Non-pharmaceutical interventions and urban vehicle mobility in Seoul during the COVID-19 pandemic.

Non-pharmaceutical interventions to control human mobility are important in preventing COVID-19 transmission. These interventions must also help effectively control the urban mobility of vehicles, which can be a safer travel mode during the pandemic, at any time and place. However, few studies have identified the effectiveness of vehicle mobility in terms of time and place. This study demonstrates the effectiveness of non-pharmaceutical interventions at both local and national levels on intra- and inter-urban vehicle mobility by time of day in Seoul, South Korea, by applying the autoregressive integrated moving average with exogenous variables. The study found that social distancing measures at the national level were effective for intra-urban vehicle mobility, especially at night-time, but not for inter-urban mobility. Information provision with emergency text messages by cell phone was effective in reducing vehicle mobility in daytime and night-time, but not during morning peak hours. At the local level, both restrictions on late-night transit operations and stricter social distancing measures were mostly significant in reducing night-time mobility only in intra-urban areas. The study also indicates when (what time of the day), where (which area within the city), and which combination strategy could be more effective in containing urban vehicle mobility. This study recommends that restrictions on human mobility should also be extended to vehicle mobility, especially in inter-urban areas and during morning peak hours, by systematically designing diverse non-pharmaceutical interventions.

Introduction

Human mobility facilitates social interaction and drives innovation and productivity improvement, but it also facilitates the spread of infectious diseases (Morens & Fauci, 2013; Peter Sands and Dzau, 2016). In fact, it is the most important driver of airborne disease transmission (Alessandretti, 2021; Cheshmehzangi, Tang, Li, & Su, 2022). During the coronavirus disease-19 (COVID-19) pandemic, all countries implemented a variety of non-pharmaceutical interventions (NPIs), which can be categorized into four types: isolation or quarantine, mask-wearing, traffic restriction, and social distancing to control the diffusion of COVID-19 (Bo et al., 2021). Especially in the early stages, when there were no vaccines and treatments for prevention, these NPIs were almost the only measure (Oh et al., 2021; Snoeijer et al., 2021). They remain important even now that vaccines and therapeutics have been developed and distributed (Guo et al., 2021).

Many studies have proven that NPIs are effective in curbing COVID-19 diffusion (Kishore et al., 2021; Lucchini et al., 2021; Wellenius et al., 2021). Stronger NPI measures induce a greater reduction in human mobility (Czech et al., 2021; Summan & Nandi, 2021). However, even when most countries have begun distributing vaccines and treatments, the number of confirmed cases of COVID-19 is still not decreasing. This is because vaccinations lead to unexpected increase in human mobility so that the rate of COVID-19 transmission is still sustained or even increased (Deb et al., 2022; Guo et al., 2021).

Declaring the end of the pandemic might be difficult if NPIs do not sufficiently control human mobility despite the development of vaccines and therapeutics. Several studies have reported that the mode shift of travel to automobiles, which people perceive as safer than public transit, has taken place worldwide, including South Korea, during the spread of infectious diseases (Abdullah et al., 2021; Anke et al., 2021; Barbieri et al., 2021; Basu & Ferreira, 2021; Habib & Anik, 2021; Harrington & Hadjiconstantinou, 2022). Despite the huge decline in public transit mobility, the shift to private vehicles, recognized as a safer mode of transport for human mobility, could exacerbate the spread of COVID-19. Because the risk of transmission of infectious diseases in mobility may be smaller than that in public transit, the use of vehicles alone cannot reduce the risk of transmission in origins and destinations. In other words, controlling the vehicle mobility as an alternative mode should be more important in containing the COVID-19 pandemic considering public perception about mode shift for safety.

Vehicle mobility is characterized by more freedom in terms of travel behavior, such as destination choice, travel distance, and trip chaining, compared to other modes of transportation such as public transit. If vehicle mobility is not controlled during the pandemic, the transmission of COVID-19 would spread more rapidly and widely. During the initial period of the COVID-19 pandemic, the decrease in car driving during rush hour in Seoul was insignificant compared to the use of subway and bus, partly due to a mode shift from public transit to cars (Kim et al., 2021). NPIs should be applied to all modes of travel to be more effective in controlling mobility at any time and place (Poletto, Tizzoni, & Colizza, 2013; Dueñas et al., 2021). However, few studies have verified the effectiveness of NPIs on the control of vehicle mobility as well as identified whether human mobility can be controlled at any time and place.

This study demonstrates the effectiveness of various NPIs at national and local levels in Seoul, South Korea by applying the autoregressive integrated moving average with exogenous variables (ARIMAX) methodology for the daily data of vehicle mobility by the time of day and area type with an approximately two-year period. The spatial spread of infectious diseases has a bidirectional relationship with human mobility (Belik et al., 2009), and a time lag in the causal relationship of influence (Oster et al., 2020). Therefore, empirical analysis using cross-sectional data prevents the exploration of these causal relationships. Therefore, a linear regression model with ARMIA error, called the ARIMAX model, using time-series data for an extended approximately two years can identify how various NPI measures affect vehicle mobility. South Korea has taken various non-pharmaceutical measures, such as social distancing and provision of information through emergency text messages to mobile phones related to COVID-19at the national level, as well as additional measures for social distancing and traffic restriction at the local level. The study also identifies the synergistic impacts of diverse NPI measures on urban vehicle mobility by the time of day and area type in Seoul. These scientific findings are expected to contribute to global knowledge of the relationship between NPI measures and human mobility.

Literature review

NPIs and COVID-19 transmission

NPIs are public health measures to contain the transmission of infectious diseases. NPIs are the most effective interventions against infectious diseases after vaccination (OECD, 2020). They are still important measures of public health for which vaccines have been developed and distributed (Guo et al., 2021). To comprehensively examine the effectiveness of NPIs implemented in 190 countries during the early COVID-19 pandemic, Bo et al. (2021) categorized them into: isolation or quarantine, face mask-wearing, traffic restrictions, and social distancing. The Oxford COVID-19 Government Response Tracker (OxCGRT) team developed a daily-based database about the NPI measures taken by each country to identify how governments responded to control its diffusion (Gokmen et al., 2021; Hale et al., 2021). The team categorized containment and closure measures into eight types: school closing, workplace closing, canceling public events, restrictions on gatherings, closing public transport, stay-at-home requirements, restrictions on internal movement, and international travel controls. NPI measures can be applied at different government levels, such as countries, regions, municipalities, and communities (OECD, 2020).

NPIs were the most effective public intervention until vaccines and treatments for COVID-19 were developed and applied (Lewnard & Lo, 2020; OECD, 2020). Many countries have implemented NPI measures to control the diffusion of COVID-19, first reported in Wuhan, China, in December 2019. The implementation of NPIs, such as social distancing and mask-wearing, has played an important role in containing the spread of confirmed COVID-19 cases and deaths, as well as mitigating the overload of the hospital system (Qian & Jiang, 2022). Bo et al. (2021) reported that diverse NPI measures are more effective in controlling the spread of COVID-19. Haug et al. (2020) also demonstrated that a combination strategy for diverse NPIs could be more suitable for controlling the transmission of COVID-19. Bo et al. (2021) also found that the measure that resulted in the greatest reduction in the number of cases was social distancing (−42.94 %), followed by mandatory mask-wearing (−15.14 %), quarantine (−11.40 %), and restricted access (−9.26 %). Investigating the impact of eight NPI measures on confirmed cases in France, Spain, China, and South Korea, Gokmen et al. (2021) demonstrated that the most effective measure was workplace closure, followed by stay-at-home and gathering restrictions. They also found that their NPI measures at the national level were more effective than those at the regional level, although their impacts were differentiated across countries.

