Changes in traffic crash patterns: Before and after the outbreak of COVID-19 in Florida

Abstract

In the twentieth year of the twenty-first century, humanity is facing an unprecedented global crisis owing to the COVID-19 pandemic. It has brought about drastic changes in the way we live and work, as well as the way we move from one place to another, namely transportation. Previous studies have preliminarily found that mobility, travel behavior, and road traffic safety status experienced great changes after the outbreak of the COVID-19. The objective of this study is to explore how crash patterns have changed, as well as the contributing factors of such changes and the heterogeneity between counties in Florida. Thus, data of COVID-19 cases, crash, socioeconomic factors, and traffic volume of 2019 and 2020 are collected. Preliminary analyses show a considerable reduction from March to June. Substantial changes are shown in the proportions of crashes by time of occurrence and injury severity. Two types of statistical models are developed to identify factors of (1) changes in the percentages of crashes by type and (2) the numbers of crashes by type. The developed models reveal various demographic, socioeconomic, and travel factors. After controlling other factors, the total numbers of crashes are 14% lower after the outbreak. The most significant reductions are observed in peak-hour (22%), while no significant change is found in fatal crashes. The results show that the number of crashes has significantly decreased even after controlling the traffic volume, but some crash types (e.g., fatal) did not show a significant reduction. The findings are expected to provide some insights into better transportation planning and management to ensure traffic safety in a possible future epidemic.

1. Introduction

In the twentieth year of the twenty-first century, the humanity faced an unprecedented global crisis owing to the COVID-19 pandemic. It upended societies around the world, causing significant disruptions to every aspect of human life. According to the World Health Organization (WHO), more than 750 million confirmed cases were reported as of April 2023 (WHO, 2023). Governments implemented a range of measures to combat the spread of the virus, including lockdowns, travel restrictions, and social distancing requirements, which have impeded international and local trade and commerce, leading to significant losses. A report estimated the economic impact of COVID-19 (Fernandes, 2020). The report stated that a reduction in GDP of 3.5% to 6.0% in a mild scenario is expected while a contraction in GDP of 10.4% in an extreme scenario. What’s more, recent studies broadly agreed that the COVID-19 pandemic had significant impacts on different transportation fields including tourism, intermodal transportation, air transportation, rail transportation, et cetera (Nicola et al., 2020, Gray, 2020). Among them, the air transportation sector was hit hard by the pandemic, with flights being cancelled or minimized, and passenger traffic dropped drastically (Suau-Sanchez et al., 2020). Many research studies found that air transportation plays a key role in international disease spreading (Balcan et al., 2009, Bogoch et al., 2015, Chong and Zee, 2012, Colizza et al., 2006, Hufnagel et al., 2004; Meloni et al., 2009; Tatem et al., 2006, Tizzoni et al., 2012). If an outbreak occurs, the carrier can reach anywhere on the earth within several hours by the air transportation. Because the air transportation plays a crucial role in the spread, prevention and control of diseases, many researchers have developed models explaining the disease spread by air transportation with population, geographic characteristics, seasonality, and control measures.

Road transportation sector was also seriously affected. Specifically, travel demand decreased dramatically, and many people shifted to private modes of transport (e.g., passenger car) instead of transit due to the concern of infection (De Vos, 2020, Liu and Lee, 2023, Lee et al., 2020). Furthermore, the pandemic had a significant impact on traffic safety and crash patterns, including time of occurrence and injury severity of crashes. There has been a shift in the time of occurrence of crashes, with peak times shifting from morning and evening rush hours to midday and late-night hours. This is likely due to the changes in work schedules and restrictions on non-essential travel leading to more people being on the roads during non-peak hours. On the other hand, it’s common that there was a reduction in the total number of crashes reported after the outbreak of the virus. This can be attributed to a decrease in traffic volume due to the implementation of lockdown measures and travel restrictions globally. However, while the total number of crashes has decreased, the proportion of severe injury or fatal crashes remains uncertain. According to statistics from the National Highway Traffic Administration (NHTSA), in 2020, the United States experienced an increase of approximately 7.2% in the number of deaths resulting from motor vehicle traffic crashes, with about 38,680 fatalities being reported. Additionally, the country witnessed a reduction of approximately 13.2% in vehicle-miles traveled (VMT), totaling to 430.2 billion fewer VMTs. The mortality rate for 2020 stood at 1.37 deaths per million VMTs, marking an increase from the 2019 rate of 1.11 deaths per million VMTs (NHTSA, 2021). This can be linked to the speeding and aggresive driving that has been observed on less-congested roads due to the lower traffic volumes. This data highlights the continued need for improvements in road safety and accident prevention measures.

To date, numerous studies have targeted changes in crash patterns and explored the condition of traffic safety after the outbreak of COVID-19 pandemic. It is evident that due to variations in travel restriction and social distancing policies, traffic safety presented varying conditions across different regions and periods. However, there are few studies that have focused on the regional heterogeneity and influencing factors of traffic safety. Therefore, this study aims to analyze the changes in the crash pattern after the outbreak of the pandemic, as well as its relationship with the socio-demographic characteristics of the region. Of course, the best way to explore changes in transportation and travel patterns is to directly collect transportation data for multiple situations, which would be difficult to do so due to data limitation. In this study, traffic crash data are used for “surrogate” measures for the number of trips by type. The of trips made by specific type explains the most variations in the number of crashes involving the corresponding type (Lee et al., 2019b), thus it is possible to speculate the changes in the number of trips for different situations. Crash records, confirmed cases, socio-demographic data, and VMT were collected from 67 counties in Florida, US. Modeling and analysis were conducted to examine the changes in the crash pattern and their influencing factors, while controlling for traffic volume. The impact of the pandemic on road safety is comprehensive, the main goal of this study is to provide insights for a future possible epidemic.

