Introduction

Traffic accidents are major costs to our collective social and economic well-being. According to the Federal Highway Administration (FHWA 2015), in the United States, roadway accidents are the leading cause of death, and the total estimated cost of roadway accidents is more than $230 billion annually. Despite substantial efforts towards implementing preventive solutions for accidents, they still remain a serious problem (Plug et al. 2011). This problem becomes even more challenging and complex when aging people are considered, as they are more vulnerable to traffic accidents than other adult age groups due to their cognitive, behavioral, and health limitations (‘the aging population’ can be thought of as those people aged 65+). This is especially critical for the State of Florida as it leads the nation with having 18.2 % of its population over 65 years of age (U.S. Census 2010). This percentage will continue to grow as the U.S. Census Bureau (2010) estimates that the number of people 65+ will double in Florida by 2020. As aging people become a greater share of the population, the number of aging-involved accidents will also be more than those presently reported. Therefore, it is important for transportation system planners to understand the spatial and temporal patterns of aging-involved traffic accidents, which can help with implementing more effective preventive measures to reduce fatalities and injuries.

Given the high risk of death and injuries posed by roadway accidents, geographical information systems (GIS) has emerged as a key tool for identifying roadways associated with high accident risk and providing visual confirmation of accident clusters on maps. However, to the authors’ knowledge, there has not been any previous research done specifically focusing on the spatial and temporal patterns of accidents involving aging people. With this motivation, this exploratory study focuses on identifying and analyzing the locations and times that pose a high risk of accidents for aging populations using GIS-based spatial, temporal and spatio-temporal methods. This is the key contribution of the paper, which is an important first step towards understanding these dynamics. Following a review of the existing literature, the following methods are implemented: (a) Spatial analysis with kernel density estimation (KDE) and density ratio (DR) analysis, (b) Temporal analysis using spider graphs, and (c) Spatio-temporal analysis with the comap method. This is followed by a case study application of the Escambia, Hillsborough and Pinellas county roadway networks of Florida, for which high accident risk locations and clusters (hotspots) are identified. The cities of Pensacola, Tampa and St. Petersburg are located in these counties, respectively. These urban counties are selected due to their high numbers of aging-related accidents with respect to their population. These counties have also been identified by other local and state advocacy groups as having a high number of aging population-related accidents (e.g. by the Safe Mobility for Life Coalition Program of Florida) (Sandler et al. 2015).

This study is an important first step towards understanding the spatial and temporal patterns of accidents involving aging people. With the proposed GIS-based approach, transportation systems planners can have a means of analyzing and evaluating aging-involved accidents based on the location and times of these accidents. This approach can help decision makers with the implementation of aging-focused accident mitigation strategies. The following section presents a review on the related literature.

Literature Review

Over the last two decades, the high accident risk for aging people has been a growing concern since that population is rapidly expanding. This is especially critical for the State of Florida, where the population aged 65+ had a 16 % increase between 2000 and 2010, and their population is expected to double by the year 2020, as estimated by U.S. Census (2010). With this increasing 65+ population, statistics also show that the accident rates projected to 10,000 licensed drivers for the 65+ age group increased by 30.8 % from 2008 to 2011 (Florida Department of Highway Safety and Motor Vehicles 2013). These statistics make to the case for focusing specifically on aging-involved accidents. Several researchers have recognized this need; however, to the authors’ knowledge, most of these studies focus only on the following issues that influence the accidents rather than discerning broader trends in where accidents might actually be occurring: (a) significant factors without a specific focus on aging people, and (b) aging-related behavioral issues (e.g. cognitive ability) (Abdel-Aty et al. 1998; Abdel-Aty et al. 1999a; Abdel-Aty et al. 1999b; Abdel-Aty and Radwan 2000; Preusser et al. 1998; Hu and Baker 2010; Centers for Disease Control and Prevention 2007; National Center for Health Statistics (NCHS) 2007; Dissanayake and Lu 2002; Braitman et al. 2007; Boufous et al. 2008; Cook et al. 2000; McGwin and Brown 1991; Williams and Shabanova 2003; Alam and Spainhour 2008; Guoqing and Baker 2010). Therefore, there is a clear need for research focused on the spatial and temporal distribution of aging-involved accidents.

Traffic accidents on the roadways exhibit distinct spatial and temporal patterns, and are said to form clusters in geographical space and time (Black 1991). Extensive knowledge and analysis of these patterns is therefore critical for developing appropriate accident prevention strategies. According to McGuigan (1981), the relationship between traffic accidents and roadway attributes gives us one identification method of the locations which result in more accidents, namely ‘hotspots’. These hotspots can be determined spatially, temporally and spatio-temporally using different methods. In the following sections, we review several existing spatial, temporal, and spatio-temporal methodologies that may be useful to this end.

