Exploration of the Characteristics of Elderly-Driver-Involved Single-Vehicle Hit-Fixed-Object Crashes in Pennsylvania, USA

Abstract

With the acceleration of population aging, the elderly driving safety issue is increasingly prominent. Method: With the crash data of Pennsylvania from 2010 to 2019, this study exclusively discusses features of single-vehicle hit-fixed-object crashes (SVHFOCs), one of the most common and deadliest crash types for elderly drivers. Results: Firstly, we demonstrate that elderly drivers are much more likely to be injured and killed than young drivers in SVHFOCs by checking crash consequences. The descriptive analysis indicates that elderly drivers have very different crash features from young drivers. They are found to drive with more caution in many aspects, such as more low-speed local travels, fewer illegal behaviors, fewer nighttime travels, etc. Then, a logistic regression model is built to find the factors significantly influencing the severity of SVHFOCs from driver, vehicle, roadway, and environment. The estimation results indicate that female sex, not wearing a seatbelt, DUI, rural area, and SUV involvement tend to be associated with more severe SVHFOCs. Additionally, illumination, weather, and road type could also significantly affect crash severity. Especially, SVHFOCs in adverse weather, in dark conditions, and at intersections are found to be less severe, which implies that elderly drivers might drive more carefully in complex environments. Practical Applications: These findings are expected to provide new insights for agencies in formulating customized measures to prevent elderly drivers from being involved in SVHFOCs.

1. Introduction

With the fast development of the economy and medicine, population aging has been accelerating in the world. According to the United Nations, the elderly (≥65) population proportion was 9% globally, and 15.9% in the United States (U.S.) in 2019 [1]. There are many challenges for elderly people in life, and transportation is the primary one [2]. On one hand, declining physiological and psychological functions can impair their driving capabilities. Considering that elderly Americans travel mainly by driving [3], this is especially challenging. Many older drivers have realized this issue and take proactive actions to reduce crash risks. For example, they are found to prefer short-distance journeys in the daytime [4] and executing fewer turns at intersections [5]. On the other hand, they are also vulnerable in collisions. Several studies have shown that the mortality rate of elderly drivers increases with age in crashes [6,7,8]. Therefore, elderly driver safety has become an increasingly serious traffic safety issue and attracts much attention.

The precondition of formulating effective measures for elderly drivers to prevent crashes is to fully figure out their crash characteristics. Different crashes have different features and causations. Among various traffic crashes, a single-vehicle crash (SVC) refers to a traffic accident involving only one vehicle, which usually hits a fixed object alongside the road, such as tree, power pole, etc., leading to a single-vehicle hit-fixed-object crash (SVHFOC). In the U.S., SVCs accounted for 46% of total fatalities in crashes in 2008 [9], indicating their huge threat to traffic safety. A study on Alabama crashes demonstrates that older drivers are responsible for more SVCs [10]. Due to the vulnerability of elderly drivers, SVCs are especially dangerous for them. An analysis of California SVCs from 2003 to 2004 indicates that elderly drivers account for a greater proportion of fatal accidents than other age groups [11]. Similar findings also can be seen in another study in Singapore [12]. Although these studies have pointed out the high risk of SVCs for elderly drivers, none of them have tried to comprehensively explore features of elderly SVCs. Currently, to the best of authors’ knowledge, only a few studies exclusively discuss elderly SVCs. For example, Amiri et al. used deep learning to predict the severity of the elderly HFO crashes in California [13], but they focused more on discussing the pros and cons of the adopted models, rather than digging deep into characteristics of those crashes. Therefore, these studies might not provide much useful information for agencies to take customized measures to prevent elderly SVHFOCs.

Generally, although SVHFOCs have been found to be a great threat for elderly drivers, there is lack of studies on them. Considering the weakening physiological and psychological functions of elderly drivers, it is necessary to explore their crash features closely. Therefore, with the crash data of Pennsylvania (PA) from 2010 to 2019, this study is designed to comprehensively explore characteristics of SVHFOCs involving elderly drivers, identify the factors influencing their severity, and propose appropriate countermeasures. The following paper is organized as follows: Section 2 introduces the materials used in this study; Section 3 illustrates the methodology; Section 4 discusses the estimation results; Section 5 gives conclusions and discussions.