2. Literature review

There has been a large body of research on how COVID-19 changed transportation system. Nicola et al. (2020) asserted that social distancing, self-isolation, and travel restrictions resulted in huge economic losses, and transportation sectors including tourism and aviation sectors are the biggest victims of COVID-19. Gray (2020) evaluated the changes in transportation services related to agriculture, and revealed that agriculture access to bulk ocean freight, rail movement, and trucking has enhanced because of the declined demand for these transportation services by other economic sectors. The intermodal transportation of agriculture products has experienced troubles from the lack of available containers. On the other hand, the social distancing policy has increased the demand for retail food pick-up and delivery services, which might have increased road traffic volume. Suau-Sanchez et al. (2020) assessed the impact of COVID-19 on air transportation. The authors explored airline seat capacity and freight demand of the first months of 2020. In the Asia Pacific, air freight tonnages suddenly dropped in late January and recovered afterward. Meanwhile, air freight tonnages gradually declined from early March in North America, North America, and the Middle East. The available seat kilometers (ASK) of full-service network carriers have reduced since early February, and the ASK as of late April was 81.4% smaller compared to the previous year’s same date. On the other hand, low-cost carriers reduced the ASK from March, and the ASK has reduced by 71.6%. Because the full-service network carriers have more international supplies, they experienced an early reduction. The findings imply that international air travels are more heavily affected by COVID-19 than domestic air travels. Alessio et al. (2020) summarized several important changes in transportation after the outbreak. Because of the lockdown situation, profound reductions in general consumption were observed; conversely, e-commerce and digital services are expected to continuously grow. The COVID-19 outbreak halted passengers’ mobility and restricted air and sea cargo capacity; while trans-Eurasian rail lines have been intact. Moreover, greenhouse gas (GHG) emission was considerably reduced due to the reduced mobility. Muhammad, Long, and Salman (2020) stated that nearly half of the population in the world is under lockdown because of COVID-19, and resulted in 90% reduction in total mobility. The authors used satellite data released by NASA (National Aeronautics and Space Administration) and ESA (European Space Agency) and showed that pollution in the epicenters in China, Europe, the United States, etc. has decreased up to 30%. De Vos (2020) explored how COVID-19 and following social distancing affect travel behavior. The social distancing practice might change the number and types of activities outside home. Not surprisingly, travel demands would reduce, and less people would use public transportation under the social distancing regulation. Furthermore, the authors asserted that the social distancing practice might result in limited physical activity and social isolation. Zheng et al. (2022) studied the spatial transmission of COVID-19 transportation in China. The authors revealed a positive association between the frequency of flights, trains, and buses departing from Wuhan (the epicenter in China) to other cities and the daily cumulative numbers of COVID-19 confirmed cases in those cities. The distance between Wuhan and other cities is inversely related to the number of confirmed cases in the corresponding city. Mogaji (2020) analyzed impacts of COVID-19 on transportation in Lagos, Nigeria. The author claimed that the impact in developing economies, where lockdowns and restrictions on the movement might be ineffective. The significant impacts include the increased cost of transportation, shortage of transportation mode and congestion. H. Lee et al. (2020) examined the relationship between COVID-19 and traffic levels in Korea. The authors showed a 9.7% reduction of the nationwide traffic volume for the first three months of 2020 compared to the previous year’s same period. Confirmed COVID-19 cases in Korea have progressively gradually reduced since March while traffic volume has increased.

As shown in the above-mentioned studies, it is obvious that COVID-19 has drastic impacts on transportation and mobility. Earlier studies mainly focused on analyzing effects of transportation systems on disease spreads. For the interregional level disease spreads, and public transportation plays a role such as bus, metro (Andrews et al., 2013). Andrews et al. (2013) estimated the risk of tuberculosis transmission on three modes of public transit (minibus taxis, buses, and trains). Environmental risk was highest in minibus taxis and lowest in trains; however, the average number of passengers sharing an indoor space was highest in trains and lowest in minibus taxis.

In addition, the outbreak of COVID-19 has caused widespread disruptions in road traffic safety worldwide. Numerous studies have investigated the changes in traffic crashes, utilizing various methodologies and data sources. During lockdown periods, there was a significant decrease in traffic crashes and traffic volumes in many countries. This is because the pandemic led to restrictions on movement and reduced traffic volumes, resulting in a lower likelihood of crashes occurring. However, the long-term impact of the pandemic on road safety seems uncertain, as traffic patterns and behaviors may revert to pre-pandemic levels once lockdowns are lifted, and vaccines become available. Some studies show that there was an increase in dangerous driving behavior and an uptick in fatal crashes. For instance, Yao et al. (2021) analyzed the data including traffic volume, daily cases, weather, social distancing index, and crashes from January 2019 to June 2020 in Detroit, and found that the angle crashes have increased significantly. Doucette et al. (2021) investigated the impact of the pandemic and stay-at-home orders on crash patterns in Connecticut and found that the vehicle mileage traveled (VMT) and crash numbers have significantly reduced but the crash rate has increased in consideration of the reduced VMT. Koloushani et al. (2021) identified the significant socio-demographic and transportation-related factors contributing to crash count decrease during COVID-19 in four Florida counties by negative binomial regression. Stiles et al. (2021) studied road traffic safety in Ohio’s Franklin County and confirmed a lower volume and a relatively higher crash severity during the stay-at-home period. Since the increase of fatal crashes in the US has been initially proved by statistical analysis, Adanu et al. (2021) further explored the associated factors. They found that the majority of the increased crashes in Alabama are speeding, DUI (drunk and drug-impaired driving), and weekend crashes. Alhajyaseen et al. (2022) investigated the road safety status during the pandemic by collecting road crash data (from 2015 to 2020) and traffic violation data (from 2019 to 2020) from Qatar. Statistical analysis and Z-tests were conducted, and the results show that the rates of minor/major and fatal injuries per 1000 crashes climbed up during the highest restrictions period.