However, NPIs can be both direct and indirect expensive strategies for controlling infectious diseases. Reviewing the research studies containing cost data for NPIs from 1990 to 2020 (Skarp et al., 2021) estimated costs per case for isolation (US$141.18 to US$1042.68), tracing and quarantine (US$40.73 to US$93.99), and social distancing (US$33.76 to US$167.92). Conversely, the implementation of NPIs, such as social distancing and quarantine, has had a negative impact on mental health (Koch & Park, 2022; Venkatesh & Edirappuli, 2020), quality of life (Fink et al., 2022) and economic activity (Thunström et al., 2020). In addition, Gokmen et al. (2021) found that the effectiveness of NPI measures across countries could be differentiated depending on public acceptance and human behavior response. Thus, the OECD (2020) recommends that public authorities carefully implement NPI measures understood by the population so that they can be more effective against COVID-19 but less harmful to people, society, and the economy.

NPIs, human mobility and travel behavior

Human mobility is a major cause of the spread of COVID-19 (Cheshmehzangi et al., 2022; Nouvellet et al., 2021). This is because its infection mainly occurs through close human contact, and it spreads through human movement through the transport network (Alessandretti, 2021). Additionally, human mobility is closely associated with the spread of the virus in one or both directions through transportation networks (Borkowski et al., 2021). Many countries experiencing the pandemic have implemented various NPIs measures to restrict human mobility (Nouvellet et al., 2021; Snoeijer et al., 2021).

Many studies have demonstrated that NPIs are effective in reducing human mobility to control the transmission of COVID-19 (e.g., Kishore et al., 2021; Lucchini et al., 2021; Wellenius et al., 2021). Analyzing the relationship between human mobility and the spread of COVID-19 in 52 countries worldwide, Nouvellet et al. (2021) found that its transmission decreased significantly with an initial decline in mobility in 73 % of the countries. Oh et al. (2021) analyzed 34 countries and demonstrated that restrictions on human mobility caused a reduction in the number of confirmed COVID-19 cases by at least 20 % to 40 %. Abulibdeh and Mansour (2022) found that the strength of six lockdown measures, such as restrictions on workplaces, parks and outdoor activities, public transport, self-quarantine, grocery and pharmacy, and retail and recreation, were negatively related to the number of confirmed COVID-19 cases in global south countries. Similarly, examining a cross-country cluster analysis, Czech et al. (2021) demonstrated that a more stringent NPI policy resulted in a greater decline in mobility. Investigating the impact of social distancing measures on mobility and the number of confirmed cases in the United States (US) using mobility data from Google users, Wellenius et al. (2021) identified that a 10 % reduction in mobility caused a 17.5 % decrease in the number of confirmed COVID-19 cases.

The implementation of diverse NPIs to control the spread of COVID-19 can influence human mobility through changes in travel behaviors such as trip frequency, destination choice, distance, mode choice, and travel rescheduling. In particular, the implementation of restrictions on human mobility in response to the COVID-19 pandemic has caused a sudden change in the travel behaviors of people worldwide (Barbieri et al., 2021). Ehsani et al. (2021), based on a national survey of a representative sample of U.S. adults, reported that the frequency of travel decreased by 10.36 % compared to the pre-pandemic period, which was also observed in public transit, personal vehicle use, and walking. Conducting an online survey of 9394 people in 10 countries, including Australia, China, Ghana, Brazil, and the US, Barbieri et al. (2021) revealed that mobility restrictions resulted in a decrease in the frequency of use of all modes of transportation.

NPI measures on the restriction of human mobility may have had a differential effect on the choice of transport mode during the COVID-19 pandemic. Habib and Anik (2021), who analyzed public discourse on Twitter, revealed that people avoided public transportation and switched to non-public transport modes such as private cars, bicycles, and walking. Similarly, Anke et al. (2021) conducted an online survey on the impact of COVID-19 on mobility behavior changes in Germany, and found a profound impact of mobility on the mode shift from public transport to driving, walking, and cycling. Alatawi et al. (2020) reported a decline of 19 % to 50 % by country in private car use after analyzing the change in the frequency of routing requests by users of Apple Maps. Lee et al. (2020) found that the daily traffic volume for three months from January 1, 2020, for a vehicle detection system (VDS) on roads decreased by 9.7 % compared to the same period in 2019 in South Korea.

However, depending on the degree of fear of COVID-19 transmission by transport mode, substitute relationships among them may have emerged. Among all transport modes, public transit was considered to be the greatest risk of COVID-19 infection in Sicily, Italy, during the COVID-19 pandemic (Campisi et al., 2022), and even after the NPI measures were completed in Spain and Portugal (Paiva et al., 2022). Harrington and Hadjiconstantinou (2022) conducted an online survey in the United Kingdom and found that public transit users changed their commuting mode to a private car for safety reasons. Barbieri et al. (2021) found that public transit avoidance was consistently observed across countries. Büchel et al. (2022) reported that the share of public transit significantly decreased compared to private modes based on a travel survey with GPS tracking in Switzerland. Basu and Ferreira (2021) confirmed that there was a decrease in public transportation in the Boston metropolitan area during the COVID-19 pandemic owing to uncertainty in public transportation services and fear of overcrowding.

The purpose and distance of travel during the COVID-19 pandemic may have had a differentiating impact of NPI measures on the success of restrictions on human mobility. Yilmazkuday (2020), analyzing the relationship between daily inter-county travel and the spread of COVID-19 in the US, confirmed that staying in the same county decreased the number of weekly confirmed cases and deaths. Reviewing numerous studies on how infectious diseases had been diffused by travel networks from past experiences, Borkowski et al. (2021) identified that they could be spread more by human mobility behavior, such as long-distance travel and commuting.

Research gap and my contribution

Numerous empirical studies have investigated the relationship between COVID-19, human mobility, and travel behavior during the pandemic. Many studies have demonstrated that NPI measures are effective in reducing the transmission of COVID-19 by restricting human mobility (e.g., Alessandretti, 2021; Cheshmehzangi et al., 2022; Kishore et al., 2021; Lucchini et al., 2021; Oh et al., 2021; Snoeijer et al., 2021; Wellenius et al., 2021).

However, there are research gaps between the empirical findings of current studies. First, the study empirically investigates the impacts of NPI measures on daily vehicle mobility during the two-year pandemic period. The NPI measures implemented by government authorities could be more successful in controlling human mobility, varied by the mode shift from public transit to private cars. A shift from public transit to private cars occurred to avoid infection risk during the COVID-19 pandemic (Basu & Ferreira, 2021; Harrington & Hadjiconstantinou, 2022). Lee et al. (2020) reported a decoupling pattern between confirmed COVID-19 and vehicle volume on roads three months after the first case of COVID-19 was reported in South Korea in mid-January 2020. Such an increasing pattern of vehicle mobility might have played a significant role in the increasing COVID-19 cases and deaths (Borkowski et al., 2021). Nonetheless, few empirical studies have explored the impact of NPI measures on the variation in vehicle mobility.