2. Materials

PA is a state in the northeastern United States with an area of 44,742 square miles and a population of 12.8 million [14]. By 2019, the proportion of the elderly (≥65) population in PA had reached up to 19.0%, much higher than that of the whole U.S. (15.9%). Traffic crash data of PA from 2010 to 2019 were retrieved from the PA Department of Transportation. The raw crash data consist of multiple files, including crash, person, vehicle, roadway, etc., which could be joined by the unique crash record number (CRN). Firstly, SVHFOCs were identified by checking the number of involved vehicles and the collision type. In total, there were 225,762 SVHFOCs. Although they only accounted for 21.9% of total crashes, they led to 29.6% of total fatalities, indicating their great threat. Then, if the driver is aged between 65 and 98, it is thought to be an elderly SVHFOC; if the driver is aged between 25 and 64, it is a young one. Finally, 24,240 elderly SVHFOCs and 201,522 young SVHFOCs were identified.

Table 1. Summary statistics of the SVHFOCs in PA from 2010 to 2019.

Variable	Definition	Elderly Drivers ($\ge 65$)	Young Drivers (25–64)
Severity	PDO	49.3%	59.0%
	Injury	48.4%	39.9%
	Fatal	2.3%	1.1%
Object type	Tree	22.1%	20.9%
	Pole	21.2%	20.6%
	Embankment	14.0%	15.6%
	Others	42.7%	42.9%
Gender	Male	60.3%	64.7%
	Female	39.6%	35.2%
Unbelted	Unbelted in a crash	9.7%	13.8%
Driving Under Influence (DUI)	Yes	6.0%	22.0%
	No	94.0%	78.0%
Speeding related	Yes	23.2%	40.6%
	No	76.8%	59.4%
Vehicle type	Car	58.5%	53.8%
	Truck	14.0%	17.6%
	SUV	20.4%	21.9%
	van	5.4%	4.3%
Vehicle age	Crash year minus model year	Min = 0, mean = 9.2, var = 6.3	Min = 0, mean = 9.7, var = 6.0
Area	Urban	60.1%	59.9%
	Rural	39.9%	40.1%
Road type	State road	63.0%	61.5%
	Local road	28.2%	25.5%
	Interstate	6.8%	10.2%
Intersection	Crash took place at an intersection
	Yes	17.9%	15.7%
	No	82.1%	84.3%
Curved roads	Crash happened on a curved road
	Yes	33.3%	39.4%
	No	66.7%	60.6%
Illumination	Bright	77.8%	53.8%
	Dark	22.2%	46.2%
Weather	Sunny	74.5%	67.0%
	Rain	14.2%	17.1%
	Snow	9.3%	13.4%

shows the summary statistics of those crashes. To check whether elderly drivers have different traffic safety features from young drivers, their SVHFOCs were compared from multiple perspectives.

Crash severity is the primary concern for traffic safety analysis. Here, the proportions of injury crashes and fatal crashes are 50.7% and 2.3% for elderly SVHFOCs, but only 41.0% and 1.1% for young ones. That is, elderly drivers seem much more likely to be injured in SVHFOCs. Especially, the fatality rate of elderly SVHFOCs is more than 2 times that of young ones. This might be mainly attributed to their fragile bodies. Regarding the collision object, the most common fixed objects are trees, poles, and embankments. Among them, trees are the most dangerous (22.1%), followed by poles (21.2%) and embankments (14.0%). The casualty rate (61.6%) of vehicles driven by elderly drivers hitting trees is the highest, significantly higher than that of young drivers (49.9%). Meanwhile, the accident fatality rates caused by collisions with embankments (58.2%) and poles (57.1%) are significantly higher than those of young drivers (47.5% and 44.4%). In addition, car accidents are highly likely to have serious consequences, with the mortality rate of elderly drivers almost twice that of young drivers. It can be seen that SVHFOCs pose a huge threat to elderly drivers.