The increase in dangerous driving behavior may be attributed to several factors. One is the psychological impact of the pandemic and the lockdown, which may have led to pent up frustrations and stress. Drivers might have driven more aggressively or taken more risks because of fewer vehicles on the road. Katrakazas et al. (2020) investigated the impact of COVID-19 on driving behavior and safety indicators using the data captured through a special smartphone application in Greece and the Kingdom of Saudi Arabia. They found that the traffic volume reduced drastically but dangerous driving behavior increased significantly after the outbreak of COVID-19, including speeding, harsh acceleration and harsh braking, and phone use. Vingilis et al. (2020) used interactionist model to identify the potential person and situation factors associated with COVID-19 that may affect traffic safety, and found that economic factors, increased stress and anxiety brought by the pandemic, more “free” time and more dangerous driving behavior may have different impacts on road traffic safety.

In conclusion, though there has been a series of studies demonstrating the universal reduction in both traffic volume and crash frequency, as well as the probable rising trends of severe and fatal crashes after the outbreak of COVID-19, rare research has been found to investigate the contributing factors related to socio-demographic aspects on the inter-county perspective. Thus, this study aims to figure out how has road traffic safety in the state of Florida changed after the outbreak of COVID-19, which emphasized the analysis of socio-demographic factors and the heterogeneity between different counties in Florida. It is essential to gain a better understanding of the complex relationship between the COVID-19 pandemic and road safety and to take area heterogeneity into consideration when developing partial and resilient traffic restrictions and management policies and measure in a possible future epidemic.

3. Data collection

The study area is Florida, which includes sixty-seven counties. Florida has experienced rapid growth of confirmed cases and deaths of COVID-19, and it shows disparities in outcomes by race and ethnicity ( Samuels-Staple, 2020 ). The unit of analysis is county.

Data used in this study were collected from four sources. First, crash data were obtained from the Signal Four Analytics database of the Florida Department of Transportation. Two-year crash data of 2019 and 2020 were used for the analysis. The crash types were classified by time of occurrence and injury severity. Second, socio-demographic data were acquired from the U.S. Census Bureau’s American Community Survey 2018. The data include population, education attainment, commuting mode of transportation, occupation, household income, health insurance coverage, and male population. The COVID-19 data regularly released by the Florida Department of Health were also collected. They are the numbers of confirmed cases and deaths. Last, the travel-related data was obtained from the Florida Department of Transportation. Vehicle mileage traveled of each county is selected for analysis in order to control the traffic volume and figure out whether traffic safety is worsened or improved. The descriptive statistics of the processed data are presented in Table 1 .

Table 1.

Descriptive Statistics of the Processed Data (N = 67).

	Variable	Mean	S.D.	Min	Max	Source
Crash	Total (2019)	11100.0	22347.9	39	143,751	Florida Department of Transportation (Signal Four Analytics)
	Total (2020)	8740.7	16425.5	45	103,453
	Peak (7–9 a.m. and 4–6p.m.) (2019)	1158.3	2352.9	4	14,528
	Peak (7–9 a.m. and 4–6p.m.) (2020)	731.1	1378.7	2	8443
	Off-peak (9 a.m. – 4p.m.) (2019)	4779.8	9490.5	10	61,280
	Off-peak (9 a.m. – 4p.m.) (2020)	3872.6	7200.0	19	45,489
	Fatal (2019)	44.1	56.8	1	281
	Fatal (2020)	45.8	60.5	0	318
	Injury (2019)	2511.5	4202.5	19	21,479
	Injury (2020)	2100.2	3334.1	22	16,512
	PDO (property damage only) (2019)	8563.5	18311.4	19	122,087
	PDO (property damage only) (2020)	6612.4	13225.8	23	86,731
Socio-demographic	Population (2019)	316547.8	499413.5	8500	2,812,100	U.S. Census Bureau American Community Survey 2018
	Population (2020)	322349.3	505194.8	8600	2,833,200
	% of unemployment	3.1	0.8	0.8	4.7
	% of public transport commuters	0.9	1.2	0	6.6
	% of walking commuters	1.4	0.9	0.2	91.2
	% of people working at home	5.4	2.1	0.6	11
	% of tertiary industry occupations	75.6	7.2	52.1	88.4
	% of natural resources, construction, and maintenance occupations	12.5	5.1	5.1	32.4
	% of production, transportation, and material moving occupations	11.9	3.2	6.1	19.1
	Median household income (in 1,000 USD)	51.3	10.2	35.438	82.252
	% of families below poverty level	11.4	4.3	4.6	21.9
	% of families with no health insurance coverage	12.9	3.5	5.6	25.6
	Male-female ratio	106.9	17.9	90.2	187.4
	Older dependency ratio	38.5	20.7	17.6	157.5
	Child dependency ratio	32.9	4.5	20.2	45.2
COVID-19	Confirmed cases of COVID-19 per population (per 100 persons)	6.4	2.2	16.3	3.0	Florida Department of Health
	Death cases of COVID-19 per population (per 100 persons)	0.1	0.1	0.4	0.0
VMT	VMT (2019)	7606374.0	10717610.4	215155.1	48092927.2	Florida Department of Transportation
	VMT (2020)	7029373.7	9,713,336	205161.2	41978345.3
	Percentage change of VMT	−0.058	0.043	−0.165	0.101