Spatial Analysis

Any statistical technique that incorporates spatial data characteristics, such as coordinate geometry or other geographic properties such as adjacency, falls under the broad rubric of ‘spatial analysis’ methods. Many spatial analysis methods have been applied in GIS to identify ‘hotspots’ (i.e. locations with an unusually high occurrence of a particular phenomenon), across a range of disciplines. Several examples include the following: (a) crime-related hotspots (Nakaya and Yano 2010; Turnbull et al. 2000), (b) disease hotspots (Kulldorff et al. 2005; Song and Kulldorff 2003; Jarrahi et al. 2007; Molina et al. 2013). Similarly, GIS is also used to analyze roadway accidents in order to identify high risk locations, hotspots and/or clusters. Previously, hotspots have been evaluated in order to determine possible deficiencies in roadway design (Dai et al. 2010; Steenberghen et al. 2010; Plug et al. 2011; Larsen 2010; Pulugurtha et al. 2007). A number of studies have also been conducted to identify the accident hotspots using GIS techniques in order to examine pedestrian-, truck-, young driver-, and weather-related accidents (Dai 2012; Abdel-Aty et al. 2007; Khan et al. 2008; Huang et al. 2010; Siddiqui et al. 2011; Shalini and Geetam 2013). But none have explored the patterns of accidents associated with aging adults.

There are a variety of spatial analysis methods found in the literature including kernel density estimation (KDE), Getis-Ord (Gi*) statistics, K-means clustering, nearest neighbor methods, and others which can potentially be used to identify accident hotspots (Larsen 2010). This study will utilize one of the more popular approaches for spatial hotspot analysis, namely kernel density estimation (KDE). KDE is fundamentally a density estimation technique which uses a distance-based control for analyzing a point dataset, and it creates a density surface for each point (i.e., accident occurrence). For each individual point, a Kernel density surface is defined with the highest value at its location center, the density value decreases as one moves away from the center, and finally becomes zero after it reaches a pre-specified radius. Conceptually, each of these individual density surfaces are added to create a continuous smooth curved surface across the entire study area (Silverman 1992).

For the KDE-based spatial approach, distances between two accident locations can be calculated in two distinct ways: (a) Euclidean (planar) distance (ED), (b) Roadway network distance (RND). The first one uses a Euclidian distance measure while analyzing the traffic accidents (Pulugurtha et al. 2007; Flahaut et al. 2003; Shalini and Geetam 2013). This approach calculates the accident density in a circular window moving across the study area, and the accidents inside the window area are weighted based on their Euclidean (planar) distances from the center where the density values are assigned. This produces a biased estimation by allowing the estimation of the Kernel density values anywhere in the plane. With the RND approach, accidents are weighed based on their network distances along the roadway, not by the Euclidean distance. That is, this approach assumes that the accidents occur on or alongside the roadways. The RND approach was applied on GIS-based roadway networks by Okabe et al. (1995, 2006, 2009) as an improvement to the ED method, and also in order to overcome the biased estimation encountered when the ED method is used for roadway networks. For this purpose, Okabe et al. (2006) created a GIS-based application named as SANET. For a more detailed discussion on the SANET tool, please see Okabe et al. (2009). Recently, several researchers employed this approach in order to calculate the kernel density distributions for accidents (Larsen 2010; Xie and Yan 2008). Comparisons between ED and RND were also presented while analyzing accident hotspots for the whole population (i.e. all driving ages) by Mohaymany et al. (2013) and Dai (2012). A recent paper (Lamb et al. 2015) studied the network K-functions using network distance measures for the following types of networks: the actual roadway network, topologically correct network, and directionally constrained network. They clearly stated the overestimation of the planar K-function when network constrained events such as accidents are being clustered. These studies show the relatively better performance of the RND approach over the ED approach, while estimating the accident distributions more accurately. Note that, since the choice of bandwidth has a significant effect on the density estimation, studies suggest using a 100–300 m bandwidth (Okabe et al. 2009), which can be taken as a rule of thumb when applying the methodology especially in the urban areas.

Temporal Analysis

The majority of existing studies use spatial clustering for the analysis of traffic accidents. Few researchers focus on investigating the temporal patterns inherent to accident data sets, which are also arguably as critical as spatial patterns (Li et al. 2007; Plug et al. 2011). Temporal analysis concentrates on units of time, such as hourly, daily, monthly, and yearly measures, in order to visualize and understand accident rates and frequencies (Plug et al. 2011). There are several ways of capturing the temporal dimension such as line graphs, bar graphs and spider plots. Several recent studies discuss the usefulness of spider graphs illustrating the chronological nature of the temporal data, highlighting the temporal hotspots, and providing a better visualization and understanding of accident variation over time (Plug et al. 2011; Li et al. 2007; Asgary et al. 2010).