2.1. Driver Features

Sex often exhibits great impacts on driver behaviors. Here, male drivers account for 60.3% of elderly SVHFOCs, slightly lower than with young ones. Age could also influence driving behaviors. Figure 1 shows the distribution of the elderly SVHFOCs by their age. It indicates a linearly decreasing trend, which is thought to be reasonable. With the increase in age, the elderly driver population decreases, and their driving activities are also expected to reduce.

2.2. Vehicle Features

Vehicles also have great effects on crash consequences. Cars, SUVs, vans, and trucks are the four main vehicle types for SVHFOCs. Figure 2 shows the compositions of elderly SVHFOCs by vehicle type per year. It can be found that the SUV proportion increased quickly from 14.8% in 2010 to 25.2% in 2019, while the car proportion decreased from 65.0% in 2010 to 53.6% in 2019. This finding is consistent with the increasing trend of SUVs in the U.S. auto market [15]. Although SUVs might be more maneuverable, they are also taller and prone to have rollover accidents, posing a greater risk to drivers [16]. Here, SUVs are only involved in 20.4% of elderly SVHFOCs, but account for 29.4% of rollover collisions.

Vehicle age might also influence crash risks. Old vehicles are thought to be riskier. On one hand, vehicle parts would be worn with aging, impairing their functions. On the other hand, old vehicles are equipped with fewer advanced driver assistance systems (ADASs), such as lane departure warning, blind spot detection, etc., which have been proved to be greatly helpful in collision prevention [17]. Here, the average vehicle age is 9.2 years for elderly SVHFOCs, slightly smaller than that of young SVHFOCs (9.7 years). Figure 3 shows the distribution of vehicle usage years for elderly and young drivers. It can be observed that elderly drivers drive vehicles that are newer compared to young drivers, which might be mainly attributed to their higher disposable incomes.

2.3. Roadway Features

Roadway might also influence traffic safety in many aspects. Here, SVHFOCs mainly occurred in urban areas, and most SVHFOCs occurred on state roads, followed by local roads and interstates. Especially, 10.2% of young SVHFOCs occur on interstates, but this is only 6.8% for elderly ones, implying that elderly drivers might drive much less on interstates. Additionally, regarding roadway geometry, 33.3% of elderly SVHFOCs occur on curved roads, obviously smaller than 39.4% of young ones. As is known, interstate roadways have big traffic volumes and higher speed limits, and vehicles are easily run off roads on curves, both of which are challenging for elderly drivers. Therefore, they might tend to avoid traveling in those complex scenarios to avoid crash risks.

2.4. Environment Features

With aging, visual functions of the elderly often decline quickly due to illnesses, such as presbyopia, glaucoma, etc. Therefore, nighttime driving is always a big challenge for elderly drivers. Here, 22.2% of elderly SVHFOCs occurred in dark conditions, less than half of that of young ones, which verifies that elderly drivers travel much less at nighttime somehow. Furthermore, Figure 4 shows the distributions of SVHFOCs over time of day. Elderly SVHFOCs are concentrated in the daytime with one peak at noon, whereas young SVHFOCs are evenly distributed, with an AM peak and a PM peak. This implies that elderly drivers also avoid travelling at peak hours. Similarly, adverse weather could also make driving risky. Here, 13.5% and 9.3% of elderly SVHFOCs occurred on rainy days and snowy days, respectively, both of which are obviously smaller than those of the young ones (16.3% and 13.4%). These findings confirm that elderly drivers prefer to not drive in risky environments, such as during nighttime and in adverse weather [18]. Even when driving in adverse weather, they are inclined to be more cautions [19]. It should be noted that most elderly Americans are expected to be retired, which is the precondition for them to travel more flexibly to avoid risky environments.