# VMT means vehicle mileage traveled as mentioned above.

4. Temporal and spatial analyses

Temporal and spatial analyses are conducted to visually present the changes after the outbreak. Fig. 1 depicts significant reductions in the number of total crashes of the same months between 2019 and 2020 after March, as well as the percentage change in each month. The number of crashes slightly increased after May in 2020 but did not exceed 2019.

Fig. 1.

Changes in the monthly number of total crashes between 2019 and 2020. # The left axis stands for the number of crashes while the right one stands for the percentage change. (Note: the monthly number of crashes is based on the last day of the months).

Fig. 2 compares the relative percentage changes by crash type. The relative change refers to the difference between the number of crashes of a specific type in 2019 and 2020, divided by the total number of crashes of the same type in 2019 as in the following equation.

$$\begin{array}{c}{\mathit{PC}}_{\mathit{tm}}=\frac{N_{tm2020}-N_{tm2019}}{N_{tm2019}}\end{array}$$ (1)

where ${\mathit{PC}}_{\mathit{tm}}$ is the relative percentage change of crashes of type $t$ of month $m$, $N_{tm2020}$ is the number of crashes of type $t$ of month $m$ in 2020, and $N_{tm2019}$ is the number of crashes of type $t$ of month $m$ in 2019.

Fig. 2.

Relative changes of crashes by type between 2019 and 2020 (Note: the monthly number of crashes is based on the last day of the months).

This analysis was performed to understand the changes in crash patterns in Florida after the outbreak of COVID-19 by comparing the relative percentage changes of each type of crash. From the perspective of the time of occurrence, the reduction rate of peak crashes is obviously more drastic than off-peak crashes. Additionally, from the perspective of crash severity, the decrement of PDO (property damage only) crashes is the most obvious, followed by injury crashes. It is noticeable that the number of fatal crashes even increased after May.

Fig. 3 compares the proportion of crashes by type between 2019 and 2020 in different counties of Florida. Relative changes of each kind of crash are shown according to the gradient of color. The urban area in Florida is also presented. The number of total crashes has decreased in urban areas after the outbreak more than in rural areas, except for fatal crashes. Moreover, the sources of heterogeneity among counties need to be explored through models.

Fig. 3.

Relative change of crashes between 2019 and 2020 by type. # The red color stands for increased crashes whereas the green color for decreased crashes, and the shade of color indicates the gradient of change. (For interpretation of the references to color in this figure legend, the reader is referred to the web version of this article.)

To explore the if the influences of the COVID-19 still exist, crash data from 2021 and 2022 in Florida were collected from the Signal Four Analytics dataset, categorized by occurrence time and severity of injury, aggregated by county. Paired t-tests are conducted to figure out if there is significant difference between the crash frequencies before and after the COVID-19 outbreak.

According to Table 2 , for total crash, off-peak crash and injury crash, there is a significant difference between 2019 and 2020, while the differences in 2019 vs. 2021 and 2019 vs. 2022 is insignificant. For peak crash, all paired t-tests showed a significant difference. The mean peak crash frequencies after the COVID-19 outbreak (2020–2022) are fewer than that before the outbreak (2019). It might be because many people still telework even after the endemic. For fatal crash, the difference between 2019 and 2020 is insignificant, while there are significant differences in 2019 vs. 2021 and 2019 vs. 2022. The number of fatal crashes after the outbreak is larger than that before the outbreak. For PDO crash, the crash frequencies after the outbreak are significantly fewer than that before the outbreak.

Table 2.

Paired t-test results of crash frequencies before the COVID-19 outbreak (2019) vs. after the COVID-19 (2020, 2021, and 2022) outbreak (N = 67).

Crash type	Year	2019	2020	2021	2022
Total	Mean	11100.0	8740.7	10469.2	10390.2
	Standard deviation	22516.5	16549.5	20228.8	20364.3
	t-statistic	–	3.164	1.955	1.969
	p-value	–	0.002***	0.055*	0.053*
Off-peak	Mean	4779.8	3872.6	4555.0	4478.4
	Standard deviation	9562.1	7254.4	8692.2	8613.7
	t-statistic	–	3.132	1.756	1.907
	p-value	–	0.003***	0.084*	0.061*
Peak	Mean	2965.4	2117.7	2634.6	2676.9
	Standard deviation	5952.1	3961.2	5006.1	5175.0
	t-statistic	–	3.441	2.645	2.450
	p-value	–	0.001***	0.010**	0.017**
Fatal	Mean	44.1	45.8	51.5	47.9
	Standard deviation	57.2	60.9	69.9	62.6
	t-statistic	–	−1.684	−3.941	−3.093
	p-value	–	0.097*	<0.001***	0.003***
Injury	Mean	2511.5	2100.2	2440.9	2395.8
	Standard deviation	4234.2	3359.3	3936.4	3841.6
	t-statistic	–	3.688	1.371	1.626
	p-value	–	<0.001***	0.175	0.109
PDO	Mean	8563.5	6612.4	7976.9	7946.4
	Standard deviation	18449.6	13325.6	16460.5	16705.0
	t-statistic	–	3.055	2.109	2.076
	p-value	–	0.003***	0.039**	0.042**

(Note: *** significant at 0.01, ** significant at 0.05, and * significant at 0.1).