Second, this study demonstrates how NPI measures differentially affect daily vehicle mobility by time zone and area in Seoul, South Korea. Some studies have confirmed that NPI measures could be effective in human mobility at the city level (Basu & Ferreira, 2021; Campisi et al., 2022; level). However, none of the studies have explored the impact of NPI measures on vehicle mobility by time and area in a city. Borkowski et al. (2021) reported that long-distance and commuting in human mobility plays a more decisive role in the transmission of airborne infectious diseases. This study is expected to fill the research gap of current studies by revealing the impacts of NPI measures on daily vehicle mobility by time and area in Seoul, South Korea.

Third, the study focuses on the effects of various NPI measures on daily vehicle mobility at the local and national levels. Many studies have found that NPI measures could be effective in improving human mobility at the country level (Abulibdeh & Mansour, 2022; Anke et al., 2021; Czech et al., 2021; Nouvellet et al., 2021; Oh et al., 2021; Wellenius et al., 2021; Yilmazkuday, 2020). However, none of them except Gokmen et al. (2021) demonstrate the effect of local-and national-level NPI measures on vehicle mobility simultaneously. Gokmen et al. (2021) compared the effectiveness of countrywide and regional measures and identified that the former approach could be more effective than the latter. This study explores the impacts of NPI measures at the national and local levels from the perspective of complementation rather than competitiveness.

Fourth, the study further explored the effect of the soft NPI measure, which is an emergency message service by cell phone. This measure has not been applied in most other countries. Information provision on COVID-19 using emergency text alert messages via location-based cell phones has been implemented in South Korea. Although this measure has been tested in countries such as the US and Canada (Abueg et al., 2020), South Korea has implemented its adoption since 2004. This service allows the public to quickly recognize the situation of either a disaster or an emerging risk and take countermeasures through an immediate text message. In South Korea, various pieces of information related to COVID-19 have been provided to the public through cell broadcast service technology. Local and central governments have actively applied it during the pandemic (Ju et al., 2021). The study is expected to motivate its introduction in other countries by demonstrating its impact on vehicle mobility.

Finally, the study investigates the effectiveness of various NPIs at national and local levels in Seoul, South Korea, by applying ARIMAX models for daily data for approximately two-year by the time of day and area type in Seoul. The spatial spread of infectious diseases has a bidirectional and delayed causal relationship with human mobility (Belik et al., 2009; Lee et al., 2020; Oster et al., 2020). The bidirectional feedback process can be modelled as a chain rule: COVID-19 transmission – NPIs implementation – vehicle mobility change – COVID transmission change. It is difficult to employ complex dynamic modelling to allow this process. Therefore, this study employed a moving average method for exogenous variables, such as NPI measures, to investigate the causal impacts of vehicle mobility.

Material and method

This study covered a three-year period, including one year before the onset of COVID-19 and two years after onset (study period: 1 January 2019 to 31 December 2021). The inclusion of pre-onset data is to effectively control the covariance for exogenous factors other than NPIs. This study examines the effects of NPIs on urban vehicle mobility after controlling for exogenous conditions such as day of week, holidays, temperature, weather, air pollution, population and unemployment rate in Seoul, and gasoline price.

Data on vehicle mobility

This study utilizes traffic volume data provided by the Seoul Transport Operation & Information Service (TOPIS) for urban vehicle mobility. TOPIS collects hourly traffic volume at spots on arterial roads within Seoul, 37 roads in the city centre, and 45 roads on the city boundary, as shown in Fig. 1 .

Fig. 1

Spatial distribution of traffic volume measurement points in Seoul.

Note: The purple dots indicate measurement spots on the city centre boundaries, light blue dots those within the city, and red dots those at the city limits. (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.)

This study employs daily traffic volumes by summing them up for each of the five types (within, in- and out-bound from Central Business District [CBD], and in- and out-bound from Seoul city limits). The traffic volume was converted to logarithms. This study further divided daily travel volume into three time zones: morning peak hour (07:00–09:00 h.), daytime (10:00–15:00 h.), and night-time (21:00–24:00 h.), and the entire day (24 h) for five spatial zones to identify the effects of NPI measures on urban vehicle mobility by time and place. Table 1 summarizes the statistics on the average and standard deviation of the daily traffic volume at each time zone for each spot and bound. Traffic volumes were converted into logarithms in the model, which makes interpretating the results easier because it is expressed as the effect of a percentage change in traffic volume by one unit change of NPI measures.

Table 1

Summary statistics on daily traffic by time and spot.

	(a) Within-Seoul		(b) In-bound CBD boundary		(c) Out-bound CBD boundary		(d) Out-bound Seoul boundary		(e) In-bound Seoul boundary
	Mean	Std. dev.	Mean	Std. dev.	Mean	Std. dev.	Mean	Std. dev.	Mean	Std. Dev.
Entire Day (00–24 h)	2,739,468	319,689	423,503	71,683	442,185	77,087	1,282,949	120,118	1,312,878	140,260
A.M. Peak Hour (07–09 h)	378,628	101,902	69,193	23,785	50,879	15,788	200,966	55,292	193,227	47,920
Daytime (10–15 h)	905,000	66,071	148,137	17,690	150,055	20,028	427,339	23,502	431,043	29,748
Night-time (21–24 h)	309,883	47,660	41,941	8742	51,214	11,650	126,406	16,529	141,471	20,286

Fig. 2 displays the daily trend of traffic volume by time zone as well as the entire day, which is log-transformed by summing the traffic volumes at all spots. Compared to 2019, the traffic volume showed a relatively decreasing trend from January 2020, when the COVID-19 outbreak occurred, and the largest decrease occurred in January 2021. The change in traffic volume by time zone indicates that the smallest decrease is in the morning peak hours and is larger at night.

Fig. 2

Daily variation on log-transformed total traffic volume by time of day.

Data on non-pharmaceutical interventions

South Korea have implemented social distancing measures such as school closing, workplace closing, canceling public events, restriction on gathering size, stay-at-home requirement, and restriction on internal movement. These measures are highly correlated because they were implemented together (Snoeijer et al., 2021). Many countries have also implemented various NPIs, including social distancing, while repeatedly strengthening and mitigating their steps on a case-by-case basis. The OxCGRT has built a global panel database of the policies on the COVID-19 pandemic, updating it daily and making it public (Hale et al., 2021). This database has been applied to research such as an analysis of the relationship between a non-pharmaceutical intervention measure and the increased rate of COVID-19 (Gokmen et al., 2021; Huy et al., 2022) and economic growth (König & Winkler, 2021). Among these policies, social distancing measures include school closing, workplace closing, canceling public events, restrictions on gathering size, closing public transport, restrictions on internal movements, and restrictions on international movement. They were graded according to their level. This study utilized a database on social distancing measures for South Korea at the national level that OxCGRT provides. Table 3 shows that the average social distancing index is 64.526, and the maximum is 160. Fig. 3 indicates that the highest level of social distancing in South Korea was for 10 days from 6 to 17 April 2020, in the early stages of the pandemic.

Table 3

Model statistics on alternative models by time and place.

