How has COVID-19 changed private car use in European urban areas? An analysis of the effect of socio-economic characteristics and mobility habits

Abstract

The private car has been identified as the main winner among transport modes in urban areas during the COVID-19 pandemic. The fear of contagion when using public transport or the decrease in road congestion are likely to have induced changes in citizens’ travel habits with respect to cars. This work investigates the impact of the pandemic on individuals’ habits and preferences regarding their car ownership levels and car usage in the European urban context, with a special focus on the role played by individual socio-demographics and urban mobility patterns. For this purpose, a Path Analysis approach has been adopted to model car ownership and use before and after COVID-19. The main data source employed in this research is an EU-Wide Urban Mobility Survey that collects detailed information (individual and household socio-economic characteristics, built environment attributes and mobility habits) of 10,152 individuals from a total of 21 European urban areas of different sizes, geographical locations, and urban forms. The survey data has been complemented with city-level variables that account for differences across the cities that may explain changes in car-related behaviour. The results show that the pandemic has induced an increase in car use among socio-economic groups that are generally associated with low car-dependent behaviour, revealing that policy instruments that discourage the use of the private car in urban areas are needed to avoid reversing past trends in the reduction of urban transport emissions. High-income, well-educated teleworkers are observed to be the ones that have reduced their car use to a larger extent. On the contrary, low-income individuals are mostly maintaining similar levels of car mobility. Finally, frequent public transport users are more likely than occasional users to have substituted this mode by the private car.

1. Introduction

The COVID-19 pandemic has induced stark behavioural changes that heavily disrupted urban transport. Initially, the decline in mobility, the widespread application of teleworking regimes and the increased adoption of online shopping were seen as an opportunity to advance towards the decarbonization of urban transport and a more sustainable urban mobility (Zhang & Zhang, 2021). At the same time, the fear of contagion and the desire to keep social distancing have pushed travellers to individual modes and away from public transport (Almlöf et al., 2021). In fact, in 2021 urban congestion levels were observed to be back to normal, in many cases even surpassing those from 2019 (TomTom, 2021), thus suggesting that at the aggregate level the effect of the pandemic on private car usage has not lasted. Nevertheless, the assumption that household mobility habits have remained unchanged after a crisis of such dimension is probably unrealistic. Citizens faced an uncertain situation that forced them to rethink all aspects of daily life, with mobility choices being one of the most affected areas. The nature and extent of these changes in travel behaviour are not homogeneous across socio-economic groups or geographical locations (Dueñas et al., 2021). Factors such as the type of employment, the income level, the place of residence within the city or the sanitary measures deployed in each region, are key constrains for the possibilities of adaptation of households to the new situation.

In particular, the ownership of private cars - being perceived as the most secure mode of transport in a moment in which any personal contact entailed a health risk - has been crucial for household mobility during the pandemic (Moody et al., 2021). This increase in the perceived value of having access to a car may have led to changes in car ownership and frequency of use, as well as the explanatory variables influencing them. For instance, car ownership and use are usually higher among high-income households, but in the context of the pandemic, these groups have been observed to telework more often (Olde Kalter et al., 2021). By contrast, low-income and low-qualified employees, more dependent on public transport and with more difficulties to telework, have in some cases reduced their mobility and car use to a lesser extent during the pandemic (Beck & Hensher, 2020). An additional issue to be addressed in this context is the role of shared mobility (i.e, bike-sharing, car-sharing, scooter-sharing, etc.) as an instrument to mitigate this increase in car dependency. Before the pandemic, these services were often considered to be sustainable substitutes of private cars in urban areas (see e.g. Clewlow, 2016) and to help reducing motorized travel (Midgley, 2011). However, it is likely that the pandemic has also affected travellers' perceptions and habits regarding these mobility options, thus changing their interrelation with car use.

Practitioners and policymakers all over the world are particularly concerned about the impacts of COVID-19 on urban mobility, and how to mitigate them through the design of appropriate policy measures to move urban transportation towards higher sustainability standards. For instance, the European Union (EU), both at the national and supranational level, is making an unprecedented financial and political effort to ensure a recovery that fits the environmental and social sustainability principles established in the European Green Deal (European Commission, 2019). The successful and efficient deployment of the post-pandemic recovery plans is largely dependent on the capacity of institutions to support households in the transition to a more sustainable way of moving. This capability can only be achieved through a sound understanding of the current preferences and habits of citizens with respect to private cars, and the identification of the factors that push them to rely on this mode to fulfil daily life activities.

This work explores the extent to which socio-economic attributes and past travel habits (i.e., frequency of use of the private car, public transport, and shared mobility) of individuals and households have affected changes in private car use and ownership after the advent of COVID-19 in European cities. To achieve this goal, we analyse an EU-wide Travel Survey which covers 21 European urban areas of different sizes, geographical locations, and urban forms. The methodology used for the analysis is based on a Path Analysis that simultaneously encompasses four main variables: (1) household car ownership before COVID-19; (2) frequency of use of private car before COVID-19; (3) changes in household car ownership levels due to COVID-19; and (4) changes in frequency of private car use in the post-pandemic period.

This paper is organized as follows. After this first introductory section, a literature review is presented in Section 2, covering the main existing evidence on car ownership, car use and the effect of the COVID-19 pandemic in urban mobility. Section 3 describes and analyses the survey used in this research, as well as the city level data collected to characterize the multiple cities included in the study. Section 4 explains the modelling framework that has been adopted to achieve the proposed goals and provides further details on the variables and techniques employed for the modelling phase. In Section 5, the modelling results are presented, together with a critical comparison with the previous literature. The discussion of those results, jointly with the main lessons and policy recommendation that can be draw from them, are explained in Section 6. Finally, Section 7 compiles the conclusions that can be drawn from the analysis, discusses its limitations and suggests avenues for further research.

2. Literature review

Car ownership and car use are inevitable entangled, as the chosen frequency of car use is inevitably preceded and constrained by the number of cars the individual has access to. However, most of the existing literature focuses on one of those choices and, in the case of contributions modelling car use, they tend to consider car ownership solely as an explicative variable of travel behaviour (Buehler, 2011, He and Thøgersen, 2017, Nolan, 2010). A smaller group of research studies have built different modelling frameworks decoupling the explicative effects of car ownership and travel behaviour, by considering car ownership as endogenous to socio-economic and built environment characteristics while being exogenous to car use (see e.g., Ben-Akiva and Lerman, 1974, Ding et al., 2018, van Acker and Witlox, 2010). Both approaches seem to obtain similar results regarding the effect of car ownership on car use: households that own more cars rely on this mode more frequently and drive longer distances (Dieleman et al., 2016, van Acker and Witlox, 2010).

In the following, the factors that have been previously observed to influence car ownership and car use are revised (2.1, 2.2), together with an overview of the existing research addressing the effect of COVID-19 on car use in the urban context (Section 2.3).

2.1. Factors affecting car ownership

According to the existing literature, socio-demographic characteristics have a clear impact on car ownership levels. Income plays a key role in the number of cars per household, with wealthier households owning higher number of cars (Dargay, 2001, Nolan, 2010, Potoglou and Kanaroglou, 2008). The income level is related to education level and employment status, so that qualified workers receive higher salaries than individuals in less-qualified positions, and therefore tend to own more cars (Chu, 2002, Ding et al., 2018). Additionally, GDP per capita at the national or regional level has also been observed to affect car ownership levels (Dargay et al., 2007, Dargay and Gately, 1999). The age and number of members in the household also influence car ownership. Moreover, they are inevitably linked to life-cycle effects that are also related to car ownership changes. For instance, Dargay (2001) observed a clear rise in car ownership for young adults in the UK and a decline when the head of the household reaches the retirement age, and the children start leaving the household. This decrease in car ownership for older adults, as well as the positive relationship between the number of adults in the household and the number of cars, is corroborated in Cornut (2016) for the Paris region, and in Matas & Raymond (2008) for Spain. For the case of children, their presence in the household seems to increase the probability to own a car (Dieleman et al., 2016, Oakil et al., 2016), but many authors have concluded that the number of children does not have a significant effect on the number of cars (Cornut, 2016, Yang et al., 2021).

Furthermore, the built environment inevitably influences the number of cars owned per household. Land-mix (Jiang et al., 2017, Zegras, 2010), as well as population density and employment density, are found to reduce car ownership (Ding and Cao, 2019, van Acker and Witlox, 2010). The same trend is observed regarding household proximity to public transport (Zhang et al., 2017) or proximity to the city centre (Ding and Cao, 2019, Cao et al., 2019). As for the existing shared mobility services, the only one that seems to affect car ownership is car-sharing, although differing results have been obtained. For instance, Becker et al., 2018, Martin et al., 2010 have found significant reductions in car ownership among car-sharing users in Basel (Switzerland) and US cities respectively, while Haustein, 2021, Zhou et al., 2020 concluded a non-significant effect on the number of cars given the low levels of adoption generally observed for this service in Copenhagen (Denmark) and major Australian cities.