2.5. Summary

Generally, elderly SVHFOCs and young SVHFOCs show some big differences in many aspects, including crash outcome, driving behaviors, environment, etc., which demonstrates that elderly drivers and young drivers might have very different traffic safety approaches. Since most existing studies target young drivers, their conclusions might not work for elderly drivers. Therefore, it is necessary and important to exclusively explore the traffic safety of elderly drivers to obtain more refined results.

3. Methods

Due to the great threat of SVHFOCs to elderly drivers, it is important to figure out what factors contribute most to their severity. Due to the small proportion of fatal crashes, here, they are merged with injury ones. Therefore, all elderly SVHFOCs were reclassified into two types by severity: PDO and casualty. Then, a logistic regression model was built to identify the important factors influencing the crash severity in R [20]. The model is shown as follows.

$$P\left(Y=1|X\right)={\displaystyle \frac{\mathrm{e}\mathrm{x}\mathrm{p} ({\beta }_0+\sum _{i=1}^n{\beta }_i{x}_i)}{1+\mathrm{e}\mathrm{x}\mathrm{p} ({\beta }_0+\sum _{i=1}^n{\beta }_i{x}_i)}}$$

(1)

where $Y$ is the binary dependent variable, with 1 for casualty (Casualty) and 0 for PDO;

$n$ is the number of independent variables;

$X=(x_1,x_2,\dots ,x_n)$, is the independent variable vector;

${\beta }_0$ is a constant term, and ${\beta }_i$ are the regression coefficients.

To ensure the accuracy of the modeling results, the data were processed before modeling. Crashes with unknown variables were removed first. Then, some variables were reorganized to show their features more accurately. Finally, 22,276 crashes were retained for regression analysis, and they occupy 91.9% of the raw data.

Table 2. Summary of elderly SVHFOCs for regression analysis.

Variable	Definition	Proportion
Dependent
Crash Severity	1 for Casualty 0 for PDO	51.4%
		48.6%
Independent
Gender	0 for Female (Baseline)	40.8%
	1 for Male	59.2%
Driver Age	Min = 65, mean = 74.1, var = 73
	0 for Belted (Baseline)	89.9%
	1 for Unbelted	10.1%
DUI	0 for No DUI (Baseline)	93.9%
	1 for DUI impaired	6.1%
Speeding	No (Baseline)	77.1%
	Yes	22.9%
Vehicle type	Car (Baseline)	59.8%
	Truck	14.0%
	SUV	20.8%
	Van	5.4%
Vehicle age	Min = 0, mean = 9.2, var = 6.3
Area	0 for Urban (Baseline)	40.4%
	1 for Rural	59.6%
Roadway type	State Road (Baseline)	64.1%
	Local Road	28.8%
	Interstate	7.1%
Intersection	No (Baseline)	81.9%
	Yes	18.1%
Curved roads	No (Baseline)	67.1%
	Yes	32.9%
Illumination	Bright (Baseline)	78.1%
	Dark	21.9%
Weather	Sunny (Baseline)	75.9%
	Rain	14.4%
	Snow	9.7%

shows the summary statistics of these crashes.

The estimation results of the logistic regression model are shown in

Table 3. The estimation results of the logistic model with all variables.

Variable	Estimate	Standard Error	Z Value	Pr (>|z|)	VIF
Intercept	−1.006	0.155	−6.482	<0.001 *	-
Gender—Male	−0.171	0.030	−5.751	<0.001 *	1.059
Driver Age	0.015	0.002	7.560	<0.001 *	1.042
Unbelted	1.420	0.055	25.872	<0.001 *	1.010
DUI	0.202	0.061	3.300	<0.001 *	1.042
Speeding	0.021	0.036	0.573	0.566	1.100
Vehicle type—Truck	−0.035	0.044	−0.798	0.425	1.028
Vehicle type—SUV	0.162	0.036	4.538	<0.001 *	1.028
Vehicle type—van	−0.085	0.062	−1.356	0.175	1.028
Vehicle age	0.003	0.002	1.156	0.248	1.020
Area—rural	0.096	0.030	3.220	0.001 *	1.063
Roadway—Interstate	−0.094	0.055	−1.711	0.087	1.018
Roadway—Local Road	−0.094	0.032	−2.956	0.003 *	1.018
Intersection	−0.140	0.037	−3.781	<0.001 *	1.031
Curved roads	0.031	0.030	1.038	0.299	1.032
Illumination—Dark	−0.314	0.035	−9.033	<0.001 *	1.035
Weather—Rain	−0.259	0.041	−6.321	<0.001 *	1.040
Weather—Snow	−0.528	0.050	−10.508	<0.001 *	1.040