Type	Time of day	Model (p, d, q)	Standard deviation of error term	Log-likelihood	AIC	BIC	Q statistic for Ljung–Box Test	P-value
Within-Seoul	Entire Day	LM	0.047	1799.8	−3551.6	−3431.6	208.29	0
		ARIMA (4, 1, 4)	0.085	1148.9	−2279.7	−2234.7	25.886	0
		ARIMAX (4, 0, 1)	0.04	1982.9	−3907.7	−3762.7	0.134	0.715
	A.M. Peak Hour	LM	0.094	1052.1	−2056.1	−1936.1	16.929	0
		ARIMA (5, 1, 5)	0.19	261.3	−500.5	−445.5	1.994	0.158
		ARIMAX (4, 0, 1)	0.091	1086.7	−2115.3	−1970.3	0.194	0.66
	Daytime	LM	0.042	1928.1	−3808.3	−3688.3	200.136	0
		ARIMA (5, 1, 4)	0.047	1785.8	−3551.6	−3501.6	13.902	0
		ARIMAX (5, 0, 0)	0.036	2094.4	−4130.8	−3985.8	0.705	0.401
	Nighttime	LM	0.065	1450.1	−2852.3	−2732.3	256.726	0
		ARIMA (0,1,0)	0.141	594.4	−1186.9	−1181.9	18.172	0
		ARIMAX (4, 0, 1)	0.053	1679	−3299.9	−3154.9	0.032	0.858
In-bound CBD boundary	Entire Day	LM	0.047	1799.8	−3551.6	−3431.6	208.29	0
		ARIMA (4, 1, 4)	0.085	1148.9	−2279.7	−2234.7	25.886	0
		ARIMAX (2,0,1)	0.04	1982.9	−3907.7	−3762.7	0.134	0.715
	A.M. Peak Hour	LM	0.094	1052.1	−2056.1	−1936.1	16.929	0
		ARIMA (5, 1, 5)	0.19	261.3	−500.5	−445.5	1.994	0.158
		ARIMAX (4, 0, 1)	0.091	1086.7	−2115.3	−1970.3	0.194	0.66
	Daytime	LM	0.042	1928.1	−3808.3	−3688.3	200.136	0
		ARIMA (5, 1, 4)	0.047	1785.8	−3551.6	−3501.6	13.902	0
		ARIMAX (3,0,2)	0.036	2094.4	−4130.8	−3985.8	0.705	0.401
	Nighttime	LM	0.065	1450.1	−2852.3	−2732.3	256.726	0
		ARIMA (0, 1, 0)	0.141	594.4	−1186.9	−1181.9	18.172	0
		ARIMAX (2,0,3)	0.053	1679	−3299.9	−3154.9	0.032	0.858
Out-bound CBD boundary	Entire Day	LM	0.063	1483.4	−2918.8	−2798.7	205.082	0
		ARIMA (4, 1, 4)	0.113	831.1	−1644.2	−1599.2	0.099	0.753
		ARIMAX (2, 0, 3)	0.054	1654.1	−3250.2	−3105.2	0.061	0.806
	A.M. Peak Hour	LM	0.105	927.5	−1807	−1686.9	61.531	0
		ARIMA (5, 1, 5)	0.209	153	−284	−229	2.37	0.124
		ARIMAX (4, 0, 1)	0.099	991.2	−1924.4	−1779.4	0.123	0.726
	Daytime	LM	0.061	1526.1	−3004.2	−2884.2	169.318	0
		ARIMA (5, 1, 4)	0.075	1279.5	−2539	−2489	8.867	0.003
		ARIMAX (2, 0, 3)	0.053	1681.4	−3308.9	−3173.8	0	0.998
	Nighttime	LM	0.081	1213.4	−2378.8	−2258.8	191.137	0
		ARIMA (0, 1, 0)	0.205	180.5	−359	−354	14.579	0
		ARIMAX (5, 0, 0)	0.07	1378.1	−2698.2	−2553.1	0.073	0.787
In-bound Seoul boundary	Entire Day	LM	0.048	1780	−3511.9	−3391.9	384.556	0
		ARIMA (4, 1, 4)	0.069	1368.9	−2719.7	−2674.7	26.749	0
		ARIMAX (4, 0, 1)	0.036	2093.3	−4132.7	−3997.6	0.001	0.981
	A.M. Peak Hour	LM	0.103	951.3	−1854.5	−1734.5	16.447	0
		ARIMA (5, 1, 5)	0.196	227.4	−432.9	−377.9	2.435	0.119
		ARIMAX (4, 0, 1)	0.099	993	−1928	−1783	0.06	0.806
	Daytime	LM	0.044	1882.8	−3717.6	−3597.6	433.568	0
		ARIMA (5, 1, 4)	0.036	2076.1	−4132.2	−4082.2	1.762	0.184
		ARIMAX (5, 0, 0)	0.031	2257.7	−4457.5	−4312.4	0	0.992
	Nighttime	LM	0.078	1255.1	−2462.2	−2342.2	346.053	0
		ARIMA (0, 1, 0)	0.105	917.4	−1832.8	−1827.8	38.099	0
		ARIMAX (4, 0, 1)	0.06	1537.2	−3016.4	−2871.4	0	0.985
Out-bound Seoul boundary	Entire Day	LM	0.05	1753.7	−3459.4	−3339.4	344.64	0
		ARIMA (4, 1, 4)	0.091	1070.3	−2122.6	−2077.6	0.139	0.709
		ARIMAX (4, 0, 1)	0.037	2063.5	−4068.9	−3923.9	0.007	0.933
	A.M. Peak Hour	LM	0.096	1031.5	−2014.9	−1894.9	67.291	0
		ARIMA (5, 1, 5)	0.173	361.6	−701.3	−646.3	1.242	0.265
		ARIMAX (4, 0, 1)	0.089	1116.1	−2174.2	−2029.2	0.31	0.578
	Daytime	LM	0.043	1897.2	−3746.5	−3626.4	331.496	0
		ARIMA (5, 1, 4)	0.041	1939.5	−3859.1	−3809.1	13.624	0
		ARIMAX (1,0,2)	0.032	2218.5	−4379	−4234	0.491	0.484
	Nighttime	LM	0.073	1333.6	−2619.1	−2499.1	341.964	0
		ARIMA (0, 1, 0)	0.145	561.2	−1120.3	−1115.3	14.305	0
		ARIMAX (3, 0, 2)	0.056	1625	−3191.9	−3046.9	0.435	0.509

Note: LM = Linear Model, ARIMA = AutoRegressive Integrated Moving Average, and ARIMAX = AutoRegressive Integrated Moving Average with eXogenous variables.

Fig. 3

Daily trends on social distancing, emergency text message information, and the number of newly confirmed cases.

Note: The blue line represents the scores of social distancing measures at the national level, the gold line represents the number of daily emergency disaster text messages, and the brown line represents the number of new confirmed cases per day. (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.)

The measure that can be unique to South Korea is information provision on COVID-19 using emergency text alert messages by location-based cell phones, even though this measure has been tested in countries such as the United States and Canada (Abueg et al., 2020). Public authorities have actively used it to prevent the spread and prevention of COVID-19. Since December 2004, South Korea has established a service system that provides immediate text messages for information provision during emergency disasters. This service is provided by the Ministry of Public Administration and Security, local governments, and the Korea Meteorological Administration. It is the Cell Broadcasting Service (CBS), which transmits disaster messages through a mobile communication base station during disaster or danger. Thus, a text message was sent to people with all types of cell phones in that area at that time. As of 2020, the penetration rate of cell phones for people over 18 years in South Korea is 100 % (Kim et al., 2021). Thereby, Ju et al. (2021) reported that delivering emergency disaster and safety messages using cell phones is more effective than other media sources. During the COVID-19 pandemic, the number of emergency text messages was highest since 2004. These data were acquired by downloading data from the National Disaster Safety Portal (http://www.safekorea.go.kr) using an API key. In this study, only texts sent to the Seoul area were selected through text mining.