2.2. Factors affecting private vehicle use

In general, socio-economic factors affecting car ownership also influence car usage but not always in the same way. Income, employment, and level of education are positively correlated with more frequent and longer car trips (Jiang et al., 2017, Schwanen et al., 2004, Shen et al., 2016). By contrast, the net impact of teleworking on car use remains still unclear (Lopez Soler et al., 2021). Early studies on this topic identified a small but significant negative effect of telecommuting on the number of trips and vehicle-kilometres travelled by car (Choo et al., 2005, Koenig et al., 1996). More recent studies suggest that reductions in commuting trips are accompanied by a reallocation to other purposes, such as longer distances travelled during the weekend or more fragmented trips (see e.g., Kim et al, 2015 for the Washington area and de Abreu e Silva and Melo, 2017 for the UK). Regarding the effect of household size and structure, it has been observed that the number of children increases car demand (Dargay and Hanly, 2007, Kim and Ulfarsson, 2008), while the effect of the overall household size is not yet clear (Ding et al., 2018, Yang et al., 2021). The study of the relationship between age and car use has yielded two theories. Some studies found that car use peaks for middle-aged adults and starts declining afterwards (Shen et al., 2016, Yang et al., 2021), while other authors obtained a monotonous positive effect of this variable (Nolan, 2010, van Acker and Witlox, 2010).

As for the relation between built environment and car use, existing contributions have generally concluded similar effects than those of car ownership (Chen et al., 2008, de Abreu e Silvaet al., 2012). Nevertheless, some authors suggest that its impact on car use is mainly indirect and occurring through a mechanism of modified car ownership levels (Cao et al., 2007, Ding et al., 2018, van Acker and Witlox, 2010). Commuting distance is often found to have a positive effect on car use (de Abreu e Silvaet al., 2012, Ding et al., 2014, Ding et al., 2017). At the city level, high car accessibility (Jun 2008), urban sprawl and low land-mix are observed to increase car use (García-Palomares, 2010, Yang, 2008).

Finally, the effect of shared mobility services on car use seems to be rather limited. The effect of car-sharing on private car use has received little attention, but in general, existing results point out a moderate reduction on the use of private car among carsharing adopters (Clewlow, 2016, Kopp et al., 2015). The theoretical potential of bike sharing systems in reducing car use in urban areas is often highlighted (Midgley, 2011, Shaheen et al., 2010). However, the empirical contributions analysing the real impact on car reduction are scarce. The literature review carried out by Fishman et al. (2013) showed that shared bikes are more likely to substitute other sustainable modes than private cars. Later, the same authors found that the substitution rate of car trips by bike-sharing is largely dependent on the overall proportion of trips that are done by car in the city (Fishman et al., 2014). As for scooter-sharing, the authors have not been able to find any existing research addressing the interrelations of these services with private car use.

2.3. The impact of the COVID-19 pandemic on urban mobility

As soon as the global health emergency was declared, and the enforcement of lockdowns and mobility restrictions spread throughout the world, a massive effort was made by the transport research community to gather evidence and provide insights on how this crisis was affecting transport behaviour. Existing research works have shown that the effect of lockdowns and restriction were largely dependent on socio-economic characteristics (Almlöf et al., 2021, Fatmi, 2020) and that public transport trips dropped significantly less than car trips (Aloi et al., 2020, Eisenmann et al., 2021, Labonté-Lemoyne et al., 2020). Interestingly, for the case of Bogotá (Colombia), Dueñas et al. (2021) observed that high income residents, who were observed to move more before the pandemic, reduced their mobility to a much greater extent than lower income groups. Schaefer et al. (2021) studied the effect of COVID-19 in mobility in the Hanover Region (Germany) in June 2020 and observed that residents in the city centre were more likely to increase their car use than individuals living in less dense areas. Recently, Basu & Ferreira (2021) carried out a survey among residents of the metropolitan area of Boston (US) between April and October 2020, finding out that 20 % of zero-car household included in the sample intended to buy a car because of COVID-19.

The characterization of mobility trends right after the first COVID-19 outbreak reveals diverging consequences of the pandemic across sociodemographic groups. In this respect, individuals traveling less or relying in sustainable mobility options before the pandemic, have been pushed towards an increase in car dependency or to higher levels of mobility than other groups. However, the extent to which this tendency disruption will have a lasting effect on urban mobility in coming years remains an open question. The number of research works addressing these long terms impacts is still scarce. Christidis et al. (2021) combined pre-pandemic data and the TRIMODE model to predict the post-pandemic recovery of transport activity in EU Member States. Currie et al. (2021) carried out a survey in Melbourne (Australia) during the summer of 2020 on expected travel behaviour once the virus is no longer present and found that car use is expected to grow with respect to pre-pandemic times even with the mobility reduction associated to teleworking. Similar research in terms of survey timing and expectations analysis was performed by Angell & Potoglou (2022) in the Cardiff Region (UK), also finding evidence that suggests a car-dependent mobility recovery. The review of these works shows that the current knowledge of the long-term effect of the pandemic on urban mobility has been mainly based on expected behaviour of the individuals when the pandemic had just started, and most European countries were still under strict lockdowns or applied restrictions to mobility.

The above literature review points out some research gaps to be addressed. First, there is a need to explore individual urban travel behaviour when lockdowns where already over, but some sanitary measures were still in place. Secondly, since the private car has been signalled as the main winner among urban transport modes during the pandemic, a specialized study on how the COVID-19 pandemic has changed the determinants of household car ownership and usage is especially relevant to understand its consequences for urban mobility in coming years. Third, the analysis of the relationship between changes in car use after the end of lockdowns and previous frequency of use of public transport and shared mobility will provide relevant insights to better understand how the car is interacting with more sustainable modes after the start of the pandemic. Last, this paper represents an innovative contribution that exploits an EU-wide survey, with more than 10,000 observations from a total of 21 European cities, that simultaneously accounts for both variability across individuals and urban areas of different profiles. This confers this research a high level of generality and a remarkable value for the understanding and regulating of pandemic urban mobility in the European Union after the start of the COVID-19 pandemic.

3. Data – An EU-Wide urban mobility survey

This section presents the data employed for this research. The main source is an EU-wide travel survey conducted in 21 European cities, which is described in Section 3.1 and explored preliminarily in Section 3.2. This information has been complemented with data at the city level, that captures the economic, demographic, and urbanistic attributes of the urban areas included in the study, which is reported in Section 3.3.

3.1. Survey description

The main data source exploited in this research is a cross-sectional EU-wide survey carried out in 2021 by the European Commission in 21 Functional Urban Areas1 (FUAs, from now on) throughout the EU (Christidis et al., 2022). The FUAs surveyed are represented in Fig. 1 and include: Paris, Lille and Calais (France), Charleroi (Belgium), Madrid and Málaga (Spain), Lisbon and Porto (Portugal), Milano and Catania (Italy), Berlin and Dresden (Germany), Stockholm and Malmö (Sweden), Bacau and Cluj-Napoca (Romania), Poznan and Krakow (Poland), Brno and Praha (Czech Republic) and Dublin (Ireland). As can be observed, the FUAs surveyed are heterogeneous in terms of size, geographical location, and urban form (population density, urban sprawl, etc.).

Fig. 1.

Geographical coverage of the survey.

The survey campaign was carried out from March to May 2021, when lockdowns were no longer in place, but some mobility and sanitary restrictions were still maintained by governments. A total of 10,152 individuals (approximately 500 individuals per city) were surveyed, but 89 observations were removed due to missing data or lack of consistency in the responses. For each city, 400 responses were collected using a CAWI (Computer-Assisted Web Interviewing) methodology. The remaining 100 responses were collected through CATI (Computer-Assisted Telephone Interviewing) questionnaires to ensure that socio-economic groups that are not reachable through online surveys were adequality represented in the sample.

The survey collected information concerning: i) individual and household socio-economic characteristics (e.g. gender, age, level of income, etc.); ii) built environment attributes (e.g. residential location, household distance to public transport stops or stations, etc.); and iii) self- reported mobility habits before and after the advent of COVID-192 (e.g. frequency of use of private car, shared mobility services, public transport, etc. before and after the advent of COVID-19 periods, reasons behind the changes in travel behaviour, etc.). The distribution of the main of these variables for the sample is displayed in Table 1 .

Table 1.

Distribution of socio-economic characteristics in the sample.

