2.1. Impact of Mobility Restrictions on Road Safety During Pandemic

Research studies have highlighted the significant impact of mobility restrictions on reducing the rate of crashes and injuries during the COVID-19 pandemic. For instance, a 30% reduction in traffic volume in Qatar resulted in a remarkable 37% decrease in crash frequency [20]. Similarly, in Spain, Saladié et al. found that a 12-week containment order led to a 34% reduction in overall crashes compared to the average of 2018–2019 [21]. Another study conducted in Greece observed a substantial decline in crashes, with a 62% reduction, along with a 68% decrease in fatalities and a 48% drop in serious injuries during the pandemic [22]. In a comprehensive study conducted by Gupta et al., several countries, including the United States of America, Australia, Brazil, Russia, Denmark, Finland, France, Germany, Greece, Hungary, Italy, Japan, Poland, Spain, and Sweden, were examined to assess the impact of strict lockdown policies and reduced mobility on road fatalities during the COVID-19 pandemic [23]. According to this study, fewer travels for shopping and leisure activities have resulted in a notable decline in road fatalities across all nations. These findings highlighted that implementing rigorous measures of lockdown and limiting mobility have proven to be effective strategies in promoting road safety by minimizing vehicular traffic and associated risks.

Despite the results of the studies mentioned earlier, some researchers have observed only slight alterations in crash rates amidst the COVID-19 pandemic. Studies carried out in Nepal showed that the predicted decrease in road traffic accidents, injuries, and fatalities was not as substantial as expected at the time of the pandemic [24]. Similarly, despite a decrease in the overall crash rate, the number of fatalities from traffic crashes in Northern Ireland remained unchanged compared to prepandemic levels [25]. In Missouri State, minor injury crashes experienced reductions, whereas fatal crash patterns did not change [26]. In addition, there has been an escalation in the number of fatalities observed in various regions worldwide [27]. According to research conducted in Alabama, although there was a significant decrease in traffic flow and distance covered by vehicles, an unfortunate surge in crashes, injuries (major), and fatalities was reported [28]. Higher fatality rates were also reported in Queensland, Australia, during the pandemic [29]. According to most of the literature, the leading cause of the poor road safety situation is attributed to dangerous driving habits, such as driving under the influence of alcohol and exceeding the speed limit [27, 30, 31].

Incorporating the detailed statistics into the comparative analysis between the Global North and Global South, we observe a diverse impact of the pandemic on road traffic collisions/crashes (RTCs) [32]. The significant reduction in RTCs in Spain (67%), France (74%), and Germany (23%) reflects the effectiveness of lockdown measures in the Global North, where advanced transport systems and strict adherence to restrictions contributed to a pronounced decrease in traffic volumes. Conversely, in the Global South, countries like Nepal (48%) and UAE (84%) also saw significant reductions, yet the variations within and between these regions underline the influence of local transport systems, behaviors, and enforcement of lockdown measures. This disparity emphasizes the critical need for tailored road safety strategies that consider regional transport characteristics and pandemic response effectiveness.

2.2. Motorcycle-Involved Crashes During COVID-19

During the COVID-19 pandemic, motorcycle-related crashes increased significantly in various regions, including Türkiye and the Middle East. In Türkiye (2024), a study revealed a notable rise in motorcycle accidents, particularly among motor couriers, during quarantine periods, with an increase in injuries among motorcyclists who relied on two wheelers for essential services [33]. In Qatar, another study found that the surge in demand for motorcycle delivery drivers during the pandemic led to a 66.3% increase in motorcycle-related crashes, primarily among young, work-related riders, compared to the prepandemic period [34]. In Kentucky, a separate investigation compared pre- and postpandemic motorcycle crashes along urban roadways, finding that factors like higher traffic volumes and sharp horizontal curves contributed to increased crash frequencies in both periods [35]. In Taiwan, motorcycle crashes increased from 47,108 in 2019 to 51,441 in 2020, with a rise in injuries and fatalities, particularly among food delivery riders [36]. Interestingly, in Florida while motorcycle crashes decreased during the pandemic, fatality rates rose [37]. These findings emphasize the increased vulnerability of motorcyclists during the pandemic due to heightened demand for delivery services and reduced public transport.