This study employed two indicators: the number of messages sent per day and the number of newly confirmed cases. Disaster messages sent to cell phones are divided into three types: an emergency disaster used when a war situation occurs; an emergency disaster used when natural disasters such as earthquakes and typhoons occur; and a safety guide used when special weather warnings such as heatwaves, yellow dust, and safety precautions are required (Kim et al., 2021). Emergency text services related to COVID-19 have provided information on emergencies and safety to the public.

It is not possible to determine how many of these sent messages have been received by one person. Therefore, this study employs the number of confirmed cases per day and the number of text messages sent by the central government, quarantine authorities, Seoul, and 24 local governments as indicators. The most frequently provided information was the number of newly confirmed COVID-19 cases in each municipality. The higher this number, the more likely it is that more residents will be able to self-restrict movement, if possible. Therefore, this study expected that the provision of information could affect vehicle mobility. Another indicator of text messaging services is how many text messages were transmitted daily by various public authorities to the public. This information mainly includes the places that the confirmed patient has visited, and prevention and hygiene rules. The former specifies the location and time of the COVID-19 outbreak in a specific place so that it could affect vehicle mobility, preventing vehicles from visiting the place. In addition, the provision of information on the guidance of prevention and hygiene rules could be considered a warning about the risk of COVID-19. In this regard, this study selected the number of text messages transmitted per day as an indicator of NPI measures.

As the response of people in changing their travel behaviors using a vehicle may occur with a time lag, the moving average of the past seven days was calculated for two indicators. Zhang et al. (2021) demonstrated that confirmed COVID-19 cases were positively associated with an approximately 7-day earlier human mobility. Oster et al. (2020) also used a 7-day moving average, which can deviate by the day of the week, to identify transmission dynamics by age group in COVID-19 hotspot counties in the US. Yilmazkuday (2020) revealed that staying in the same county decreases the number of weekly confirmed cases and deaths. Based on these, the study employed the values of 7-day moving averages for the two indicators in the models. A change in travel behavior can be decided a few days later, as well as immediately when text messages are received. Table 3 shows that the average number of emergency text messages sent per day is 8.965, with a maximum of 44.286, while the average number of new confirmed cases per day in Seoul is 8.965, with a maximum of 44.286.

In addition to non-pharmaceutical measures such as social distancing and emergency message services at the national level, two additional measures at the local level were implemented to contain COVID-19 diffusion. One was a measure to reduce the operation of public transit managed by the Seoul metropolitan city, such as buses and subways, by up to 20 % during late night hours after 10 pm. This measure was temporarily implemented from 1 July to 24 October 2021. The other was a stricter social distancing measure than the national level (16 August–13 September 2020; 24 November 2020–14 July 2021) in the Seoul metropolitan area. These regionally differentiated stricter measures may also have affected the urban vehicle mobility. Therefore, these measures were applied to the analysis by treating the corresponding days as a dummy variable.

Data on control variables

Table 3 summarizes the statistics for the mean, standard deviation, and minimum and maximum values of the exogenous variables. Here, except for non-pharmaceutical intervention measures, the other variables were the control variables. These variables are known to affect travel demand (Akin et al., 2011; Kwak et al., 2017; Sung, 2017; Wets et al., 2010). In this study, factors such as days of the week, public holidays, traditional holidays, highest daily temperature, adverse weather factors such as rain and snow, and air quality factors such as fine and ultrafine dust were used as control factors affecting traffic volume in the models. In addition, exogenous factors, such as the monthly unemployment rate and average price per liter of gasoline, were also incorporated into the model to control them. However, there are missing values of traffic volume measured by detectors such as loops, images, and geomagnetic. Therefore, the study imputed the average traffic volume at the same hour on the same day of the week within the same month at the same point for the hourly time where the missing value occurred. However, because the substitution of the average of these missing values is not perfect, a bias in traffic volume prediction may occur. Therefore, this study employed the number of missing values to improve the accuracy of the prediction.

Methodology and model selection

This study applies autoregressive integrated moving average with exogenous variables (ARIMAX) modelling to examines NPIs' effectiveness on urban vehicle mobility in Seoul. This model is a statistical model that analyses changes over time along with exogenous variables. It was also employed in the prediction of traffic volume (Nagy & Simon, 2018), the effect of weather on infectious diseases (Akin et al., 2011), the relationship between traffic volume and infectious diseases (Shi & Fang, 2020), and the resilience of the air transportation industry due to an infectious disease pandemic (Shi & Fang, 2020). This ARIMAX model is applied for the combination of four time zones (entire day, morning peak hour, daytime, and night-time) and five spatial zones (within-city, in- and out-bound from CBD, and in- and out-bound from city limits).

The most important factor in considering the methodology of this study is the autocorrelation of daily time series data, while non-constant aperiodic factors such as holidays, as well as NPI measures, and periodic factors that take day and month fluctuations into account are input as explanatory variables (Anggraeni et al., 2015; Arunraj & Ahrens, 2015). The alternative methods reviewed in this study are the multiple linear regression model, univariate autoregressive integrated moving average (ARIMA) model, and the ARIMAX model. The ordinary least squares linear regression model, which inputs periodic and aperiodic factors as explanatory variables, has a disadvantage in that it is difficult to control the time series autocorrelation. The ARIMA model, which can effectively control time-series autocorrelation, controls trends and periodic fluctuations; however, it does not consider aperiodic and exogenous factors. The ARIMAX model considers both periodic and aperiodic variability, and the effect of exogenous intervention. This model is suitable for the occurrence of fluctuations due to events that do not have the same effect and timing, such as periodic repetition and trends.

The ARIMAX model in the study, also called a regression model with the ARIMA errors, combines two statistical models, namely, multiple linear regression model and ARIMA, into a single regression model for time series data (Hyndman & Athanasopoulos, 2021), with the equation:

Here, the response variable log(y ) is the daily vehicle mobility converted to logarithms at time t, β 0 is a constant term, and β is the regression coefficient of an exogenous variable x at time t. This can be described as a multiple linear regression model with exogenous variables. Finally, η is an error term at time t, which can be estimated using the autoregressive moving average (ARMA) model. If differencing (d) is required, then the constant term disappears. Error η , can also be formulated as follows:

Here, the error η , is composed of the autoregression ϕ of order p, and the moving average θ of order q and z at time t, which has a zero mean and an independent and identically distributed white noise process.

Based on the regression model with ARIMA errors, this study constructed 20 ARIMAX models with five places and four time zones. In each ARIMAX model, the outcome variable is the logarithmic traffic volume in a specific time zone in a specific area, as presented in Table 1. Table 2 presents summary statistics for the NPI indicators and input variables controlled in the models. Control variables input to the models are dummy variables on the day of the week and holiday dummy variables, weather and atmospheric environment, unemployment rate, gasoline price, etc. The NPI indicators at the national level are the social distancing index and the number of text messages and confirmed COVID-19 cases on the 7-day moving average. The NPI indicators at the local level are dummy variables for the measures of the 20 % reduction in late-night operations in Seoul and the stronger distance in the Seoul metropolitan area.