			Respondents/Mean	% (EU average*)/SD
Socio-economic characteristics	Gender	Male	4,463	44.4 (51.1)
		Female	5,600	55.6 (48.9)
	Age	16 – 25 years old	1,597	15.9 (10.6)
		26 – 35 years old	2,017	20.0 (32.7)**
		36 – 49 years old	2,994	29.8 (32.7)**
		Over 49 years old	3,455	34.3 (35.5)
	Employment & Teleworking conditions (pre-COVID)	Employed – No teleworking	5,430	54.0
		Employed – Teleworking	954	9.5
		Unemployed	803	8.0
		Student	889	8.8
		Other	1,987	19.8
	Employment & Teleworking conditions (post-COVID)	Employed – No teleworking	3,260	32.4
		Employed – Partial teleworking	1,574	15.6
		Employed – Full time teleworking	1,550	15.4
		Unemployed	803	8.0 (6.6)
		Student	889	8.8
		Other	1,987	19.8
	Level of Education	Primary and secondary	1,190	11.8 (23.6)
		Post-secondary and Tertiary	4,431	44.0 (45.3)
		University	4,302	42.8 (31.1)
		Prefer not to say	140	1.4
	Household Income	Low	1,993	19.8
		Lower – middle	1,764	17.5
		Middle	1,372	13.6
		Higher – middle	1,259	12.5
		High	1,184	11.8
		Do not know	1,370	13.6
		Prefer not to answer	1,121	11.1
	No. Children in the household (numerical)		0.69	1.087
Built environment attributes	Residential location	Core city	7,092	70.5
		Commuting area	2,971	29.5
	HH Distance to Public Transport	Walking distance	7,976	79.3
		Further than walking distance	1,545	15.3
		Do not know	542	5.4
	HH Distance to the work/study centre	<5 km	3,091	30.7
		Between 5 km and 20 km	3,046	30.3
		Between 20 and 50 km	858	8.5
		More than 50 km	278	2.8
		Others	2,790	27.7
Mobility habits	Private car usage (pre-COVID)	Never	2,590	25.7
		Less than once a week	914	9.1
		About once a week	907	9.0
		2–3 days per week	1,813	18.0
		5 days per week or more	3,839	38.1
	Public transport usage (pre-COVID)	Never	1,977	19.6
		Less than once a week	3,145	31.3
		About once a week	1,330	13.2
		2–3 days per week	1,472	14.6
		5 days per week or more	2,139	21.3
	Change in private car usage (post-COVID)	Decrease	2,441	24.3
		Same	5,536	55.0
		Increase	2,086	20.7
	Number of cars per adult in household pre-COVID (numerical)		0.641	0.431
	Change in car ownership due to COVID	No change in car ownership	8,988	89.3
		Bought a car	812	8.1
		Sold a car	263	2.6
	Bike-sharing usage (pre-COVID)	Non-users	8,002	79.5
		Occasional	1,540	15.3
		Frequent	521	5.2
	Scooter-sharing usage (pre-COVID)	Non-users	7,778	77.3
		Occasional	985	9.8
		Frequent	309	3.1
		Not available	991	9.8
	Car-sharing usage (pre-COVID)	Non-users	7,439	73.9
		Occasional	1,232	12.2
		Frequent	401	4.0
		Not available	991	9.9

^*The percentage for the whole EU population according to Eurostat database has been included when available.

^**Eurostat data group individuals between 26 and 49 years old into a single category.

The preliminary analysis of the sample shows that the survey provides an accurate representation of European socio-demographics and includes an adequate level of heterogeneity for the purposes of this research. Nevertheless, some groups seem to be over-represented when compared to official EU statistics (Eurostat, 2021b), which may be due to the fact that the survey was exclusively conducted in urban areas. This may explain the somehow higher presence of individuals aged between 24 and 49 in the sample (50 %) compared to official EU statistics (32 %). This is presumably due to urban areas having a younger population than rural areas, as noted by The World Bank (2016). The same is observed for e.g. the level of education, since 42.7 % of surveyed individuals reported to have a university degree, while this percentage goes down to 31.1 % for the EU population according to Eurostat, 2021a. Again, this is probably due to the higher concentration of qualified jobs in urban areas that tend to attract more educated workers (Hartal et al., 2021).

The analysis of the level of income is particularly difficult given the heterogeneity in prices and purchasing power across EU countries. To allow for a fair comparison, income levels originally collected (with numerical categories) are adjusted as relative income levels considering the living standards in each country. It is worth noting that low-income households are the most frequent group in the sample (19.8 %). Percentages decrease progressively as the income level increases, with high-income households being the smallest category (11.7 % of the sample). The survey also provides interesting input concerning employment and teleworking conditions, as well as changes observed during COVID-19, since previous and posterior employment situations were collected. This temporal distinction allowed the quantification of the increase in teleworking potentially induced by COVID. Interestingly, the percentage of workers benefiting from any type of teleworking regime rose from 10 % to 30 % of the sample during COVID. Results concerning other occupation categories, such as unemployed people, are perfectly consistent with the official unemployment rates in the EU Area at the time the survey was carried out (Eurostat, 2021c).

Table 1 also provides an overview of the household-related built environment across individuals in the sample. Most of them (70.5 %) reported that they live within the core city of the Functional Urban Area (FUA) and within walking distance from a public transport station or stop (79.3 %). These inputs suggest that a significant part of the surveyed individuals live in high density areas, well connected to public transport, in line with empirical findings as regards European urban layouts (see e.g., Bassolas et al., 2019). Furthermore, 61 % of respondents have declared to live closer than 20 km away from their workplace.

Finally, Table 1 includes the sample statistics concerning mobility habits before and after the advent of COVID-19. The results indicate high motorization rates among surveyed individuals. The average number of cars per adult in the household pre-COVID is 0.64 cars per household. We can also observe that the majority pre-COVID profile includes users of private vehicle who travelled by car 5 or more days per week (38.1 %). The second biggest group corresponds to individuals who did not use private car before COVID-19 (25.7 %). Findings particularly interesting are provided concerning changes in private car usage induced by COVID-19. Almost half of the individuals (45.0 %) reported that the pandemic has induced changes in their frequency of use of private car. Specifically, 24.3 % of respondents declared that they use their car less frequently compared to pre-COVID times, while 20.7 % reported that their usage of private car has increased. It is worth noting that the built environment conditions described above would imply that most of the surveyed individuals have the conditions in place to reduce their car dependency, given their proximity to the public transport network in their city. As for the use of public transport, almost half of individuals in the sample (44.5 %) used it occasionally (about once per week or less), and 35.9 % of them chose this mode 3 times per week or more. This is consistent with the modal share generally observed in European cities, despite the great variability across cities (EIT Urban Mobility, 2021).

The results concerning the frequency of use of shared mobility 3 indicate the still growing adoption of these services. For all the three mobility options reported (bike-sharing, scooter sharing and carsharing), the majority of respondents (percentages between 70 % and 80 %) stated that they do not use these services, which seems reasonable given the still minor presence of these modes in urban modal split. The most frequently used service seems to be bike-sharing, with 15.3 % of occasional users and a 5.2 % of frequent users in the sample. For scooter-sharing and car-sharing, the category not available has been included to capture the fact that these services are not available in certain FUAs surveyed in this campaign.

3.2. Survey analysis – Preliminary findings

This section conducts a preliminary analysis on the sample distribution concerning the main variables of interest for this research, particularly change in car ownership and car use derived from COVID-19. To that end, the distribution of these two variables have been graphically represented across the most relevant socio-economic and mobility variables collected in the survey.

Fig. 2 displays the distribution for the variable capturing change in car ownership due to COVID-19. We can observe that the effect of age is clearly (and surprisingly) negative, so that those households with individuals between 16 and 25 years old present a higher proportion of car buyers. Interestingly, full-time teleworkers also purchased more cars than any other employment category, followed by partial teleworkers and students. Similarly, middle- and middle-low-income households and residents living in the city core are more likely to have bought a car, although the differences with the rest of the categories included in the corresponding variables are less pronounced. As for mobility variables, it is worth mentioning the clear positive effect of initial levels of car ownership on the probability of having bought a car due to the pandemic. In addition, the distribution for the variable capturing the use of private car in pre-COVID times shows that car purchases are more frequent among individuals who used their private car only occasionally (about once per week or less) before the pandemic. This may suggest an increase in the motorization rates of citizens that seldom relied on their private car before the pandemic.

Fig. 2.

Sample distribution of the change in car ownership due to COVID.

Furthermore, Fig. 3 presents the distribution of the variable change in car use after COVID-19 across some of the most important socio-economic and mobility variables. It can be clearly observed that the proportion of individuals who have decreased their car use is higher among older adults, compared to their younger counterparts. Additionally, it can be observed that the age group between 26 and 35 years old presents a higher proportion of individuals who have used their car more often after COVID-19. The effect of income is also worth highlighting, as while households with higher incomes are the ones experiencing larger reductions of car use, low-income households are the most likely ones to keep similar levels of car-based mobility. The relationship between the number of cars owned by the household before the pandemic and changes in car use does not seem to follow any clear trend. By contrast, households having purchased a car due to the pandemic tend to have increased their car use, and those that have sold a car have decreased its use, as seems reasonable. The distribution for the variable controlling for car use before COVID shows that those households that have increased more their use of car were frequent users (more than 3 days per week), while occasional users (about once a week or less) are the ones who display bigger reductions. Finally, the plot shows that frequent users of public transport before the pandemic are more likely to have increased their car use, except for those that used it five times per week or more. For this group, the observed increase in the frequency of car use is lower than for all the other profile. On the other hand, they have also decreased their car use the least.