2.3. Challenges in Data Availability for Developing Countries and Alternative Data Sources

The adoption of mobile sensors and publicly available data from companies such as Google, Facebook, and Cuebiq has revolutionized the evaluation of transportation systems and mobility changes in real time, allowing for detailed assessments of the impact of COVID-19 on human mobility and traffic safety [38]. This approach, alongside a systematic literature review on big data applications during the pandemic, underscores the critical role of technological advancements in understanding and mitigating transport safety impacts. However, it also highlights the need for further research, especially in developing countries where traditional data sources may not fully capture the extent of road traffic injuries and fatalities [39].

The limited data availability in developing countries poses a significant challenge to comprehensive road safety analysis. This review highlights innovative approaches taken by researchers to overcome data gaps, such as the utilization of newspaper reports and social media content to infer traffic accident trends and causes. Yet, the scarcity of systematic and reliable data remains a critical barrier to understanding the full impact of the pandemic on road safety in these regions. The lack of evidence underscores the urgent need for investment in robust data collection and analysis systems that can provide actionable insights for policymakers and road safety practitioners.

The impact of restrictive measures implemented to combat the transmission of the coronavirus has demonstrated both positive and negative consequences for road safety. Hence, understanding these impacts and their relationship with road crash patterns is essential. However, limited data availability in Bangladesh poses a significant challenge to road safety analysis. In response, researchers are turning to alternative sources, such as newspapers, to bridge the gap caused by data unavailability [40, 41]. For example, the research by Ahmed et al. utilized newspaper-based data to examine the causes and trends of fatal vehicle crashes involving truck in Bangladesh [40]. This approach is common in other countries as well [42–44]. Interestingly, many studies have employed time series analysis to analyze newspaper data. Jünger and Gärtner used time series analysis to examine the trend of media coverage and reporting on terrorism, while Daw et al. conducted a monthly time series analysis to quantify changes in newspaper coverage of drug financing over time [45, 46].

Motorcycles play a crucial mode of transportation in low-income developing countries, including Bangladesh. However, they are also associated with high levels of risk [47]. Hence, this study holds immense significance in guiding policymakers on how to handle future waves or related situations, intending to reduce the occurrences of crashes and deaths. By understanding the factors and patterns related to motorcycle crashes, appropriate measures can be implemented to enhance road safety and mitigate the risks associated with motorcycle transportation.

3.1. Study Area

The study was conducted in Bangladesh, a South Asian nation of more than 165 million people [48]. Geographically, Bangladesh is surrounded by Myanmar and India and is located in the fertile Bengal Delta. The study area encompasses all the urban, suburban, and rural regions across the country.

The study area, as depicted in the accompanying map (Figure 1), covers the entire nation of Bangladesh, divided into eight major administrative divisions: Dhaka, Chattogram, Khulna, Barishal, Rajshahi, Sylhet, Rangpur, and Mymensingh. These regions encompass a range of urban, suburban, and rural environments, providing a comprehensive framework for assessing motorcycle-related road traffic accidents. The map highlights the geographical distribution of these divisions, categorizing motorcycle crashes from March 2020 to February 2023. Dhaka and Rajshahi exhibit the highest crash rates, while Sylhet, Barisal, and Rangpur report fewer incidents. In addition, population figures for each division are noted, helping contextualize crash data within the broader regional demographics, thus facilitating a clearer understanding of the study’s findings.