Note: *, significant at 95% confidence interval.

. All the variance inflation factors (VIFs) are far less than 5, which indicates weak correlations between independent variables. Since many variables are insignificant, the model was rebuilt with significant variables only, and the estimation results of the new model are shown in

Table 4. The estimation results of the logistic model with significant variables only.

Variable	Estimate	Standard Error	Z Value	Pr (>|z|)	VIF	Odds Ratio
(Intercept)	−1.026	0.151	−6.775	<0.001 *	-	0.358
Gender—Male	−0.179	0.028	−6.280	<0.001 *	1.017	0.836
Driver Age	0.016	0.002	7.926	<0.001 *	1.026	1.016
Unbelted	1.427	0.055	26.097	<0.001 *	1.006	4.166
DUI	0.212	0.061	3.474	<0.001 *	1.040	1.236
Vehicle type—SUV	0.171	0.034	4.989	<0.001 *	1.008	1.186
Area—Rural	0.099	0.029	3.380	<0.001 *	1.046	1.104
Roadway—Local Road	−0.083	0.031	−2.666	0.008 *	1.025	0.920
Intersection	−0.137	0.037	−3.730	<0.001 *	1.026	0.872
Illumination—Dark	−0.313	0.035	−9.036	<0.001 *	1.035	0.731
Weather—Rain	−0.250	0.040	−6.294	<0.001 *	1.008	0.779
Weather—Snow	−0.519	0.048	−10.778	<0.001 *	1.008	0.595

Note: *, significant at 95% confidence interval.

.

Table 5. The likelihood ratio test of the two models.

Model	AIC	#Df	Loglik	Df	Chisq	p-Value
Model 1—All variables	29,630	18	−14,797
Model 2—Significant variables only	29,627	12	−14,801	−6	8.421	0.209

shows that the AIC values of the two models are very close, and the likelihood ratio test also shows no significant difference, which confirms the credibility of the parsimonious model. Therefore, all the following analyses are conducted based on the estimation results of the new model in Table 4.

4. Results

4.1. Driver Features Analysis

The estimated results show that both driver gender and driver age could significantly affect crash severity. Although males are more likely to be involved in elderly SVHFOCs, they are 16.4% less likely to be injured than females. Many reasons might contribute to this result. Firstly, females are generally lighter and smaller, and thus more likely to suffer severe injuries with the same protective measures [21]. Secondly, women are increasingly engaging in risky driving behaviors (e.g., alcohol-related), which can lead to more severe accident consequences [22]. Finally, women are more likely to drive smaller, less safe cars, and are therefore more likely to be injured in an accident [23]. Driver age shows significantly positive effects. For every year of age, elderly drivers were 1.6% more likely to be injured in SVHFOCs. As is known, elderly people are frailer and more vulnerable [24] and thus are more likely to suffer severe damages in collisions. The finding confirms the presence of heterogeneities among elderly drivers over age again.