Fig. 3.

Sample distribution of the change in car use after COVID-19.

3.3. City level data

The information collected from the European Urban Mobility Survey was complemented by a set of city-level variables to take into account the potential influence of the specific context of each urban area on car usage and ownership changes. These variables are intended to capture to some extent the demographic, economic, urban and accessibility features of each city, and include: i) city population, which allows to explore whether COVID-19 might have affected car ownership and usage in FUAs of different sizes; ii) city GDP per capita (US$), aiming to explore whether the economic conditions of a FUA might partly explain the effect of COVID-19 in urban mobility; and iii) the urban form measured as the share of the city core with respect to the whole area of the FUA. In this respect, since commuting areas are generally characterized by less density and lower land-use mixes, their relative size is likely to influence car dependence both before and after the pandemic. These three variables have been obtained from the OECD Regions and Cities database (OECD, 2021a).

The analysis also included the modal performance of car, bike and public transport in each city, according to the Urban Access Framework proposed by the OECD (OECD, 2021a). These performance indices are aimed at capturing the speed, frequency, and density of the network for biking, public transport, and private car. Mode performance of a certain mode is defined as the ratio of the population that can be reached in 30 min by this mode, and the population within 8 km (2 km for walking). A complete explanation of the methodology adopted to calculate mode performance can be found in ITF (2019). These variables were firstly proposed by OECD in 2019, therefore they do not capture changes in accessibility during or due to COVID-19.

All city-level attributes have been transformed into nominal variables. For each variable, there are four categories, each of which group 25 % of the cities, and one category that includes the cities for which the variable was not available. This categorization has been done because, as the delimitations of the cities in the survey follow the definition of FUAs provided by the OECD, a set of city-level variables consistent with this delimitation was only found in the OECD Regions and Cities database. This database does not provide any variables for non-OECD territories (Romania), and only a few of them for some smaller cities (Charleroi or Calais). Therefore, to keep these FUAs in the study, it was necessary to create a “no data available” category to group these cases.

4. Methodology

4.1. Modelling framework – Path analysis

In order to explore the potential changes induced by COVID-19 on car ownership and car usage in EU urban areas, a Path Analysis framework has been adopted in this research. Path Analysis (PA) is a statistical technique that allows to model situations in which a certain variable is dependent on a set of explanatory factors and can be, in turn, considered as an explanatory factor for another dependent variable (Streiner, 2005).

This approach allows the disentanglement of the direct and indirect effects of an exogenous factor through an intermediate outcome variable. It should be noted that a series of assumptions regarding the causal relationships that exist among the endogenous variables included in the modelling is necessary (Lleras, 2005). PA can be considered as an extension of the traditional regression models in which several sub-models are estimated at the same time, accounting for the correlations between all the endogenous and exogenous variables included in the model (Land, 1969). PA is an econometric technique widely adopted in the field of transport research, and the reader is referred to e.g., Streiner (2005) for further information.

Initially, together with the PA, a nested multilevel model with a random intercept for each of the four endogenous variables was proposed to account for potential differences across European macro regions (Mediterranean, Western, Eastern and Northern), countries, and cities. For further information on multilevel modelling, the reader is referred to Peugh, 2010. The unconditional means model (also known as null model) was estimated by including only the random intercepts in order to test whether the variation between European macro regions, countries and cities was significant. Thereafter, the rest of the exogenous variables were included.

4.2. Model structure

The model defined for this research follows the sequential structure represented in Fig. 4 and includes four endogenous or outcome variables. The first two outcome variables capture individuals’ car ownership and frequency of use of private car before the pandemic, while the last two capture potential changes in car ownership and usage induced by COVID-19.

Fig. 4.

Path diagram of the proposed model.

The exogenous variables included as explanatory factors can be grouped in three types. Firstly, the individual and household’s socio-economic characteristics and travel behaviour before COVID-19 (frequency of use of public transport and shared mobility) constitute the main focus of this work. Nevertheless, the research controls for the effect of other additional variables. Secondly, some additional variables capturing the level of accessibility of the household have been included, given that a clear influence of this factor on car use and ownership has been previously observed in the literature (van Acker and Witlox, 2010, Ding and Cao, 2019). Third, a set of city-level attributes has been included to capture certain aspects that may influence private car ownership and usage (population, level of urban sprawl or performance indicators of different transport modes), for each of the 21 cities included in the study.

As can be observed in Fig. 4, the socio-demographics of individuals and households, together with built environment and city attributes, influence car ownership during pre-COVID times. Subsequently, private car usage during pre-COVID times is considered influenced by individuals’ and household socio-demographics, built environment and city attributes, usage of shared mobility services, and car ownership for pre-COVID times. Next, individuals’ and household socio-demographics, together with built environment and city attributes, influence car ownership after the advent of COVID-19. Finally, all the variables mentioned above are considered to influence the change in private car use after the start of the pandemic.

Each of the four outcome variables included in the Path Analysis is described below:

• i.Car ownership before COVID-19. Numerical variable that captures the number of cars per adult in the household before the beginning of the pandemic.
• ii.Frequency of use of private car before COVID-19. Ordinal variable with the following five categories: (1) Never used, (2) Less than once a week, (3) About once a week, (4) Between 2 and 3 days per week, and (5) 5 days per week or more.
• iii.Changes in car ownership due to COVID-19. Categorical variable with three options: (1) No change in car ownership, including those individuals who have neither bought nor sold a car or that have declared to both sell and buy a car, as a result of the pandemic; (2) Bought a car, including those individuals who reported having bought a car (either first-hand or second hand,) due to the COVID-19 pandemic; and (3) Sold a car, including individuals who have sold a car due to the pandemic. It is worth noting that the survey questions employed to build this variable, made explicit reference to COVID-19 as the trigger for the decision to make changes in household car ownership.
• iv.Change in car use frequency after the start of the pandemic. Categorical variable that captures individuals’ frequency of use of private car after the start of the pandemic, compared to pre-COVID times. This variable has three categories: (1) Decreased use; (2) About the same use (reference category); and (3) Increased use.

These four co-endogenous variables are of three different types: linear (‘Number of cars per adult’), ordinal (‘Frequency of private car before COVID-19′) and nominal (‘Changes in car ownership due to COVID-19′ and ‘Change in car use frequency after the start of the pandemic’). The specification of the sub-models for each of these types are linear regression, ordinal logit regression and multinomial logit regression, respectively.

5. Modelling results

This section comments the modelling results obtained from the PA exploring changes in private car ownership and usage motivated by COVID-19. Detailed results for the full model and the whole set of explanatory variables can be found in Appendix A. Nevertheless, for length reasons, only the modelling results for the variables of greater interest for this study (socio-economic characteristics and pre-pandemic travel habits) are displayed in Table 2 below. The results include the regression coefficients for each of the co-endogenous variables included in the model, the p-values associated with each coefficient and the odds ratio (OR). The final specification of the model has been obtained through a process of testing alternative combinations of variables. The variables that were not significant at the 90 % level and for which the likelihood ratio tests showed that their removal from the model did not significantly affect the goodness of fit, were gradually removed from the model to obtain a more parsimonious specification. This approach has been widely adopted in the field of transport research (see e.g., Alemi et al., 2018, Aguilera-García et al., 2020). Additionally, some of the coefficients in the final specification of the model are duplicated, which shows that two or more categories have been merged.4

Table 2.

Modelling results for the number of cars per adult and the frequency of car use before COVID-19.