Table 2

Summary statistics on and variation inflation factors of exogenous variables.

Variables		Mean	Std. dev.	Min	Max	VIF
Control variables	Mon (=1, Sun = 0)	0.142	0.349	0	1	1.72
	Tue (=1, Sun = 0)	0.143	0.35	0	1	1.73
	Wed (=1, Sun = 0)	0.145	0.352	0	1	1.73
	Thu (=1, Sun = 0)	0.143	0.35	0	1	1.73
	Fri (=1, Sun = 0)	0.143	0.35	0	1	1.74
	Sat (=1, Sun = 0)	0.142	0.349	0	1	1.72
	Holiday (yes =1, no = 0)	0.026	0.16	0	1	1.01
	Korean Traditional Day (yes =1, no = 0)	0.022	0.146	0	1	1.04
	Sandwich Day (yes =1, no = 0)	0.004	0.06	0	1	1.03
	High Temperature (°C)	18.317	10.15	−10.7	36.8	1.48
	Rainy Day (yes =1, no = 0)	0.293	0.455	0	1	1.12
	Snowy Day (yes =1, no = 0)	0.022	0.146	0	1	1.17
	PM 10 (>81 = 1, ≤81 = 0)	0.05	0.218	0	1	1.29
	PM 2.5 (>36 = 1, ≤36 = 0)	0.131	0.338	0	1	1.3
	Unemployment rate	4.587	0.807	3.3	6.5	1.81
	Gas Price (KWR, Log)	7.356	0.074	7.2	7.544	2.98
	Counts of NAs on Traffic Measurement Points	40.84	33.786	0	150	1.7
NPI measures at the national level	Social Distancing Index Score	64.526	51.973	0	160	3.07
	No. Massages (MA-7)	8.965	11.413	0	44.286	4.24
	No. New patients (MA-7)	201.957	437.007	0	2720.571	1.77
NPI measures at the local level	Restriction on Transit Operation in Seoul	0.097	0.297	0	1	1.78
	Enhanced Intervention in Capital Region	0.238	0.426	0	1	3.13

Note: VIF = variance inflation factor.

This study reviewed and tested all three alternative models: multiple linear regression models, ARIMA models, and ARIMAX models for each time zone in an area. In the ARIMA and ARIMAX models, the study estimated the appropriate AR(p) and MA(q) with differencing (d), if necessary, to choose the best model. The parentheses in the model column in Table 3 represent the appropriate orders of p, d, and q for both the ARIMA and ARIMAX models. Finally, this study employed ARIMAX models as the most suitable model for this study. For an appropriate model selection procedure, this study considered the standard deviation, log-likelihood, Akaike's Information Criteria (AIC), and Bayesian Information (BIC) of the residuals of all models as model statistics, and the Ljung–Box Q statistic as an autocorrelation test statistic of the model residuals. Table 3 summarizes the test statistics of the three alternative models for each of the five places and the four time zones. The smaller the standard deviation of the residual, the AIC, and the BIC statistic, the larger the log-likelihood statistic and the more suitable the model. All models indicate that the ARIMAX model is more suitable for predicting urban vehicle mobility than multiple linear regression and ARIMA models. For example, the standard deviation of the error term of the ARIMAX model for the intra-urban traffic volume for the whole day was 0.04, the AIC was −3907.7, and the BIC was −3762.7, which is the lowest value compared to other models, and the log-likelihood was the highest at 1982.9.

This study diagnosed changes in urban vehicle mobility using periodic and aperiodic exogenous control variables and NPI indicators. The model employed in this study is a regression model with ARIMA error. Therefore, in the same way as the diagnosis of the linear regression model, it is necessary to diagnose multicollinearity as to whether there is a strong correlation between these explanatory variables. This study employed the variance inflation factor (VIF) statistic. If the VIF is greater than five or, sometimes, ten, serious multicollinearity among the exogenous variables in the model can be detected (Ihueze & Onwurah, 2018; Kawakita & Takahashi, 2022). In Table 2, all of these values are less than five, indicating that the degree of multicollinearity between the explanatory variables is insignificant. If the multicollinearity among the exogenous variables was severe in the model, it indicated that the combination of NPI measures that the study employed would have synergistic effects on controlling vehicle mobility.

Results

Fig. 4 displays the results of ARIMAX modelling for social distancing (A), number of daily emergency messages (B), number of new confirmed cases (C), reduction of public transit operation in Seoul (D), and stricter social distancing measures in the Seoul metropolitan area (E). The dots represent the regression coefficients, and the whisk represents the 95 % confidence interval. The grey dashed vertical line indicates that the regression coefficient is zero. If the confidence interval of the regression coefficient spans this line, it means that the regression coefficient is not different from zero at the significance level of 0.05.

Fig. 4

ARIMAX modelling results on the impacts of the NPI measures on vehicle mobility.

The effect of the index of social distancing measures at the national level on urban vehicle mobility is summarized in Fig. 4(A). All the regression coefficients have negative values, except for the results of the morning peak hour, indicating that they can always have a reduction effect to some extent and points. However, when diagnosing whether it was statistically significant with 95 % confidence interval, the effects of reducing vehicle mobility spots both within the city and in- and out-bound from city centres were mostly significant. However, the reduction impacts of vehicle mobility on both bounds of the city limits are not statistically significant.

Considering the magnitude of the regression coefficient, the effect on vehicle mobility for the out-bound city centre was the greatest, followed by those of the in-bound city centre and within the city. Table 1 presents the quantified effectiveness with only statistically significant variables by ARIMAX modelling at a 90 % confidence interval, where the average and maximum values of the corresponding indicators are multiplied by the regression coefficients. Table 1 indicates that the measures of social distancing induced a reduction in vehicle mobility of 1.3 % to 5.2 %, on average, and at most 12.8 %. In particular, the effect of reducing vehicle mobility for the out-bound city centre at night was relatively large.

To measure the effects of emergency text messages, this study employed two indicators—number of daily text messages transmitted to citizens in Seoul based on the cell phone location (Fig. 4B) and number of newly confirmed cases in Seoul per day (Fig. 4C). Since people's responses to travel decisions based on text messages may occur with time lag, the study applied the moving average of the past seven days to these indicators. Fig. 4(B) indicates that the effect of reducing vehicle mobility on days with many COVID-19-related emergency text messages was statistically significant mainly during daytime and night-time, but not for the morning peak hour. For each spot, the effect was greatest mainly in the direction of outflow from the downtown area.

Table 4 shows that the vehicle mobility reduction effect of the number of text messages sent over the past seven days on average was, on average, 1.0 % to 3.9 %, and at most 19.0 % at night-time. The most frequently and continuously provided information among COVID-19-related emergency text messages is the number of new confirmed cases per day. Fig. 4(C) also indicates that information provision was not statistically significant in suppressing vehicle mobility by time of day for all spots.

Table 4

Effectiveness size of NPI measures on vehicle mobility by time and place.