The plot (Figure 2) visualized significant COVID-19 events in Bangladesh from the start of 2020 through 2021, highlighting key moments in the pandemic’s progression. It began with authorities at Dhaka airport beginning screening in January 2020, followed by the evacuation of Bangladeshis from Wuhan in February. The timeline marked the first confirmed COVID-19 case on March 8, 2020, signifying the beginning of the pandemic in the country. This event coincided with the start of the first wave, which lasted until May 31, 2020. The plot also illustrated the implementation and conclusion of a general holiday (lockdown) to curb the virus spread. It continued to track the pandemic through the second and third waves, marked by their start and end dates in 2021, alongside other pivotal moments such as the crossing of 100,000 and 200,000 confirmed cases and the start of the vaccination program in January 2021. The timeline ended with the third wave concluding in August 2021, providing a succinct overview of the pandemic’s key phases in Bangladesh within the given timeframe.

3.2. Data Source

The data used in this study were obtained from the Accident Research Institute (ARI), an organization supervised by the Bangladesh University of Engineering and Technology (BUET). ARI is the most authentic and reliable source of crash data, which has been manually collecting newspaper-based data since January 2016 as a response to the challenges associated with accessing official accident data in Bangladesh. The researchers and data collectors working in ARI first identify the newspapers relevant to RTCs. This includes reliable national newspapers that regularly cover all types of traffic crashes nationwide and have a wide readership. Once relevant articles are identified, data collectors extract key crash-related information. This includes details such as the date of the accident, location (e.g., district, division, road name, and hospital name), type of transport mode (e.g., road traffic, railway, and waterway), type of accident (e.g., head on, rear end, fall from the vehicle, and lost control), vehicles involved, number of injuries or fatalities, and any additional relevant information reported in the article. The extracted information is then recorded in a standardized format. Researchers can analyze the accident data using various statistical techniques once the accident data are compiled. We only extracted the reports from the whole dataset that involved at least one motorcycle, focusing our analysis on motorcycle-related crashes. This approach allowed us to narrow down the dataset and focus on the specific context of motorcycle crashes. The final dataset comprised monthly estimates of total motorcycle RTCs, injuries, and fatalities, which were treated as the dependent variables in our analysis.

The study period for this research encompassed 86 months, from January 2016 to February 2023. This timeframe aligns closely with the period of significant motorcycle registration, as the BRTA reported, with approximately 2.6 million new motorcycles registered between 2016 and 2022 (BRTA, 2023). The inclusion of this specific timeframe allows us to capture the consequences of the substantial growth in motorcycle numbers and their impact on road safety before and during the pandemic.

Included in the supplementary materials is a table of pseudo data (see Appendix Table A1), which emulates the structure of our comprehensive accident dataset. This pseudo dataset includes a variety of fields such as the date of the accident, time of day, district, Thana (subdistrict), accident location, road name, name of the hospital, numbers killed and injured, age of those killed, accident type, and vehicles involved. Each record in the pseudo dataset is formatted to closely resemble the actual data collected, albeit with anonymized and nonspecific data to maintain confidentiality and privacy. The format of this table allows readers to appreciate the breadth and depth of the data collected for this study without revealing sensitive information.

3.3. Statistical Analysis

The specific method that studies the collected data points over a period is called a time series analysis. Since variables may be analyzed to illustrate how they vary over time, time series data differ from other data types in this way. It forecasts data for the future based on past data and ensures that any trends or patterns are not outliers and can consider seasonal variation. One popular time series analysis method is ARIMA. The efficiency and theoretical foundation of ARIMA have made it a popular choice in transportation research, i.e., forecasting and modeling of vehicle emissions [49], air transportation research [50, 51], traffic congestion [52], traffic flow [53], traffic volume [54], and RTCs [55, 56]. In addition, ARIMA has been effectively employed in analyzing intervention-impacted crash patterns [57–60].

Despite the widespread application of the ARIMA model in studying traffic collisions, there is a need for more literature that specifically focuses on utilizing ARIMA to analyze motorcycle crash patterns and trends [61, 62]. The influence of the coronavirus pandemic on the frequency and severity of motorcycle accidents has received very little attention. This study will add to the expanding knowledge based on the diverse impacts of the coronavirus pandemic on worldwide transportation.