Variable (base category)	Number of cars per adult before COVID-19 (numerical)	Frequency of car use before COVID-19 (ordinal1)
	Coeff.	Coeff. (OR)
SOCIO-ECONOMIC ATTRIBUTES
Gender (Female)
Male	–	0.353*** (1.423)
Age (16–25 years old)
26–35 years old	−0.033*	0.327*** (1.387)
36–49 years old	−0.037*	0.337*** (1.400)
Over 50 years old	−0.042**	0.570*** (1.768)
Education (Primary and Secondary school)
Post-Secondary and tertiary school	–	0.230*** (1.348)
University	0.048***	0.504*** (1.656)
I prefer not to say	–	0.404* (1.498)
Employment post-COVID (Employed – no TW)
Employed - TW	–	−0.256*** (0.774)
Student	−0.049**	−0.761*** (0.467)
Unemployed	−0.062***	−0.443*** (0.642)
Others	−0.050***	−0.562*** (0.570)
Income (Low income)
Middle – low income	0.068***	0.213*** (1.238)
Middle income	0.113***	0.273*** (1.315)
Middle high income	0.152***	0.139** (1.149)
High income	0.191***	0.270*** (1.309)
I don’t know	0.060***	–
I prefer not to say	0.098***	–
Kids in household	0.055***	0.224*** (1.251)
Kids in household2	−0.008***	−0.027*** (0.973)
TRAVEL BEHAVIOUR BEFORE COVID-19
Public transport use before COVID-19 (5 or more days per week)
3–5 days per week	0.106***	0.605*** (1.832)
About once a week	0.154***	0.945*** (2.572)
Less than once a week	0.235***	1.740*** (5.696)
Never	0.297***	2.013*** (7.485)
Bike sharing use before COVID-19 (non-users)
Occasional	n/a	–
Frequent	n/a	−0.200* (0.818)
Car sharing use before COVID-19 (non-users)
Occasional	n/a	0.219*** (1.244)
Frequent	n/a	0.349*** (1.417)
Not available	n/a	0.256** (1.292)
CO-ENDOGENOUS VARIABLES
Number of cars per adult before COVID-19	n/a	1.995*** (7.354)
Constant	0.291***	n/a
Thresholds
Threshold 1	n/a	1.643***
Threshold 2	n/a	2.290***
Threshold 3	n/a	2.854***
Threshold 4	n/a	3.907***

*p ≤ 0.05, ^**p ≤ 0.01, ^***p ≤ 0.001.

¹This ordinal variable contains 5 categories. Never used < 3–5 days per week < About once a week < Less than once a week.

The results yielded by the construction of the unconditional means model mentioned in the previous section (Section 4.1) revealed that there only exist significant differences across cities for all the four endogenous variables (the random intercepts were not significant at the macro region and country level). However, when the city level attributes were introduced as exogenous variables, the variance of the random intercepts across cities lost its significance. This effect proved that the set of city attributes were able to capture the geographical differences across cities and therefore, the use of a multilevel model was no longer required.

5.1. Cars per adult in the household before COVID-19

The modelling results confirm the strong influence of socio-economic characteristics on car ownership, as previous research contributions have largely concluded. For instance, the influence of income and level of education on the number of cars in the household before the pandemic is in line with previous results obtained in the literature, since both higher incomes and higher levels of education have been associated with higher car ownership (see e.g., Chu, 2002 for New York or Potoglou & Kanaroglou, 2008 for Hamilton). According to the modelling results, respondents aged under 25 have higher levels of car ownership than older respondents (particularly, those between 26 and 49 years old). Given that previous contributions suggest the peak for car ownership in older ages (see e.g., Cornut, 2016 for Paris or Matas & Raymond, 2008 for Spain), further investigation of the data has been done to interpret this result. Younger adults (between 16 and 25 years old) in our sample live in households with an average number of adults significantly higher than other age groups. In this respect, more adults in the household have been associated to a higher individual travel demand and higher levels of car ownership, as noted by e.g., Cornut (2016). As for the effect of mobility patterns, it is worth noting the statistically significant relationship between less frequent use of public transport and the number of cars per adult in the household. In terms of size effects, the analysis of the marginal effects shows that while the predicted average number of cars per adult for those that never use public transport is 0.780, this number goes down to 0.473 for those that use it every day.

The results concerning the effect of the built environment on car ownership levels are coherent with the previous literature. Individuals living in the commuting area, which are characterized by lower densities and less accessibility to public transport, owned more cars before the pandemic. The same trend is observed for those who cannot reach public transport on foot (Zhang et al., 2017). Longer distances to the place of work are clearly related to higher number of cars in the household, also in agreement with other authors such as Bhat & Guo (2007).

5.2. Frequency of car use before COVID-19

The second model in the PA explores the frequency of use of private car before the COVID-19 pandemic, which is captured through an ordinal variable. The modelling results are consistent with the previous literature in the sense that older wealthier males, working in fully in-person regimes and with high levels of education, were the most frequent car users in pre-COVID times, similarly to e.g., Shen et al. (2016) or Yang et al. (2021). For example, an individual belonging to a high-income household is 17 % more likely to use the car 5 or more days than someone living in a low-income household. Furthermore, the results for the age variable are in line with the findings by Nolan, 2010, van Acker and Witlox, 2010, and suggest a positive effect of age in car usage. As for the number of children in the household, they significantly increase the frequency of car use in pre-COVID times. An additional kid in the household entails a 14 % increase of the probability of using the car 5 days per week or more. This is probably because the use of public transport or active modes is less convenient when having to accommodate children. This positive effect has already been observed in several previous studies such as Dargay & Hanly (2007) or Kim & Ulfarsson (2008).

Regarding the effect of public transport frequency of use, there is again a very clear and negative relationship between these two variables. For non-users of public transport before COVID-19, the odds ratio of using the car more frequently is 7.49 higher than for those who use public transport every weekday or more. This value decreases steadily as the frequency of use of public transport increases, suggestion a substitution effect between these two modes. Concerning the effect of shared mobility, the results suggest that the use of these services had a significant effect on the frequency of car use. According to the results, the frequency of use of private car is lower among intensive users of bike sharing, while both frequent and occasional users of car sharing seem to use the car more often than non-users of these services. The analysis of the marginal effects shows that the frequent and occasional use of car-sharing increases, respectively, by 21 % and 13 % the likelihood of using the car five times per week.

5.3. Change in car ownership due to COVID-19

The third sub-model of the PA (see Table 3 ) explores changes in household car ownership due to COVID-19. It should be reminded (see Section 4.2) that this categorical variable included three levels: no change in car ownership (reference category), the household purchased a car due to COVID-19, and the household disposed of a car due to COVID-19.

Table 3.

Modelling results for the car ownership change and the frequency of car use after the start of the pandemic.

Variable (base category)	Car-ownership change due to COVID (No change in car ownership)		Change in frequency of car use after the start of the pandemic (About the same use)
	Bought car	Sold car	Decrease use	Increased use
	Coeff. (OR)	Coeff. (OR)	Coeff. (OR)	Coeff. (OR)
SOCIO-ECONOMIC ATTRIBUTES
Gender (Female)
Male	–	–	–	–
Age (16 – 25 years old)
26 – 35 years old	−0.558*** (0.572)	–	–	–
36 – 49 years old	−0.820*** (0.441)	–	–	−0.266*** (0.767)
Over 50 years old	−1.170*** (0.310)	−0.481*** (0.618)	–	−0.479*** (0.619)
Education (Primary and Secondary school)
Post-Secondary and tertiary
school	–	–	–	–
University	–	–	0.236***	–
I prefer not to say	–	–	–	–
Employment post-COVID (Employed – no TW)
Employed - full TW	0.522*** (1.686)	0.316* (1.372)	1.033*** (2.811)	–
Employed - partial TW	0.332*** (1.394)	0.316* (1.372)	0.721*** (2.057)	–
Student	–	–	–	–
Unemployed	–	0.569** (1.767)	0.390*** (1.477)	–
Others	–	–	0.492*** (1.635)	–
Income (Low income)
Middle – low income	0.309** (1.362)	–	0.172** (1.188)	0.187** (1.205)
Middle income	0.282** (1.326)	–	0.172** (1.188)	–
Middle high income	–	−0.461** (0.631)	0.179* (1.196)	0.187* (1.206)
High income	–	−0.461** (0.631)	0.607*** (1.835)	–
I don’t know	–	–	–	–
I prefer not to say	–	–	–	−0.207* (0.813)
Kids in household	0.309*** (1.362)	–	–	–
Kids in household2	−0.032** (0.969)	–	–	–
TRAVEL BEHAVIOUR BEFORE COVID-19
Public transport use before COVID-19 (5 or more days per week)
3–5 days per week	–	0.329* (1.389)	–	–
About once a week	–	–	1.792* (1.215)	−0.323*** (0.724)
Less than once a week	–	–	1.278*** (3.591)	−0.549*** (0.578)
Never	−0.392*** (0.676)	–	1.410*** (4.098)	−0.802*** (0.449)
Bike sharing use before COVID-19 (non-users)
Occasional	n/a	n/a	–	0.443*** (1.556)
Frequent	n/a	n/a	–	–
Car sharing use before COVID-19 (non-users)
Occasional	n/a	n/a	0.194* (1.214)	0.463*** (1.588)
Frequent	n/a	n/a	0.503*** (1.653)	0.468*** (1.597)
Not available	n/a	n/a	–	0.429*** (1.520)
CO-ENDOGENOUS VARIABLES
Number of cars per adult	–	0.721*** (2.056)	–	–
Car use before COVID-19 (5 or more days per week)
3–5 days per week	–	–	1.792*** (15.992)	0.398*** (1.489)
About once a week	–	0.423* (2.513)		–
Less than once a week	–	0.921*** (1.526)	2.542*** (12.710)	−0.601*** (0.548)
Never	−0.414*** (0.661)	–	2.772*** (6.001)--	−3.135*** (0.043)
Car ownership change due to COVID-19 (no change)
Bought a car	n/a	n/a	–	0.622*** (1.863)
Sold a car	n/a	n/a	0.736*** (2.089)	–
Constant	−2.101***	−4.038***	−2.234***	−0.232**
Goodness of fits statistics
Log-Likelihood (model)	−29,492.2		AIC	59,358.5
Log-Likelihood (null)	−34,573.6		BIC	60,708.0
LR test	χ2 = 10162.82, p-value = 0.000

*p ≤ 0.05, ^**p ≤ 0.01, ^***p ≤ 0.001.