NPI measures	Variable	Entire day		AM peak		Daytime		Nighttime
		Mean	Max	Mean	Max	Mean	Max	Mean	Max
Social Distancing Index	Within-Seoul	−2.6 %	−6.4 %	−2.6 %	−6.4 %	−1.3 %	−3.2 %	−3.2 %	−8.0 %
	In-bound CBD boundary	−3.2 %	−8.0 %	−3.9 %	−9.6 %	−2.6 %	−6.4 %	−3.2 %	−8.0 %
	Out-bound CBD boundary	−3.9 %	−9.6 %	−5.2 %	−12.8 %	−3.9 %	−9.6 %	−5.2 %	−12.8 %
	In-bound Seoul boundary
	Out-bound Seoul boundary							−1.9 %	−4.8 %
No. Messages (MA-7)	Within-Seoul					−1.3 %	−6.2 %	−2.2 %	−10.6 %
	In-bound CBD boundary	−1.0 %	−4.9 %			−1.5 %	−7.5 %	−2.8 %	−13.7 %
	Out-bound CBD boundary	−1.1 %	−5.3 %			−1.4 %	−7.1 %	−2.7 %	−13.3 %
	In-bound Seoul boundary	−1.3 %	−6.6 %			−1.2 %	−5.8 %	−3.9 %	−19.0 %
	Out-bound Seoul boundary							−1.8 %	−8.9 %
Restriction on Transit Operation in Seoul	Within-Seoul							−0.4 %	−3.9 %
	In-bound CBD boundary
	Out-bound CBD							−0.6 %	−5.9 %
	In-bound Seoul boundary							−0.5 %	−5.3 %
	Out-bound Seoul boundary
Enhanced Intervention in Capital Region	Within-Seoul							−0.9 %	−3.6 %
	In-bound CBD boundary	−0.7 %	−2.9 %					−0.9 %	−3.6 %
	Out-bound CBD boundary	−0.7 %	−3.0 %	−1.2 %	−5.2 %			−1.2 %	−5.0 %
	In-bound Seoul boundary
	Out-bound Seoul boundary

Note: The percent values are calculated by multiplying the regression coefficient by the average and maximum values of the NPI metrics.

The Seoul Metropolitan City Government implemented a 20 % reduction in public transport operations, such as buses and subways, after 10 p.m. from 1 July to 24 October 2021. Fig. 4(D) shows that this measure mostly reduced vehicle mobility at night, especially within the city, out-bound city centre, and in-bound city boundaries. It is interpreted that this measure reduced the night-time vehicle traffic volume by raising awareness among citizens about COVID-19 diffusion, rather than causing them to use vehicles due to reduced availability of public transport. Its effectiveness is, on average, 0.4 % to 0.6 %, and up to 8.9 %, as shown in Table 1.

Measures that were more stringent than the nationwide level of distancing were mainly carried out in the early stage of the rapid increase in the number of confirmed cases in the Seoul metropolitan area. There are two periods from 16 August to 13 September 2020, and from 24 November 2020 to 14 July 2021. Fig. 4(E) indicates that there was a statistically significant effect only on reducing vehicle mobility to the out-bound city centre during the morning peak and night-time during this period and was not statistically significant at other times for any spot. The magnitude of the effect is, on average, 0.7 % to 0.9 %, and at most 5.0 % at night-time.

Discussion

Non-pharmacological interventions to contain the spread of the coronavirus by controlling human mobility were almost the only means in the early stages when no vaccines or treatments had been developed (OECD, 2020). Many countries implemented social distancing measures such as school closures, telecommuting, gathering bans, and even complete restriction on movement to control human mobility. These measures were proven effective in curbing the rapid spread of COVID-19 (Liu et al., 2022). However, it should have been applied to all modes of travel at any time in all places. Few empirical studies have investigated urban vehicle mobility based on time and place. Therefore, this study identified how effectively various NPIs played a role in controlling the mobility of vehicles by time, place, and direction during the pandemic period of about two years after the onset of COVID-19 in Seoul, South Korea. The study results conclusively prove that the NPI measures did not effectively control urban vehicle mobility; therefore, the issue needed to be discussed.

National-level social distancing measures were effective only in reducing intra-urban vehicular mobility, especially in downtown areas. Moreover, these measures succeeded in controlling vehicle mobility only during off-peak hours, especially at night. Since long-distance travel, such as commuting, is a major cause of infectious disease diffusion (Borkowski et al., 2021; Dueñas et al., 2021; Pullano et al., 2020), these results raise doubts about the effectiveness of social distancing in controlling vehicle mobility. People who perceive private cars to be safer than public transit simply switched to cars, not only for peak hour commuting but also for long-distance travel for business, shopping, and leisure, to avoid infection risks (Anke et al., 2021; Barbieri et al., 2021; Basu & Ferreira, 2021; Jang et al., 2020; Habib & Anik, 2021; Harrington & Hadjiconstantinou, 2022). People have changed their mode of transport from public transit to private cars for long-distance travel (Abdullah et al., 2021). Thus, the study results proved that social distancing measures were not effective enough to control vehicle mobility. This suggests that policy interventions should take more complex risk-avoidance behaviors, such as travel mode shift, into consideration so that the NPI of social distancing can work effectively.

In addition to social distancing measures at the national level, two additional measures at the local level played a significant role in further reducing urban vehicle mobility. First, the 20 % restriction on public transit operation after 10 pm reduced vehicle mobility during night-time. This implies that the measure signalized people who use a private car as well as public transit riders do not travel at the time. South Korea has implemented various social distancing measures during the COVID-19 pandemic without strongly restricting public transit. Whereas, some cities, such as China and Italy, have implemented shutdown measures of public transit to control human mobility (Ren, 2020). Public transit is a crucial mode of travel for the socially disadvantaged, such as children, youths, the elderly, and the low-income class, as well as for workers who cannot work from home during the COVID-19 pandemic (Dueñas et al., 2021; Rojas-Rueda & Morales-Zamora, 2021). Therefore, the shutdown of public transit to contain the spread of COVID-19 may further exacerbate social inequality at this time. Therefore, NPI measures that limit the mobility of public transportation should be the last measure.

As a second measure at the local level, the implementation of stronger social distancing measures in the Seoul metropolitan area were effective in controlling the mobility of vehicles in the city centre and within the city. These results show that the influence of NPI measures can be limited to specific human mobility by time and place (Hu et al., 2021). Therefore, both spatially and temporally differentiated responses of people to social distancing measures should be considered to have sufficient impact on human mobility, including vehicle mobility.

In addition to the social distancing measures implemented by most countries, South Korea actively utilized the emergency disaster text transmission service through cell phones to prevent COVID-19 diffusion for a year and a half after the outbreak began. The service, launched in 2004, significantly increased the number of messages sent to people with a cell phone since the outbreak of COVID-19 in 2020 (Oh et al., 2021). The effect of receiving and reading emergency disaster text messages and obtaining information on individual preventive actions in the early stages of COVID-19 was significant (Lee & You, 2021). There was a significant decrease in the floating population, which is a location-based population measured using cell phones, at the COVID-19 mass outbreak site, which suppressed the rapid spread of the infectious disease (Choi et al., 2021). The analysis results of this study also demonstrate that this measure is relatively more effective in controlling inter-urban vehicle mobility, which is not controlled adequately by social distancing. In this regard, the dissemination of COVID-19 information through emergency text messages can be considered an NPI measure that can supplement traditional social distancing measures.