For the analysis purpose, we have utilized R statistical software tools. This statistical research aims to determine the time series trend for motorcycle crashes’ frequency, injury, and fatality and compare the situation before and during the pandemic. The 86 months (January 2016 to February 2022) of data were split into two sets: the first 50 months (prepandemic: January 2016 to February 2020) were used for model fitting, and the last 36 months (during the pandemic/intervention period: March 2020 to February 2023) were used for prediction. The projected and actual values were compared to comprehend how the pandemic affected the statistics on motorcycle crashes.

3.4. Development of Time Series Model

This flowchart (Figure 3) summarizes the ARIMA model development process for motorcycle crash data. Starting with data collection and extraction of relevant variables, it proceeds through steps such as plotting trends, ensuring stationarity, and selecting model parameters using ACF/PACF and AIC/BIC. The model is finalized once residuals are confirmed as white noise. Detailed steps are discussed in the following.

ARIMA modeling is a powerful tool for assessing the impact of interruption since it accounts for underlying trends, autocorrelation, and seasonality, as well as allowing for dynamic modeling of numerous impacts [26, 63].

For the purpose of forecasting future trends, in 1976, Box and Jenkin proposed the ARIMA model [64]. This model comprises three parameters denoted by the letters p for the autoregressive term, d for the degree of integration or differencing, and q for the moving average term.

In the time series analysis, the value at time t, represented by y_t, is determined by a combination of factors. These factors include a constant term (δ), autoregressive coefficients (φ_i) that capture the relationship between current and past values, moving average coefficients (θ_j) that account for the influence of past residuals, and uncorrelated residuals (et) that represent the unexplained variability.

When dealing with time series data that exhibit seasonal patterns, such as recurring fluctuations over specific time intervals, a seasonal ARIMA (SARIMA) model can be employed. This model extends the ARIMA (p, d, and q) framework by incorporating seasonal fluctuations denoted as (P, D, and Q). Here, P represents the order of the seasonal autoregressive (SAR) term, D represents the degree of seasonal differencing or integration, and Q represents the order of the seasonal moving average (SMA) term.

4.1. Preliminary Analysis

Over the course of the 86-month observation, which spanned between January 2016 and February 2023, a total of 5771 motorcycle-associated RTCs, 3373 injuries, and 6649 fatalities were reported. The following is the summary of descriptive analysis (Table 1) for the total number of motorcycle-associated crashes, injuries, and fatalities before and at the time of the pandemic. It is important to note that during the pandemic, the minimum value for RTCs, injury, and fatality was not zero, indicating that crashes, injuries, and fatalities occurred every month. In addition, in all cases, the minimum, median, mean, and maximum values were higher during the pandemic than they were before in the pre-COVID phase.

4.2. Exploratory Analysis

Figure 4 displays the basic time series plots of RTCs, injury, and fatality, distinguishing between the pre-COVID and through-COVID timeframes. None of the plots exhibited prominent upward or downward patterns, making it challenging to discern a clear trend. Consequently, we employed the time series decomposition technique to gain further insights.

Figure 5 presents the decomposed plots for RTCs, injury, and fatality. The decomposition plot separates the time series into three key components: trend, seasonality, and random. We identified an exponential growth trend based on the observed time series, indicating a multiplicative time series. In addition, from the seasonal components of the decomposed plots, we observed seasonal trends recurring at a frequency of 12 months.