In terms of risky driving behaviors, not wearing seatbelts and DUI show significant positive effects. Here, unbelted elderly drivers are found to be four times more likely to be injured than those belted, which is understandable. Former studies have shown that unbelted drivers have more risks of head-on collisions with the steering wheel in SVHFOCs, which may fracture their ribs and cause severe damages to their chests [25,26,27], especially dangerous for elderly drivers. DUI drivers are found to be 23.6% more likely to be injured than sober ones, which is also thought to be reasonable. DUI can cause cognitive impairments to drivers, making it difficult to effectively control vehicles, which is especially dangerous for long-term alcoholic drivers [27]. However, speeding has insignificant effects. This may be attributed to the fact that the elderly generally drive cautiously. Although more than 20% of elderly SVHFOCs involve speeding, further investigation indicates that those crashes occur mainly on low-class roads with lower speed limits. That is, their real driving speeds might not be very high, which lowers the severity of collision consequences.

4.2. Vehicle Features Analysis

For vehicle type indicators, only SUVs show significant positive effects, whereas trucks and vans show insignificant effects. Compared to cars, SUVs are 18.6% more likely to lead to injuries in elderly SVHFOCs. This finding is inconsistent with conclusions of former studies, which usually think that SUVs could provide more protection to drivers [28]. It is thought that those studies usually analyze crashes involving two or more vehicles, rather than SVHFOCs. For SVHFOCs, most of them are caused by improper maneuvers, and the high gravity center would easily cause SUVs to roll over [29], thus leading to severe outcomes. However, as shown above, SUVs are increasingly popular among the elderly, which means a bigger traffic safety threat. Therefore, special education and training for elderly drivers on how to drive SUVs safely seems necessary and important. In terms of trucks, although they could protect drivers better regarding their larger size and sturdy structure, they might also lead to more severe outcomes due to their bigger mass. In addition, older truck drivers could detect and adjust their driving ability in a timely manner and drive more cautiously [30]. As to vans, they are usually driven slower as they accommodate many occupants, mostly family members.

Vehicle age had no significant effects. As shown in Figure 3, most vehicles are younger than 15 years, and their average age is only 9.2 years, implying that their technical performance has not significantly deteriorated. At the same time, the US government’s federal side impact protection regulations, which came into effect in 2010, require vehicles to be equipped with side airbags, and since 2011, it has been mandatory for all passenger cars to be equipped with an ESC (Electronic Stability Controller) [31], effectively reducing the probability of serious injuries to elderly drivers in accidents. It is thought that although older vehicles have many disadvantages, the rich driving experiences of elderly drivers might offset their weakness somehow in the collision.

4.3. Roadway Features Analysis

Area shows significantly positive effects on crash severity. The odds of elderly drivers being injured in rural areas was 10.5% higher than in urban areas. Compared with urban environments, rural roads are characterized by higher speed limits, longer police response time [32], etc., which might increase the severity of crashes [33].

In terms of road type, local roads show significantly negative effects, whereas interstates show insignificant effects. Compared to state roads, the odds of people in elderly SVHFOCs being injured is 8.0% lower than that of local roads. This should be mainly attributed to their different travel speeds. As mentioned above, compared to state roads and interstate highways, local roads have much lower speed limits. Therefore, vehicles are expected to drive slowly, which reduces the risk of severe consequences. At the same time, the odds of the elderly SVHFOCs being severe are 12.8% larger at segments than at intersections. Compared to roadway segments, vehicles usually run slower at intersections due to the presence of conflicting traffic, which might reduce the crash severity. The curved road has no significant effects either. It is thought that although it is easy to have crashes at curves, it may also prompt drivers to drive more cautiously. Studies have shown that when facing a complex driving environment on a curve, drivers would avoid traffic collisions by reducing their speed and paying attention [34], which might mitigate crash consequences.

4.4. Environment Features Analysis

Illumination shows significant negative effects, with the coefficient being −0.313. That is, for elderly drivers, the odds of suffering injuries in the dark were 26.9% less than that in bright scenarios in SVHFOCs. Driving in the dark is always a big challenge for elderly drivers. Thus, they are expected to drive more cautiously, which might reduce the crash severity instead. Similar findings are observed for adverse weather. Both rain and snow show significantly negative effects, with the coefficients being −0.250 and −0.517, respectively. Therefore, for elderly drivers, the odds of being injured on rainy and snowy days are 22.1% and 40.4% lower than that on sunny days in SVHFOCs, respectively. Adverse weather is another big challenge for elderly drivers, and they are also expected to drive with more caution in adverse weather [35].