5. Method

The main analysis of this study is to identify factors of 1) changes in the percentages of crashes by type and 2) the numbers of crashes by type. For the first analysis, linear regression models are estimated. Linear regression is widely used and simple, explanations of the techniques are omitted. Readers are referred to the works of Abdel-Aty and Radwan (2000) and Washington et al. (2020) for more information. The dependent variable of the first analysis is the percentage change between the years. The equation for calculating the percentage change of a specific type is as follows:

$$\begin{array}{c}{\mathit{PC}}_t=\frac{{(N}_{t,2020}-N_{t,2019})}{N_{t,2019}}\end{array}$$ (2)

where,${\mathit{PC}}_t$ is the percentage change of a specific type $t$, $N_{t,2020}$ is the number of crashes of a type $t$ in 2020, and $N_{t,2019}$ is the number of crashes of type $t$ in 2019. For some rare crash types, the denominator ($N_{t,2019}$) might be zero and the ${\mathit{PC}}_t$ cannot be computed. In this case, the observation is removed and unused for modeling.

The dependent variable of the second analysis is the number of crashes by each type. For this analysis, random parameter panel data negative binomial regression models are built. Crash data consist of nonnegative integer values. Count data models (i.e., Poisson regression model and negative binomial regression) are encountered frequently in the modeling of crash data. The probability of county $i$ having $y_i$ crashes per year is given by

$$\begin{array}{c}P\left(y_i\right)=\frac{\exp \left(-{\lambda }_i\right){{\lambda }_i}^{y_i}}{y_i!}\end{array}\begin{array}{c}{\lambda }_i=\exp \left(\mathit{\beta }{\mathit{X}}_i+{\epsilon }_i\right)\#\left(4\right)\end{array}$$ (3)

$$\begin{array}{c}{\lambda }_i=\exp \left(\mathit{\beta }{\mathit{X}}_i+{\epsilon }_i\right)\end{array}$$ (4)

To account for the influences of unobserved heterogeneity across counties, random parameters are incorporated into the model. It assumes that the estimated parameters vary across counties according to some specified distribution. Normal distribution is adopted in this study.

The vehicle mileage traveled (VMT) of each county was used as an exposure variable. For each observation $i$, there is

$${\lambda }_i=\exp \left(\mathit{\beta }{\mathit{X}}_i+{\epsilon }_i+{\ln \left(VMT\right)}_i\right)$$ (5)

The coefficient of exposure variable ${\mathrm{l}\mathrm{n}(VMT)}_i$ is set to be one, that is to say,

$$\begin{array}{c}\frac{{\lambda }_i}{{\mathit{VMT}}_i}=\exp \left(\mathit{\beta }{\mathit{X}}_i+{\epsilon }_i\right)\end{array}$$ (6)

It can be interpreted as the ratio of the expected crash number and VMT, which is more reasonable than crash number itself.

Also, a dummy variable indicating the period after the outbreak is attempted. The two analyses of this study are macroscopic safety investigations that analyze traffic safety at zonal level ( Abdel-Aty et al., 2013, Cai et al., 2017, Cai et al., 2016, Huang et al., 2016, Lee and Abdel-Aty, 2018, Lee et al., 2017b, Lee et al., 2018a, Lee et al., 2019a, Lee et al., 2019b, Lee et al., 2017a, Lee et al., 2018b). The macroscopic safety studies aim to identify the relationship between traffic safety level and socio-demographic characteristics of the area.

6. Results

6.1. Changes in the percentages of crashes by type

Table 3 summarizes the linear regression models for total, peak, off-peak, fatal, injury and PDO crashes after the COVID-19 outbreak. Pearson correlation test was conducted, and highly correlated independent variables were excluded (correlation coefficient > 0.7 or < -0.7). The modeling results showed that the percentage changes of crashes after the outbreak are associated with many socioeconomic, demographic and traffic related variables. Generally, counties with higher population, walking commuters, percentage of tertiary industry, and percentage of male population have experienced a more drastic reduction in the corresponding crashes. The percentage change of VMT has a positive impact on the percentage change of PDO crashes.

Table 3.

Linear regression models for percentage changes for total, peak, off-peak, fatal, injury and PDO crashes after the COVID-19 outbreak.