4.3. Stationarity Test

To ensure the suitability of the data for ARIMA modeling, it was required to confirm that the time series is stationary. To achieve this, we conducted two tests which are the ADF and KPSS tests. The ADF test indicates stationarity when the p value was less than 0.05, while the KPSS test indicates stationarity when the p value was greater than 0.10. The initial analysis revealed that all three series (RTCs, injury, and fatality) were nonstationary according to both the ADF and KPSS tests (Table 2). Therefore, first-order differencing was applied to all three series to achieve stationary time series. Subsequently, conflicting outcomes were observed in the ADF and KPSS test results. While the KPSS test indicated stationarity after first-degree differencing (p value > 0.01), the ADF test suggested nonstationarity (p value > 0.05), suggesting the need for second-order differencing.

4.4. Fitting Time Series Models

Figures 6, 7, and 8 depict the ACF and PACF plots. These plots are constructed for the RTC, injury, and fatality time series before and after applying first-order differencing. These plots were utilized to determine the appropriate degree of differencing needed to achieve stationarity in all of the three time series, as ADF and KPSS test results were contradictory. ACF and PACF plots serve as visual indicators of stationarity in the data. Stationarity is confirmed when the ACF plot exhibits a gradual decrease within the confidence interval, and the PACF plot displays a sharp cutoff. As illustrated in Figures 4, 5, and 6, after applying first-order differencing, the RTC, injury, and fatality time series exhibited the desired features of stationarity. Therefore, no further differencing was required.

The next stage was to choose parameters for the ARIMA model as the data were prepared and satisfied the modeling requirements of being stationary [65]. This has also been accomplished with the help of ACF and PACF plots.

4.5. Model Selection

The ACF and PACF plots were used to assist us in finding some potential models. Also, we incorporated seasonal parameters in the model because of seasonality being discovered in the time series through the decomposed plots. The models were further assessed using the AIC and BIC values, with lower values indicating better fits, as shown in Table 3.

Specifically, for the RTC, injury, and fatality time series, we selected the ARIMA (1,1,1) (1,1,0) [12], ARIMA (0,1,1) (1,1,0) [12], and ARIMA (0,1,1) (1,1,0) [12] models, respectively, as they had the lowest AIC values. When considering the lowest BIC values, we found that the parameters of the chosen RTC model (“ARIMA (1,1,1) (1,1,0) [12] and fatality model (“ARIMA (0,1,1) (1,1,0)[12]”) were significant. However, the injury model (“ARIMA (1,1,1) (1,1,0) [12]”) with the lowest BIC value demonstrated insignificant parameters. Therefore, we selected the ARIMA (0,1,1) (1,1,0) [12] model for the injury time series, which had the lowest AIC value and the second-lowest BIC value. Finally, the parameters of the selected models for RTCs, injury, and fatality time series are statistically significant as presented in Table 4 (p value < 0.05).

4.6. Model Diagnostics

Figure 9 shows residual plots for RTCs, injury, and fatality variables. The residuals display no discernible pattern or substantial autocorrelation and follow a normal distribution. The Ljung–Box test was conducted, yielding p values of 0.81 (RTCs), 0.15 (injury), and 0.17 (fatality) (Table 4). As all p-values exceed 0.05, the null hypothesis of no autocorrelation is not rejected. The absence of evident patterns, the normal distribution of residuals, and the nonrejection of the null hypothesis in the Ljung–Box test support the suitability of the models for forecasting RTCs, injury, and fatality variables.

4.7. Forecasting

The models were fitted for a duration of 50 months, and the subsequent 36 months of the pandemic were projected and compared to the actual values, as presented in Figure 10.

We have delineated the timeframes of three COVID-19 waves experienced in Bangladesh. During the first wave, crashes, injuries, and fatalities decreased significantly due to a 66-day government-imposed lockdown. As the second wave subsided, the number of crashes and fatalities approached predicted levels, while injuries remained notably lower than anticipated. In the third wave, there was an initial sharp decline followed by a rapid increase in crashes, injuries, and fatalities. Interestingly, during the rising phase of the third wave, the number of crashes was lower than predicted, whereas injuries and fatalities exceeded forecasts. In addition, these waves reveal a consistent pattern: in the first and second waves, there is a significant surge in the number of crashes, injuries, and fatalities toward the end of each wave, whereas in the third wave, a similar surge occurs approximately at the midpoint.