Fig. 3 displays the daily trend in the level of social distancing measures at the national level, the number of emergency disaster text messages, and the number of new confirmed cases in Seoul. The level of social distancing measures at the national level was the highest at the beginning of the COVID-19 pandemic, and thereafter, easing and strengthening were repeated and more relaxed by the end of 2021. In addition, emergency text message services related to COVID-19 were sent immediately and frequently as needed by the central government and local governments from the beginning of the pandemic. However, as people grew tired of reading so many messages, from 6 April 2021, the central government changed the guidelines. Local governments were asked to refrain from sending messages unless urgent and provide information on the number of confirmed cases once a day. As a result, as shown in Fig. 3, the text message service has significantly decreased from that point on.

The study found that the provision of information on the number of newly confirmed cases did not significantly reduce vehicle mobility at any time or at any point. This indicates that the once-a-day messages, sent based on the revised guidelines, probably made no contribution to containing COVID-19 diffusion. The restriction on the frequency and content of emergency text messages along with more relaxed social distancing measures might have resulted in failure to contain the rapid spread of COVID-19, which has been in progress since July 2021. The more highly contagious mutant virus, Omicron, was first reported to the WHO in South Africa on 24 November 2021, and South Korea confirmed its first case on 1 December 2021. In other words, there is a temporal gap between the spread of Omicron mutations in South Korea and the rapid spread of COVID-19 after July 2021. One of the main causes of the rapid spread of COVID-19 currently was the easing of NPI measures such as social distancing and COVID-19-related emergency text messages.

The study identified that the multicollinearity among exogenous variables was severe, with the VIF statistics in Table 2. This indicates that diverse combination strategies of NPI measures can have enhanced synergistic impacts on controlling urban vehicle mobility. Fig. 5 shows the summing effects of the average and maximum measures of the four NPI measures on controlling urban vehicle mobility by time and place. In Fig. 5, the most effective reduction in vehicle mobility by the combination of the four NPI measures is the outflow vehicle mobility in the CBD at night, indicating that its mobility has decreased by 9.70 % on average and by up to 37 %. The results show how and to what extent vehicle mobility can be controlled when and where through various combinations of the four NPI measures. Haug et al. (2020) also demonstrated that a combination of diverse NPI measures could have a greater impact on the transmission of COVID-19. The study also indicates when (what time of the day), where (which city area), and which combination strategy could be more effective in containing urban vehicle mobility.

Fig. 5

Synergy impacts of the four NPI measures on vehicle mobility.

This study empirically identified the effectiveness of diverse NPI measures implemented by governments in South Korea on urban vehicle mobility by time and place. However, its effectiveness in restricting human mobility during COVID-19 may be differentiated depending on public compliance and acceptance by country, which can have different social norms and cultures. For example, it was demonstrated that differences in public compliance with face mask-wearing in the US and South Korea during the COVID-19 pandemic stem from differences in social norms in those countries (Chang et al., 2021; Kang et al., 2021). Therefore, it is important to understand the differences in cultures and norms in each country and develop a way to communicate them more effectively to promote the effectiveness of NPI measures. In addition, Bicchieri et al. (2021), who explored public compliance with NPI measures in nine countries, including China, Colombia, Germany, Italy, Mexico, South Korea, Spain, the United Kingdom, and the US of America, argued that this could be further promoted through the improvement of trust in science and government.

This study is valuable because it demonstrates how various NPI measures affect urban vehicle mobility by time and place to derive policy implications from a global perspective. However, it is also worth mentioning that the limitations of this study were derived from the analysis process. First, the study measured the effect of the emergency text message service related to COVID-19 using only the information of the text message sent by the supplier, that is, the government. It would be more valuable to empirically identify the effect of this service based on the quantity and quality of text messages, that is, the frequency of COVID-19 per person and its contents. A more effective strategy for the service measure against COVID-19 can be formulated from further study examining in more detail how many people receive this service, and how behavior in human mobility, including vehicle movement, could be changed due to the frequencies and contents of the text messages from the consumer perspective, rather than the supplier.

Second, the study may have a limitation in a temporal change of urban vehicle mobility resulting from NPI measures along with an inability to grasp the effect of transportation mode shift due to the fear of contracting COVID-19. Many studies have identified the greatest decline in the use of public transit and its transition to non-public transportation such as walking, bicycling, and private cars. The study demonstrated that diverse NPI measures did not significantly affect reducing inter-urban vehicle mobility during A.M. peak hours. This result might be due to the effect of the mode shift from public transit to private cars. Therefore, future studies should investigate this moderating effect on vehicle mobility.

Finally, the study is limited as it does not explore whether demographic socioeconomic factors such as sex, age, income, and occupation have differentially affected urban vehicle mobility by the NPI measures. The data on vehicle mobility in this study were the number of passing vehicles on roads collected through vehicle detectors. Therefore, these data cannot indicate who used which type of vehicle. Thus, it is necessary to conduct a survey on travel behavior for vehicle use and link it with the analysis results of this study in the near future. Further study also needs to investigate what type of vehicle mobility could be more varied by NPI measures during the pandemic. It is expected that a more specific NPI strategy will be designed if it is possible to identify which demographic and socioeconomic groups are more effective in controlling urban vehicle mobility by time and area.

Conclusions

Although this study targeted only Seoul, South Korea, the implications based on the results provide opportunities for the combined application of diverse NPI measures on urban vehicle mobility by time of day in a certain area in cities globally. The study found that diverse NPI measures, such as social distancing at the national level, dissemination of COVID-19 information through emergency text messages, reducing public transport operations, and imposing stricter social distancing measures at the local level, have been effectively implemented to control urban vehicle mobility, especially at the beginning of the COVID-19 pandemic. It can be said that if these measures are comprehensively and systematically designed, and policy interventions are made, they would be more effective in suppressing COVID-19 diffusion caused by human mobility.

Social distancing measures may not be sustainable in the long term, as they incur huge socio-economic costs and negative effects such as social inequality (Chang et al., 2021). This can be especially true when vaccines are developed for the prevention of infectious diseases, and therapeutic agents are disseminated. However, alternative NPIs can be implemented without incurring significant socio-economic costs. One such alternative is to actively use information and communication technology to provide information about COVID-19 to citizens immediately, so that citizens can voluntarily suppress their mobility and avoid places where there is a risk of infection. In addition, in certain areas where COVID-19 is rapidly spreading or there is such a concern, if stronger social distancing measures at the local level are temporarily implemented, additional mobility control effects can be expected.

Funding

None.

Institutional review board statement

Not applicable.

Informed consent statement

Not applicable.

Data availability

The original data that support the findings of this study are available in the Seoul TOPIS [https://topis.seoul.go.kr/eng/english.jsp] for hourly traffic volumes, in the Seoul Open Data Plaza with either sheets or permitted API key for newly confirmed patients [https://data.seoul.go.kr], in the OxCGRT [https://www.bsg.ox.ac.uk/research/research-projects/covid-19-government-response-tracker] for daily social distancing index, in the National Disaster and Safety Portal with an API key [https://www.safekorea.go.kr] for real-time disaster alert, in the in the Korean Statistical Information Service (KOSIS) [https://kosis.kr/index/index.do] for exogenous variables on unemployment rate, and in the Open MET Data Portal [https://data.kma.go.kr] for weather and air quality conditions. The processed data for the analysis of this study are available from the author upon request.

CRediT authorship contribution statement

The author designed this study, collected the data, performed analysis, and wrote the paper.

Declaration of competing interest

The author declares that (i) no support, financial or otherwise, has been received from any organization that may have an interest in the submitted work; and (ii) there are no other relationships or activities that could appear to have influenced the submitted work.