Crash type	Coefficient (p)
Variable	Total	Peak	Off-peak	Fatal	Injury	PDO
Intercept	0.70133 (<0.001)	0.17931 (0.3)	0.28938 (0.142)	0.01649 (0.969)	0.0166 (0.91)	0.0608 (0.209)
Log population	−0.03118 (<0.001)	−0.02579 (0.038)	−0.0472 (<0.001)		−0.0354 (0.001)
Percentage change of VMT						0.8576 (0.008)
% of walking commuters	−0.03916 (0.002)	−0.06776 (<0.001)	−0.06429 (0.005)			−0.03 (0.057)
% of people working at home	0.01046 (0.063)			0.06191 (0.029)	0.0209 (0.006)	−0.0155 (0.017)
% of natural resources, construction, and maintenance occupations		0.00745 (0.054)
% of production, transportation, and material moving occupations					0.0162 (0.001)
% of tertiary industry	−0.00594 (0.002)
% of families below the poverty level				0.03709 (0.01)
Male-female ratio				−0.00701 (0.027)
Child dependency ratio			0.00794 (0.068)
R2	0.47	0.32	0.28	0.18	0.36	0.21

6.2. The number of crashes by type

Table 4 shows the negative binomial models for total, peak, off-peak, fatal, injury and PDO crashes. The independent variables with standard deviation are modeled as random parameters fitted by normal distribution. Similarly, variables with high Pearson’s coefficient correlation (>0.7 or < -0.7) were deleted before modeling. The VMT (vehicle mileage traveled) is employed as the exposure in the negative binomial model.

Table 4.

Random parameter panel data negative binomial models for numbers of total, peak, off-peak, fatal, injury and PDO crashes.

Crash type	Coefficient (p)
Variable	Total	Peak	Off-peak	Fatal	Injury	PDO
Intercept	−5.87743 (<0.01)	−7.98046 (<0.01)	−8.37741 (<0.01)	−8.27039 (<0.01)	−7.18093 (<0.01)	−6.81649 (<0.01)
Standard deviation	0.18805 (<0.01)	0.28240 (<0.01)	0.21503 (<0.01)	0.17294 (<0.01)	0.01145 (<0.1)	0.28812 (<0.01)
Year dummy (2020 = 1, 2019 = 0)	−0.14890 (<0.01)	−0.24808 (<0.01)	−0.17095 (<0.01)	0.02718 (0.3777)	−0.14704 (<0.01)	−0.16086 (<0.01)
Standard deviation			0.11856 (<0.01)		0.03395 (0.01)
Log population	1.00243 (<0.01)	1.12354 (<0.01)	1.13218 (<0.01)	0.85093 (<0.01)	1.02443 (<0.01)	1.0608 (<0.01)
Standard deviation	0.02560 (<0.01)		0.02206 (<0.01)		0.02359 (<0.01)	0.01599 (<0.01)
% of unemployment	−0.08626 (<0.01)	−0.07622 (<0.01)	−0.13324 (<0.01)			−0.09452 (<0.01)
Standard deviation		0.04961 (<0.01)				0.03576 (<0.01)
% of public transport commuters			0.04171 (<0.01)	−0.08177 (<0.01)
% of walking commuters	0.07969 (<0.01)	0.03673 (<0.01)	−0.02877 (<0.01)	0.11221 (<0.01)		0.10853 (<0.01)
Standard deviation		0.05357 (<0.01)	0.04239 (<0.01)			0.09697 (<0.01)
% of people working at home			0.0144 (<0.01)
Standard deviation			0.02049 (<0.01)
% of tertiary industry		−0.00493 (<0.1)			−0.00714 (0.01)
Median household income (in 1,000 USD)			−0.02626 (<0.01)
% of families with no health insurance coverage	−0.02387 (<0.01)	−0.03099 (<0.01)			−0.04204 (<0.01)	−0.03017 (<0.01)
Standard deviation						0.0138 (<0.01)
Older dependency ratio	−0.00285 (<0.01)	−0.00748 (<0.01)	−0.00367 (<0.01)		−0.00448 (<0.01)	−0.00269 (<0.01)
Standard deviation	0.00122 (<0.01)	0.00215 (<0.01)
Child dependency ratio					0.01469 (<0.01)
R2	0.55	0.54	0.56	0.60	0.56	0.55

(Note: The VMT (vehicle mileage traveled) is employed as the exposure.).

Multiple socio-demographic variables were found significant. The most important variable in the models is the year dummy (1: 2020, 0: 2019) indicating the time period after the outbreak of COVID-19. Pearson correlation test showed a strong correlation between year dummy and confirmed cases/death cases related to COVID-19, and the correlation coefficients are respectively 0.897 and 0.797. Thus, both year dummy and COVID-19 cases are selected to establish models, and models with year dummy show better performance.

Year dummy is significant for the most of crash types except for fatal crashes. And the coefficients of this variable in all the models is minus except for fatal crash. It implies that no significant reduction was shown after the outbreak for fatal crashes. The population is logarithmic to avoid too small coefficients. The coefficients of this variable are significantly positive in all six models, but the exact values are different, especially in fatal crashes (noticeably smaller than others). The percentage of unemployment is significant in models for total, peak, off-peak, and PDO crashes, and the coefficients are all negative. As for the percentage of public commuters, the estimation results are different. It is observed to have a positive impact on off-peak crashes, while the relationship is negative for fatal crashes. The high proportion of public transportation indicates a lower percentage of private car users, resulting in a probable reduction in traffic volume. Concerning the percentage of walking commuters, it is positively correlated in all models, except for off-peak crashes. The higher the percentage of homeworkers, the more off-peak crashes are estimated. Notice that there is little difference between the model results of total crashes and PDO crashes. It is reasonable because the vast majority of crashes are PDO crashes.