5.3.1. Car purchases

As could be expected, socio-demographic characteristics have a key role in explaining car purchases due to COVID-19. As for age, it has a statistically significant negative effect on the probability of having purchased a car and shows that households with younger adults are more likely to have bought a car due to the pandemic. In addition to the strong statistical relationship, the analysis of the average marginal effects shows that the size of this effect is considerable. For example, individuals between 36 and 49 are 73 % less likely to have bought a car than individuals under 25. The effect of teleworking becomes particularly strong when explaining car purchases due to COVID-19, revealing an unexpected pattern: full-time and partial teleworkers are respectively 46 % and 31 % more likely on average to have bought a car due to the pandemic than in-person workers. This effect may be due to the willingness to accept larger commuting distances by teleworkers and the consequent residence relocation to suburban areas with less accessibility by public transport, as suggested by e.g., de Abreu e Silva & Melo (2018) for Great Britain. The effect of the income variable on the probability of purchasing an additional car due to COVID-19 is only significant for middle low- and middle-income households, with a higher probability of having increased their car ownership than all the other groups. However, for these categories the size effects are rather small, as both groups display an increase of around 0.1 % in the probability of having bought a car with respect to low-income groups.

Past travel behaviour does not seem to have a relevant explanatory power on car purchases during COVID. The model only concludes that non-users of both private car and public transport before the pandemic are less likely to have bought a car due to COVID, suggesting that the pandemic has not led car-dependent individuals to introduce the private car in their daily transport habits.

Among the city level attributes, only city size and urban form have a statistically significant effect according to the results. The first variable shows that households in cities with more than 3 million inhabitants (Madrid, Paris, Berlin and Milan) or whose city core covers more than 30 % of the FUA surface (Dresden, Malmo, Porto and Charleroi) are less likely to have bought a car. None of the mode performance variables were found to significantly impact the decision to buy a car due to COVID-19. This may point out that these purchases were driven by changes induced by the sanitary crisis rather than by an evaluation of the most convenient mode of transport in normal conditions.

5.3.2. Sold a car

The modelling results indicate that younger and middle-aged adults, together with unemployed individuals, are the socio-demographic groups most likely to have sold a car due to the COVID-19 pandemic. As for the role of income, it can only be stated that middle- and middle-high income households are less likely to have sold a car compared to other income groups, as could be observed in the preliminary analysis conducted in Section 3.2. It should be reminded that income had a clear statistically significant positive effect on car ownership levels before COVID (see results in Section 5.1). Therefore, these results obtained for the period after the start of the pandemic might suggest that middle-high-income and high-income households had enough cars for the reduced mobility of the pandemic to allow the possibility of forgoing a car, while the car ownership levels of poorer household might not have been high enough for the pandemic to reduce them even further. This hypothesis is supported by the positive and significant, although small, effect of the number of cars per adult on the probability of selling one. An additional car per adult leads to a 2 % increase in the probability of having sold a car.

5.4. Change in frequency of car use after the start of the pandemic

Finally, we comment on the results for the last variable of interest, which is the change in the frequency of car use after COVID-19. This is captured by a categorical variable with three categories: Decreased use, About the same use (reference category) and Increased used. The analysis of the socio-economic characteristics driving the change in frequency of use points to an induced demand for cars, whereby those segments of the population that were considered to have lower levels of car use in the pre-COVID sub-models have increased their car use more than the rest of individuals. For the case of age, which had a clear positive effect on car use before COVID (that is, higher frequency among older individuals), the model finds that after the start of the pandemic adults over 39 years old are more likely to have reduced car use frequency than younger segments of the population. This “rebound effect” is also observed for the place of residence within the metropolitan area. The average marginal effects show that, before the pandemic living in the commuting area increased the probability of using the car five times per week or more by 11 %. After the advent of COVID, this group of individuals is on average 14 % less likely to have increased its car use. It is worth noting that a similar effect has been pointed out by Schaefer et al. (2021) in Hanover (Germany). An equivalent effect on the opposite direction is observed for the case of level of education and high-income. While both attributes were positively correlated with car use before the start of the pandemic, they now have a positive effect on the probability of decreasing car use. After the advent of COVID-19, the odds of having reduced their car use for individuals with university education and belonging to high-income households are, respectively, 1.3 and 1.8 times higher than for the reference categories.

The model yields very clear results for the different user profiles of public transport before COVID-19: frequent users (those who used it three times per week or more) are more likely to have increased their frequency of car use than occasional (about once per week or less) or non-users of public transport. The results for the shared mobility variables show that occasional users of bike-sharing prior to COVID-19 are more likely to have increased their car use. Being an occasional user of shared bikes before COVID-19 increased by 21 % the probability of using the car more often after the start of the pandemic. As for car-sharing services, both frequent and occasional users are more likely to have increased their frequency of car use than non-users. The analysis of the odds ratios allows to draw some conclusions regarding the relationship between shared and private cars after the advent of COVID-19. Before the start of the pandemic, the odds ratio of using the car more often than non-users was 1.42 and 1.24 for frequent and occasional users, respectively. After the start of the pandemic, all types of car-sharing users are more likely to have increased their car use, having both occasional and frequent users around 60 % (OR = 1.6) higher odds than non-users to have increased their car use.

Some interesting insights are found when analysing the relationship between car use before and after the start of the pandemic. The probability of keeping the same car mobility levels for non-users of the private car is 35 % higher that for those within the base category (5 or more days per week). On the other hand, occasional users of the car before the pandemic are significantly more likely to have decreased their car use than the base category. The effect size of this relationship is substantial: the average marginal effect for those that use the car less than once a week is 1.42 and for those that use it approximately once per week is 1.11. Finally, the effect of changes in household car ownership confirms the coherence of the model, as those who purchased a car have increased their car use and those who sold one have reduced its use.

In addition to the main effects observed in the model, a set of interactions between some socio-economic attributes and past travel habits have been introduced for the post-COVID sub-models. Nevertheless, statistically significant results were only obtained for the “Change in frequency of car use” regression. It should be noted that the interactions’ coefficients have been simultaneously estimated together with all the explanatory factors that were included in the model. However, to improve the readability of this paper, the coefficients for the interactions found statistically significant are displayed separately in Table 4 . As mentioned before, the table with the entire model can be found in the Appendix A.

Table 4.

Modelling results for the interactions included in the Change in car use after start of the pandemic.

Variable (base category)	Change in frequency of car use after the start of the pandemic (About the same use)
	Decreased use	Increased use
	Coeff.	Coeff.
Employed - partial TW#Car use pre-COVID: less than once a week	−0.838***	–
Employed - partial TW#Car use pre-COVID: about once a week	−1.059***	–
Employed - full TW#Car use pre-COVID: less than once a week	−0.989***	–
Employed - full TW#Car use pre-COVID: about once a week	−0.806***	–
Employed - full TW#Car use pre-COVID: between 3 and 5 times per week	0.617***	–
Medium-low-income#Car use pre-COVID: between 3 and 5 times	−0.447*	–
High-income#Car use pre-COVID: less than once a week	−0.744**	–
High-income#Car use pre-COVID: about once a week	−0.764**	–
High-income#Car use pre-COVID: between 3 and 5 times	−0.424*	–
Bike sharing use pre-COVID: occasional#Car use pre-COVID: less than once a week	0.562***	–
Bike sharing use pre-COVID: occasional#Car use pre-COVID: about once a week	–	−0.749***
Bike sharing use pre-COVID: occasional#Car use pre-COVID: between 3 and 5 times	−0.399**	−0.619***
PT use pre-COVID: never#Car use pre-COVID: less than once a week	−1.174***	–
PT use pre-COVID: never#Car use pre-COVID: about once a week	−1.292***	–
PT use pre-COVID: never#Car use pre-COVID: between 3 and 5 times	−1.165***	−0.583**
PT use pre-COVID: less than once a week#Car use pre-COVID: never	–	1.617***
PT use pre-COVID: less than once a week#Car use pre-COVID: less than once a week	−1.658***	–
PT use pre-COVID: less than once a week#Car use pre-COVID: about once a week	−1.556***	–
PT use pre-COVID: less than once a week#Car use pre-COVID: between 3 and 5 times	−0.747***	–

*p ≤ 0.05, ^**p ≤ 0.01, ^***p ≤ 0.001.