Furthermore, the box and whisker plot analysis were conducted to investigate the disparities between the observed and predicted values (Figure 11). The dataset encompassed the pandemic period (36 months) with predicted values derived from the forecasting models and corresponding observed values (testing data). The observed values consistently remained lower than the predicted values for all three variables (RTCs, injury, and fatality). This discrepancy can be attributed to the implementation of various containment measures during the COVID-19 pandemic. The lower medians and shorter whiskers for observed data further support this trend. These findings suggest that the COVID-19 containment measures significantly shaped the observed outcomes of motorcycle crashes, injuries, and fatalities.

6.1. Enhancing Data Collection to Improve Motorcycle Safety in Bangladesh

Collecting comprehensive data on mobility patterns and road user behavior is crucial for understanding crash risks and implementing effective road safety measures. The ITF (2021) emphasizes that analyzing pandemic data helps policymakers discern whether changes in road fatalities result from altered travel behaviors or exposure levels, enabling targeted interventions [73]. In Bangladesh, enhancing data collection will empower authorities to monitor the impact of mobility changes during the pandemic and develop evidence-based policies to significantly improve motorcycle safety outcomes.

6.2. Speed Management

To enhance traffic safety during and after pandemics, a comprehensive speed management program is recommended. This should include focused enforcement, such as deploying speed cameras and increasing police patrols on high-speed roads, along with signs that provide real-time speed feedback to encourage safer driving [73]. Engineering solutions like speed cushions in residential areas can help slow traffic, and public awareness campaigns should target risky speeding behaviors. These measures are expected to reduce speeding-related crashes, safeguard vulnerable road users, and lower fatalities and serious injuries, particularly benefiting motorcyclists who are at higher risk during high-speed incidents.

6.3. Enhancing Rider Education Through Structured Training

Rider education is essential for reducing crash risks, especially among younger and inexperienced motorcyclists. Traditional training often focuses on basic skills but lacks real-world road experience, leaving riders vulnerable. Adopting a system for motorcyclists, which emphasizes supervised road experience and hazard perception, can significantly improve safety. This structured approach helps new riders develop essential cognitive skills while gradually gaining experience [74]. Involving the community in developing these programs ensures riders are well-prepared to handle road risks from the training stage, ultimately enhancing safety outcomes.

6.4. Using Helmets

Amid the COVID-19 pandemic, enforcing the use of full-face helmets for both motorcycle riders and passengers is highly recommended. Properly worn helmets have been proven to significantly decrease the risk of fatal injuries, making their mandatory use a vital measure during this period [75, 76]. With healthcare systems strained by the surge in COVID-19 cases, reducing preventable injuries from motorcycle crashes is essential to alleviate the burden on medical facilities. In the long term, widespread helmet use can lead to sustained improvements in road safety, reduce healthcare costs associated with traffic accidents, and contribute to a safer postpandemic recovery.

Finally, it is essential to acknowledge that our research faced limitations due to the need for more crucial information in the dataset. We did not have access to data regarding the driver’s age, weather conditions (clear/cloudy/rainy), road characteristics (geometric features), impairment from drugs and alcohol, and other relevant factors. This absence of information prevents us from comprehensively understanding the driving and environmental elements involved in the crash. Therefore, we strongly advocate developing guidelines to educate journalists on accurately documenting crash reports with all the necessary information. In addition, our study was conducted at an aggregate level, focusing on Bangladesh. However, since Bangladesh has eight administrative divisions, conducting separate analyses for each division could provide valuable insights into identifying which areas are more vulnerable and require specific attention. Furthermore, it should be noted that this study does not include a detailed exploration of the interactions among various transportation modes. The impact and dynamics of these interactions remain outside the scope of this analysis, representing a significant limitation.