Furthermore, it is possible to calculate crash modification factors (CMFs) by exponentiating the estimated coefficient of the year dummy variable (Gross et al., 2010). The estimated CMFs from the developed models suggest the changes of the number of crashes after the COVID-19 outbreak, after controlling other factors. The estimated CMFs are summarized in Table 5 . Overall, 13.8% of crashes have been reduced. The greatest reduction was observed for peak crashes (22.0%), and it is followed by off-peak crashes (15.7%) and PDO crashes (14.9%). Injury crashes experienced 13.7% reduction. It is speculated that the stay-at-home order and the trend of working from home could be the major reason why peak crashes experienced the greatest reduction (Sekadakis et al., 2021, Christey et al., 2020 Apr). On the other hand, fatal crashes have not statistically significantly decreased, which is consistent with the findings in Los Angeles and New York City (Lin et al., 2021), Greece (Sekadakis et al., 2021), Alabama (Adanu et al., 2021), and Qatar (Alhajyaseen et al., 2022). Possible explanations related to fatal crashes could be a higher tendency of dangerous driving behaviors such as higher speed, harsh acceleration/deceleration, and increased use of alcohol/drugs (Sekadakis et al. 2021). Drivers are proven to tend to overspeed when there is less traffic volume during the COVID-19 pandemic (Katrakazas et al., 2020, Stiles et al., 2021). At the same time, public anxiety and depression caused by the pandemic could be a reason of increased use of alcohol and drugs, which is worthy of social concern (Adanu et al. 2021).

Table 5.

Crash modification factors and crash reduction factors of the COVID-19 outbreak.

Crash Type	CMF	Crash reduction (%)
Total	0.8617***	13.8%
Peak (7–9 a.m. and 4–6p.m.)	0.7803***	22.0%
Off-peak (9 a.m.-4p.m.)	0.8429***	15.7%
Fatal	1.0276	−2.8%
Injury	0.8633***	13.7%
PDO (property damage only)	0.8514***	14.9%

(Note: *** significant at 0.01, ** significant at 0.05, and * significant at 0.1).

7. Discussions and implications

7.1. Temporal and spatial analysis

Through descriptive statistics and temporal analysis, a remarkable reduction in all kinds of crashes is observed after the outbreak of COVID-19 in Florida (i.e., March 2020). In addition to public perception of own safety, countermeasures taken by local governments also played an important role. Referring to the timeline of state government response, public health emergency was declared on March 1, 2020. As the pandemic got worse, the stay-at-home order was issued for several South Florida counties and the entire state respectively on March 30 and April 1, 2020 (CDC, 2020). One interesting finding is that the number recovered to a relatively higher level from June, while the disease was not getting better at that moment. This recovery of transportation is probably because of policy change. executive order was issued on June 5, 2020, implementing “Phase 2″ of a reopening plan outside of Miami-Dade, Broward, and Palm Beach counties, allowing some businesses (such as bars, pubs, and indoor venues such as arcades, bowling alleys, cinemas, and concert venues) to expand to 50% capacity. The order recommended against gatherings of more than 50 people (CDC, 2020).

As for the changes in different types of crashes, results show that peak crashes experienced a sharper decline compared to non-peak crashes. Additionally, road traffic crashes particularly in PDO accidents were substantially reduced as a result of shutdown policies in Florida, not including crashes leading to fatal injuries. The number of fatal crashes even rose in fluctuation after falling to the lowest level in April. The findings indicate that the crash patterns in the study area have changed a lot after the COVID-19.

Results of spatial analysis reveal the existence of heterogeneity among counties in Florida. Furthermore, it can be visually observed that there is likely a correlation between the changes in crash patterns and the location of the urban areas. The urban areas are mainly located in central and southern Florida. In these areas, peak crashes and PDO crashes experienced a more obvious decrease, while the proportion of fatal crashes increased more significantly. Therefore, it is necessary to build county-level quantitative models to analyze how the pandemic and socioeconomic factors impact crashes by different types.

Results of the paired t-tests indicate that the long-term influences of the COVID-19 exist in certain crash types, including peak crash, fatal crash, and PDO crash. It is inferred that the trend of teleworking after the outbreak of the pandemic has been influencing people's commuting patterns. Moreover, fatal crash did not significantly increase in 2020, but rather increased significantly in 2021 and 2022 compared to pre-pandemic levels, indicating that the impact of the pandemic on drivers' dangerous driving behavior and traffic safety is likely to persist in the long term.

7.2. Statistical model results and policy implications

Linear regression models reveal the contributing factors to the changes in each type of crash. The dependent variables are negative in most types of crashes in most counties, that is to say, the models aim to estimate the degree of reduction. Moreover, the coefficients indicate that the influence level varies in different types of crashes. The results show that the factors of more population can lead to a more drastic decrement in most types of crashes except for fatal and PDO crashes. Besides, VMT, travel modes, and industry structure of the regions also have significant impacts on the percentage change of crashes.

On the other hand, the results of number of crashes by type reveal the changes in crash patterns in Florida, as well as the impact of COVID-19 and socio-economic factors on it. It is valuable to conduct a further discussion in consideration of policy changes, which is expected to provide additional insights into better transportation planning and management in order to ensure traffic safety under the premise of pandemic prevention. One reasonable action is to conduct resilient and partial traffic control policies. In terms of the occurrence time of crashes, people's commuting patterns have been affected by the pandemic, resulting in changes in crash patterns. Crashes during the peak hour have been significantly reduced compared to non-peak crashes. According to the model coefficients, off-peak crashes are most affected by regional population (β = 1.13218). Therefore, it is also important to pay attention to off-peak traffic control in regions with more population, less percentage of unemployment, more people working at home, more public transport commuters, lower household income, and more younger residents.