High-income seems to slightly moderate the decrease in car use that was observed among those who used the car less than five times per week. The interactions between car use and bike sharing use before COVID-19 show that among occasional users of shared bikes, those who used the car less than once a week are more likely to have further decreased their use of car, while occasional users of shared bikes who used the car between 3 and 5 days per week are more likely to keep the same level of car use. Some interesting results have been observed for the interactions between public transport and car use before COVID-19. Individuals who never used public transport before COVID-19, or did it less than once a week, and who did not use the car daily, are slightly more likely to maintain the same levels of car use as before the start of the pandemic. On the other hand, it is noteworthy that individuals who never used the car and used public transport less than once a week are more likely to have increased their car use.

6. Discussion and policy recommendations

The analysis of the modelling results of the PA yields some relevant conclusions regarding the mid-terms effects of the COVID-19 pandemic in car use and car ownership in European cities and gives rise to propose some policy recommendations to make urban mobility more sustainable.

• (1)Reversal effect of the COVID pandemic on previous trends in car usage. The combined analysis of travel choices before and after the advent of COVID-19 allowed to capture a somewhat reversal effect of the pandemic on several trends regarding car usage, observed both in our model and in previous literature. This effect is observed among younger individuals (aged 16–36) and residents in the core city, who were previously identified as low car dependency groups but who have significantly increased their car use during COVID-19. Addressing the growing number of users who have shifted to the private car as a result of COVID-19 should be a high priority, since it can reverse past trends in reducing the negative externalities of urban transport. The provision of good quality, affordable and perceived safe public transport is essential to invert this trend again.
• (2)Reduction in car use among high-income, well-educated teleworkers. The model shows that high-income and high levels of education are correlated with a reduction in car use after the advent of COVID-19. These attributes are also characteristics associated with the typical profile of teleworkers (Lopez Soler et al., 2021). The model corroborates this relation as it shows that both partial and full-time teleworkers are using their car less often after the start of the pandemic. Taken together, these results would suggest that the increase in teleworking induced by the pandemic has been successful as regards the reduction in car use. Nevertheless, it also shows that teleworking as a tool to reduce car use for urban trips, is a policy instrument with a limited scope, given that its benefits seem to be restricted to privileged socio-economic groups. Additionally, the model shows that low-income households are the least likely ones to have reduced their car use and are mostly maintaining similar levels of car mobility. As low-income and low-educated individuals are in general less prompt to telework (Sweet and Scott, 2022), policies should be deployed to ensure that these groups can cover their mobility needs in an environmentally and economically sustainable manner in the context of the post-pandemic.
• (3)Migration of frequent public transport users to the private car. The results for the different profiles of public transport users before and after the start of the pandemic show that the crisis has clearly changed the behavior of individuals with respect to these two modes in the urban context. While the model shows a strong negative relationship between frequency of use of public transport and car usage before the pandemic, after the lockdown, frequent users of public transport (three times per week or more) are clearly more likely to have increased their car use than occasional or non-users. This effect is probably suggesting that a large proportion of individuals who used public transport as their main mode of transport for daily trips have switched to the private car, confirming the strong substitution effect between these two modes induced by the pandemic. Interestingly, this trend is not extensible to changes in car ownership, as those who never used public transport before the pandemic are the least likely group to have bought a car due to-COVID-19.
• (4)Diverging patterns in car-sharing users. The modelling results show that the adoption of car sharing has a statistically significant effect on car use both before and after the advent of COVID-19. Before the start of the pandemic, a more frequent use of shared cars was associated with a higher frequency of car use, revealing a complementary relationship between them. During COVID-19, users of car-sharing are both more likely to have increased and decreased their car use, showing that the pandemic has led to a polarization with respect to car use among car-sharing users.

7. Conclusions and further research

This study conducted a PA model with the aim of exploring the extent to which to the COVID-19 pandemic has motivated changes in household car ownership and usage in European urban areas, as well as the factors that determine them. To that end, the research exploits an EU-wide survey of residents in 21 European cities of different sizes and urban forms. Thus, the study presents a comprehensive picture of the current role of the private car across European cities and may be a useful input for policy makers both at the national and European level. The model provides insights on how socio-demographic characteristics of the individual and the household, as well as their travel habits before the start of the pandemic, have affected the changes in their car-related mobility after the advent of the virus.

One of the main strengths of this work is the broad geographical scope, and the variety of explanatory variables and chained effects that are included in the model. However, this amplitude somewhat limits the study of the specific factors that might be affecting each particular outcome variable. In this regard, some avenues for further research can be suggested. First, an exhaustive study of the reasons that have motivated the purchases or sales of cars during the pandemic, the consideration of the type of car that has been purchased, or the inclusion of information on the main car user within the household would provide a clearer picture of the role of the car in European households. Second, the widespread adoption of teleworking regimes during the pandemic has probably induced the relocation of residence farther away from the workplace in many cases, which may have important implications for mobility patterns. The exploration of such choices in the definition of the PA model would allow the identification of further interrelations between residential changes, car ownership changes, and car usage habits.

EU disclaimer

The views expressed are purely those of the authors and may not in any circumstances be regarded as stating an official position of the European Commission.

Funding

This research was developed within the Collaborative Doctoral Partnership program between the Joint Research Centre of the European Commission and Centro de Investigación del Transporte (TRANSyT) of Universidad Politécnica de Madrid [Agreement n° 35364].

This work has been funded through the agreement between Region of Madrid and Universidad Politécnica de Madrid for the direct granting to finance research activities on SARS-CoV-2 and COVID-19 disease, financed with REACT-EU resources from the European Regional Development Fund (E.MORES project).

CRediT authorship contribution statement

Maria Vega-Gonzalo: Conceptualization, Methodology, Formal analysis, Writing – original draft. Juan Gomez: Conceptualization, Methodology, Supervision, Formal analysis, Resources. Panayotis Christidis: Conceptualization, Data curation, Supervision, Resources.

Declaration of Competing Interest

The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.

Appendix A. Modelling results including all explanatory variables for the four sub-models

Variable (base category)	Number of cars per adult before COVID-19 (numerical)	Frequency of car use before COVID-19 (ordinal)	Change in frequency of car use after the start of the pandemic (About the same use)	Change in frequency of car use after the start of the pandemic (About the same use)	Change in frequency of car use after the start of the pandemic (About the same use)
			Bought a car	Sold a car	Decreased use	Increased use
	Coeff.	Coeff. (OR)	Coeff. (OR)	Coeff. (OR)	Coeff. (OR)	Coeff. (OR)
SOCIO-ECONOMIC ATTRIBUTES
Gender (Female)
Male	–	0.353*** (1.423)	–	–	–	–
Age (16–25 years old)
26–35 years old	−0.033*	0.327*** (1.387)	−0.558*** (0.572)	–	–	–
36–49 years old	−0.037*	0.337*** (1.400)	−0.820*** (0.441)	–	–	−0.266*** (0.767)
Over 50 years old	−0.042**	0.570*** (1.768)	−1.170*** (0.310)	−0.481*** (0.618)	–	−0.479*** (0.619)
Education (Primary and Secondary school)
Post-Secondary and tertiary school	–	0.230*** (1.348)	–	–	–	–
University	0.048***	0.504*** (1.656)	–	–	0.236***	–
I prefer not to say	–	0.404* (1.498)	–	–	–	–
Employment pre-COVID (Employed – no TW)
Employed - TW	–	−0.256*** (0.774)	n/a	n/a	n/a	n/a
Student	−0.049**	−0.761*** (0.467)	n/a	n/a	n/a	n/a
Unemployed	−0.062***	−0.443*** (0.642)	n/a	n/a	n/a	n/a
Others	−0.050***	−0.562*** (0.570)	n/a	n/a	n/a	n/a
Employment post-COVID (Employed – no TW)
Employed - full TW	n/a	n/a	0.522*** (1.686)	0.316* (1.372)	1.033*** (2.811)	–
Employed - partial TW	n/a	n/a	0.332*** (1.394)	0.316* (1.372)	0.721*** (2.057)	–
Student	n/a	n/a	–	–	–	–
Unemployed	n/a	n/a	–	0.569** (1.767)	0.390*** (1.477)	–
Others	n/a	n/a	–	–	0.492*** (1.635)	–
Income (Low income)
Middle – low income	0.068***	0.213*** (1.238)	0.309** (1.362)	–	0.172** (1.188)	0.187** (1.205)
Middle income	0.113***	0.273*** (1.315)	0.282** (1.326)	–	0.172** (1.188)	–
Middle high income	0.152***	0.139** (1.149)	–	−0.461** (0.631)	0.179* (1.196)	0.187* (1.206)
High income	0.191***	0.270*** (1.309)	–	−0.461** (0.631)	0.607*** (1.835)	–
I don’t know	0.060***	–	–	–	–	–
I prefer not to say	0.098***	–	–	–	–	−0.207* (0.813)
Kids in household	0.055***	0.224*** (1.251)	0.309*** (1.362)	–	–	–
Kids in household2	−0.008***	−0.027*** (0.973)	−0.032** (0.969)	–	–	–

*p ≤ 0.05, ^**p ≤ 0.01, ^***p ≤ 0.001.

