, is not entirely independent of slow demography. In fact, temporary mobility and migration are closely linked and interconnected phenomena (Miller 2004). However, these phenomena can have important effects on the population and its overall dynamics by altering its propensities and behaviour significantly. Nor should the economic dimension of mobility be underestimated. In fact, this type of approach can also be extended to the mobility of goods and capital and, arguably, to trade, which is studied through approaches that are also used in the study of human mobility such as, for example, gravitational models. Moreover, we have to underline that changes in mobility patterns could imply significant changes in the structure of consumer behaviours (many people moved to the online market). The idea, therefore, is also to exploit this approach to predict the volume and direction of future trade in goods.