From the perspective of crash severity, fatal crash is a major concern, because it is found to increase to some extent after the outbreak of COVID-19. Increased dangerous driving behaviors during the pandemic including higher speed, harsh acceleration/deceleration, and increased use of alcohol/drugs could be the main reason (Sekadakis et al., 2021, Alhajyaseen et al., 2022). To further investigate possible reasons for the slight increase in fatal crashes, the crash dataset involving four types of traffic violations, including speeding, aggressive driving, drunk driving, and drug driving, were collected from the Florida Department of Transportation. The violation rate ${\mathit{VR}}_i$ is defined as the number of traffic violations per million VMT as:

$$\begin{array}{c}{\mathit{VR}}_i=\frac{V_i}{{\mathit{VMT}}_i\times 100000}\end{array}$$ (7)

where ${\mathit{VR}}_i$ represents the traffic violation rate of a certain type of illegal behavior in county $i$, $V_i$ represents the number of traffic violations in county $i$, and ${\mathit{VMT}}_i$ represents the annual VMT in county $i$.

After calculating the violation rates for each county in 2019 and 2020, paired-sample t-tests (one-tailed) were conducted, and the results are summarized in Table 6 . The results showed that the violation rates for speeding, aggressive driving, and drug driving in Florida counties significantly increased after the COVID-19 outbreak, while the violation rates of drunk driving did not show a significant increase.

Table 6.

Results of paired-sample t-tests of the violation rate increase of different types of traffic violations (N = 67, Unit: violations per million VMT).

Type of violations	Mean		Variance		t	p (one-tailed)
	2019	2020	2019	2020
Speeding	24.557	26.924	171.716	172.755	2.573	0.006
Aggressive driving	36.289	38.730	456.535	506.577	1.888	0.032
Drunk driving	27.557	27.920	114.982	109.265	0.341	0.367
Drug driving	9.594	10.562	20.023	31.238	1.578	0.060

Such increased violation patterns might imply that the slight increase of fatal crashes was due to reckless driving on less-congested roads. Thus, implementing stricter traffic enforcement measures are necessary, especially in areas with more population and fewer public transport commuters (Table 4). This can include empowering police enforcement on the road, installing more speed cameras and imposing larger fines for traffic violations. In addition, conducting regular education and safety campaigns can help raise awareness about the dangers of aggressive driving and encourage safer behaviors on the road, especially during an epidemic.

Since this paper emphasizes how traffic safety changed due to the COVID-19 pandemic and the contributing socio-demographic factors behind the phenomenon, there are certain limitations to this study that could bring more research insights in the future. First, due to the data limitation, it was unable to model crashes on monthly basis. Second, a comparative analysis towards the impact of COVID-19 on different road users (e.g., passenger car, transit, active modes of transportation, micro-mobility) was not conducted due to the data unavailability; Third, only one state was chosen as the study area. It is quite possible that there is a significant difference between countries/regions as their travel restriction policies were not the same.

8. Conclusions

This study aimed at analyzing the changes in traffic crash patterns before and after the outbreak of COVID-19 using crash data from Florida of 2019 and 2020. Crash and traffic volume (VMT), socio-demographic, and COVID-19 data were collected from the Florida Department of Transportation, the U.S. Census Bureau, and the Florida Department of Health, respectively. The temporal analysis showed a considerable reduction after the outbreak, compared to the previous year’s same months. Substantial changes are shown in the proportions of crashes by time of occurrence, injury severity, and crash types. The spatial analysis identified that most crashes have decreased in urban areas. Also, the heterogeneity among different counties has been initially confirmed.

Two types of statistical models are developed to identify factors of (1) changes in the percentages of crashes by type; and (2) the numbers of crashes by type. The developed models reveal various traffic, demographic, socioeconomic, and disease factors. After controlling other factors, the total numbers of crashes are 14% lower after the outbreak. The most significant reductions are observed in peak-hour (22%), while no significant change is found in fatal crashes.

According to the WHO, COVID-19 remains a global health emergency, but pandemic is at a “transition point” (Dillinger, 2023). As the pathogenicity has diminished, and countries have lifted their quarantine measures, mobility has gradually returned to pre-pandemic levels. However, it is not certain that the current traffic safety is the same as before the pandemic. The long-term impact of the pandemic on traffic safety is worth exploring. But certainly, within the scope of this study, road traffic crashes experienced a sharp reduction as the traffic volume reduced, while fatal crashes appear to have a tendency of the slight increment after the outbreak of the COVID-19 pandemic. And the changes show significant heterogeneity in different counties in Florida with different socioeconomic status. Lastly, it was shown that speeding, aggressive driving, and drug driving violation rates showed a significant increase after the outbreak, which might explain the slight increase of fatal crashes.

This study is expected to provide some insights into the flexible and locally-tailored traffic restriction measures during a possible future epidemic. It is expected that transportation engineers and practitioners can establish realistic and effective transportation plans considering the area heterogeneity. In the future, it would be necessary to conduct interdisciplinary research to investigate the association between dangerous driving behavior (especially the abuse of alcohol and drugs) and negative public feelings owing to the pandemic.

Author contributions

The authors confirm contribution to the paper as follows: study conception and design: Jaeyoung Lee, Mohamed Abdel-Aty; data collection: Jaeyoung Lee; analysis and interpretation of results: Jaeyoung Lee, Haiyan Liu, and Mohamed Abdel-Aty; draft manuscript preparation: Jaeyoung Lee and Haiyan Liu. Author. All authors reviewed the results and approved the final version of the manuscript.

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.

Data availability

The authors do not have permission to share data.