Variable (base category)	Number of cars per adult before COVID-19 (numerical)	Frequency of use before COVID-19 (ordinal)	Car-ownership change due to COVID-19 (No change in car ownership)		Change in frequency of car use after the start of the pandemic (About the same use)
			Bought a car	Sold a car	Decreased use	Increased use
	Coeff.	Coeff. (OR)	Coeff. (OR)	Coeff. (OR)	Coeff. (OR)	Coeff. (OR)
TRAVEL BEHAVIOUR BEFORE COVID-19
Public Transport use before COVID-19 (5 or more days per week)
3–5 days per week	0.106***	0.605*** (1.832)	–	0.329* (1.389)	–	–
About once a week	0.154***	0.945*** (2.572)	–	–	1.792* (1.215)	−0.323*** (0.724)
Less than once a week	0.235***	1.740*** (5.696)	–	–	1.278*** (3.591)	−0.549*** (0.578)
Never	0.297***	2.013*** (7.485)	−0.392*** (0.676)	–	1.410*** (4.098)	−0.802*** (0.449)
Bike sharing use before COVID-19 (non-users)
Occasional	n/a	–	n/a	n/a	–	0.443*** (1.556)
Frequent	n/a	−0.200* (0.818)	n/a	n/a	–	–
Car sharing use before COVID-19 (non-users)
Occasional	n/a	0.219*** (1.244)	n/a	n/a	0.194* (1.214)	0.463*** (1.588)
Frequent	n/a	0.349*** (1.417)	n/a	n/a	0.503*** (1.653)	0.468*** (1.597)
Not available	n/a	0.256** (1.292)	n/a	n/a	–	0.429*** (1.520)
CO-ENDOGENOUS VARIABLES
Number of cars per adult before COVID-19	n/a	1.995*** (7.354)	–	0.721*** (2.056)	–	–
Car use before COVID-19 (5 or more days per week)
3–5 days per week	n/a	n/a	–	–	1.792*** (15.992)	0.398*** (1.489)
About once a week	n/a	n/a	–	0.423* (2.513)	2.542*** (12.710)	–
Less than once a week	n/a	n/a	–	0.921*** (1.526)	2.772*** (6.001)	−0.601*** (0.548)
Never	n/a	n/a	−0.414*** (0.661)	–	–	−3.135*** (0.043)
Car ownership change due to COVID-19 (no change)
Bought a car	n/a	n/a	n/a	n/a	–	0.622*** (1.863)
Sold a car	n/a	n/a	n/a	n/a	0.736*** (2.089)	–
ACCESIBILITY
Residential location (Core city)
Commuting area	0.081***	0.183*** (1.200)	–	–	–	−0.177** (0.838)
Distance to public transport (Walking distance)
Farther than walking distance	0.070***	0.212*** (1.236)	0.910*** (2.483)	–	0.293*** (1.340)	–
I don’t know	–	–	0.727*** (2.070)	–	–	–
Distance to the work/study place (<5 km)
Between 5 km and 20 km	0.043***	0.508*** (1.661)	–	–	0.196** (1.217)	0.304*** (1.356)
Between 20 and 50 km	0.077***	0.562*** (1.755)	–	–	0.324*** (1.383)	–
More than 50 km	0.101***	0.651*** (1.918)	−0.580* (0.560)	–	0.324*** (1.383)	0.378** (1.460)
Others	–	–	–	–	–	–

*p ≤ 0.05, ^**p ≤ 0.01, ^***p ≤ 0.001.

Variable (base category)	Number of cars per adult before COVID-19 (numerical)	Frequency of use before COVID-19 (ordinal)	Car-ownership change due to COVID-19 (No change in car ownership)		Change in frequency of car use after the start of the pandemic (About the same use)
			Bought a car	Sold a car	Decreased use	Increased use
	Coeff.	Coeff. (OR)	Coeff. (OR)	Coeff. (OR)	Coeff. (OR)	Coeff. (OR)
CITY-LEVEL VARIABLES
Population (<850,000 inhabitants)
850,000–1,400,000	–	–	–	–	–	–
1,400,000–3,000,000	–	–	–	–	0.589*** (1.802)	–
greater than 3,000,000	–	–	–	–	0.589*** (1.802)	−0.233** (0.792)
GDP per capita (<36,000 US$)
36,000–42, 000	–	–	–	–	−0.519*** (0.595)	–
42,000–62,000	0.037*	–	–	–	−0.234** (0.791)	–
greater than62,000	0.057*	–	–	–	–	–
Core city surface share
(<9%)9% − 15 %	0.044***	–	–	–	–	–
15 % − 30 %	0.079***	0.264*** (1.303)	−0.336*** (0.715)	–	0.872*** (2.392)	–
greater than30 %	–	−0.533*** (0.587)	−0.336*** (0.715)	–	0.369** (1.446)	–
Car performance (<0.23)
0.23 – 0.29	–	–	–	–	−0.391** (0.676)	–
0.29 – 0.53	0.052***	–	–	–	−0.391** (0.676)	–
greater than0.53	0.052***	0.244* (1.277)	–	–	−0.703** (0.495)	–
Public transport performance (<0.11)
0.11 – 0.18	−0.073***	–	–	–	–	–
0.18 – 0.22	−0.069**	−0.343*** (0.710)	–	–	−1.165*** (0.312)	–
greater than0.22	–	−0.487*** (0.614)	–	–	−0.897*** (0.408)	–
Bike performance (<0.41)
0.41 – 0.58	–	−0.424*** (0.655)	–	–	−0.601*** (0.548)	–
0.58 – 0.61	–	−0.277*** (0.758)	–	–	−0.500*** (0.607)	–
greater than0.61	–	−0.555*** (0.574)	–	−0.653*** (0.520)	−1.023*** (0.360)	–

*p ≤ 0.05, ^**p ≤ 0.01, ^***p ≤ 0.001

Variable (base category)	Number of cars per adult before COVID-19 (numerical)	Frequency of use before COVID-19 (ordinal)	Car-ownership change due to COVID-19 (No change in car ownership)		Change in frequency of car use after the start of the pandemic (About the same use)
			Bought a car	Sold a car	Decreased use	Increased use
	Coeff.	Coeff.	Coeff.	Coeff.	Coeff.	Coeff.
INTERACTIONS
Employed - partial TW#Car use pre-COVID: less than once a week	–	–	–	–	−0.838***	–
Employed - partial TW#Car use pre-COVID: about once a week	–	–	–	–	−1.059***	–
Employed - full TW#Car use pre-COVID: less than once a week	–	–	–	–	−0.989***	–
Employed - full TW#Car use pre-COVID: between 3 and 5 times per					−0.806***
week	–	–	–	–	0.617***	–
Medium-low-income#Car use pre-COVID: between 3 and 5 times	–	–	–	–	−0.447*	–
High-income#Car use pre-COVID: less than once a week	–	–	–	–	−0.744**	–
High-income#Car use pre-COVID: less than once a week	–	–	–	–	−0.764**	–
High-income#Car use pre-COVID: between 3 and 5 times	–	–	–	–	−0.424*	–
Bike sharing use pre-COVID: occasional#Car use pre-COVID: less than once a week	–	–	–	–	0.562***	–
Bike sharing use pre-COVID: occasional#Car use pre-COVID: about once a week	–	–	–	–	–	−0.749***
Bike sharing use pre-COVID: occasional#Car use pre-COVID: between 3 and 5 times	–	–	–	–	−0.399**	−0.619***
PT use pre-COVID: never#Car use pre-COVID: less than once a week	–	–	–	–	−1.174***	–
PT use pre-COVID: never#Car use pre-COVID: about once a week	–	–	–	–	−1.292**	–
PT use pre-COVID: never#Car use pre-COVID: between 3 and 5 times	–	–	–	–	−1.165***	−0.583**
PT use pre-COVID: less than once a week#Car use pre-COVID: never	–	–	–	–	–	1.617***
PT use pre-COVID: less than once a week#Car use pre-COVID: less than once a week	–	–	–	–	−1.168***	–
PT use pre-COVID: less than once a week#Car use pre-COVID: about once a week	–	–	–	–	−1.556***	–
PT use pre-COVID: less than once a week#Car use pre-COVID: between 3 and 5 times	–	–	–	–	−0.747***	–
Constant	0.291***	n/a	−2.101***	−4.038***	−2.234***	−0.232**
Threshold 1	n/a	1.643***	n/a	n/a	n/a	n/a
Threshold 2	n/a	2.290***	n/a	n/a	n/a	n/a
Threshold 3	n/a	2.854***	n/a	n/a	n/a	n/a
Threshold 4	n/a	3.907***	n/a	n/a	n/a	n/a