5. Discussion

Based on descriptive analysis and regression analysis results, some countermeasures are suggested to prevent SVHFOCs to improve elderly driving safety in the future.

Firstly, many elderly drivers are found to be still engaged in risky driving behaviors, such as not wearing a seatbelt and DUI, which could significantly aggravate crash severity. Therefore, regular driving safety education and training seem still very necessary for elderly drivers, although they might have rich driving experience. Considering the declining physiological and psychological functions of elderly drivers, agencies may consider stricter law enforcement for elderly drivers in terms of risky driving.

Secondly, SUVs are found to be increasingly popular among elderly drivers, but also much riskier than passenger cars regarding crash consequences. Aiming at this issue, on one hand, automakers should optimize the structures of SUVs to lower their center of gravity to prevent roll-overs once collisions occur. Additionally, they may install SUVs with advanced electronic control systems, such as traction control systems and torque vector control systems, to allocate engine power to wheels in a more reasonable and balanced way. On the other hand, agencies should consider developing some free training programs to help elderly drivers identify features of SUVs quickly and learn how to drive SUVs safely.

Thirdly, aiming at SVHFOCs, some advanced driver assistance systems (ADASs), including forward-collision warning systems and lane-departure warning systems, have been proved to be able to effectively prevent vehicles running off roads. Automakers should consider improving these ADASs to be compatible with physiological functions of elderly drivers to make them more acceptable. Since these ADASs are generally optional and expensive, agencies might consider providing some subsidies to encourage elderly drivers to adopt them. At the same time, agencies should also improve roadway safety facilities, such as installing guardrails and flashing lights to help elderly drivers recognize fixed objects in advanced to avoid collisions [36], and installing crash cushions and barriers to mitigate crash outcomes.

Finally, crashes are found to be more severe with age for elderly drivers, which implies the presence of heterogeneities among them. However, most existing studies consider elderly drivers homogenous, which might impair the effectiveness of the proposed safety measures. In the future, age-customized measures should be developed for elderly drivers to protect them precisely and effectively.

6. Conclusions

As population aging is accelerating, elderly driver-involved traffic accidents have also been increasing quickly in the world. Identifying their characteristics and trends is essential for agencies to formulate effective countermeasures. With SVHFOC data in PA from 2010 to 2019, elderly drivers are proved to be more likely to be injured and killed in crashes than young drivers. Additionally, they are found to have very different features from young drivers in many other aspects. Then, a logistic regression model is built to identify the factors significantly influencing the severity of elderly SVHFOCs. Female, DUI, and SUV-involved SVHFOCs are found to be more severe, whereas SVHFOCs occurring in adverse weather and dark conditions are found to be less severe, implying that elderly drivers might be more cautious in risky driving scenarios [37]. Based on these findings, some countermeasures have been proposed to improve driving safety for elderly drivers.

Although our study provides many new insights regarding elderly driver safety, it still has some limitations. Future studies could be conducted in many aspects to further investigate this issue. Firstly, the driving abilities of elderly drivers are thought to be heterogenous over age, which, however, is not considered here. Future studies may further investigate whether and how elderly SVHFOCs would change by age, and then formulate appropriate measures which precisely adapt to the physical and psychological functions of different ages. Secondly, elderly people are very likely to be involved in many illnesses, which could heavily influence their driving behaviors. However, the illness information of those elderly drivers is unavailable in the data due to the issue of privacy. Future studies may consider taking the impact of illness into account if it is available, especially for common chronic illnesses. Thirdly, to formulate precise countermeasures, it is also essential to determine what exactly happens before SVHFOCs occur, which might not be available just by checking crash data. Naturalistic driving data have been proved to be greatly helpful in reproducing crash scenarios accurately [38,39]. Future research can consider collecting the natural driving data of elderly drivers to deeply analyze crash causes.