Spatio-temporal Analysis

Roadway accidents are phenomena which require a keen understanding of both the temporal and spatial components simultaneously. For example, when two or more accidents occur at a close spatial proximity but differ in their timing, they may not likely represent a significant cluster. Likewise, two accidents that occur in the same time period but differ spatially are also not likely to represent a significant cluster. Hence, solid knowledge of both the spatial and temporal dimensions are necessary for effectively understanding hotspots, and ultimately designing better accident mitigation strategies. However, relatively few researchers have concentrated on the spatio-temporal analysis for analyzing the accident clusters (Wang and Abdel-Aty 2006; Prasannakumar et al. 2011; Plug et al. 2011; Dai 2012).

Among the few currently available approaches, the comap method is one of the most effective methods for dealing with spatial and spatio-temporal information simultaneously (Brunsdon et al. 2007). This method can be used to explore accident data at ordered time intervals such as the hours of a day, days of a week or months of a year. It works by first sub-dividing the accident data according to the time of occurrence based on a chosen interval, which is then followed by application of a spatial analysis technique such as kernel density estimation (KDE). Results are presented in sequential order to highlight the changes over time in order to identify the variation in the spatial distribution of accidents over time. Plug et al. (2011) employed the comap method for investigating single vehicle accident patterns. They visually showed how accidents vary over space at different hours of a day. Several other studies (Asgary et al. 2010; Kilamanu et al. 2011) also used the comap method to successfully illustrate how the spatial distribution of accidents varies with time. Due to its simplicity of application, the comap method is commonly preferred over other approaches.

In sum, this review reveals a clear gap in terms of identifying accident hotspots associated with aging populations. As a result, there is a need for work on aging-involved accidents with an emphasis on identifying high risk locations. From a policy perspective, this type of research can help officials develop appropriate safety measures that can possibly reduce the number of aging-related accidents and fatalities. The next section provides information on the study area and data used for the proposed GIS-based implementation.

Study Area and Data Collection

The main objective of this paper is to use GIS-based spatial and temporal methods to help us better understand aging-involved roadway accidents. Unlike previous approaches found in the literature, the proposed approach focuses specifically on the accidents that involve aging populations with an emphasis on identifying clusters and hazardous locations. By the hazardous location analysis, we mean identifying the aging-involved accident hot spots and clusters on the roadway network. The following subsections focus on describing the study area and the data needed for the analysis.

Study Areas

Given the demographic trends and projections of the aging populations in Florida, the number of aging road users and accidents involving aging people on Florida roadways will increase, which makes studies on aging-involved accidents all the more critical. This study focuses on three urban counties in Florida, namely Escambia, Hillsborough and Pinellas counties for the aging-involved accident analysis. These counties include the cities of Pensacola, Tampa and St. Petersburg, respectively. Documentation shows they have high accident rates with respect to the age 65+ population, and they are also among the top urban counties at risk for aging-involved accidents in Florida, as identified by Florida’s Safe Mobility for Life Coalition Program (Sandler et al. 2015).

Figure 1 shows the locations as well as the population count and percentage demographics of those 65 years and older for the selected three counties, Escambia, Hillsborough and Pinellas, respectively (U.S. Census 2010). Figure 1 clearly indicates that these counties have high number of aging (65+) populations when compared to the overall population, and therefore deserve extra attention. Please note that count and percentage maps are created based on a standard deviation classification method, where each class indicates a different standard deviation range. Therefore, different colors on the maps show the variation with respect to the mean value of the overall 65+ population. That is, if the standard deviation value is higher (darker areas on the maps) for a certain population block, that population block has higher number of 65+ people with respect to the 65+ population average in the whole county.

Fig. 1
figure 1

County characteristics: a Locations of counties in the State of Florida, b Variation of 65+ population counts in the counties, and c Variation of 65+ population percentages in the counties

Full size image

Roadway Network and Accident Data

Roadway network and accident data for the study were obtained from the Florida Department of Transportation Safety Office in the format of GIS shape files and their respective databases, including 5 years of data between 2008 and 2012. Based on these databases, a comparative illustration of yearly accidents is presented for the selected three counties in Fig. 2. Figure 2 shows that accidents for the age group 50+ and 65+ tend to follow almost the same patterns. This is significant since the aging of the Baby Boom generation (people born between 1946 and 1964) is expected to produce a 79 % increase in the age 65+ population in the next 20 years (Koffman et al. 2010). This will consequently increase the accident risk faced by the aging.

Fig. 2
figure 2

Age-based evaluation of accidents per year for a Escambia county, b Hillsborough county, c Pinellas county

Full size image

In order to compare the accident hotspots with respect to aging populations and other age groups, a comparative density ratio approach is introduced in the next subsection.

Density Ratio Approach

In order to understand how accidents involving aging people occur at different locations than those of other age groups, a comparative accident density analysis is introduced through the development of a “density ratio” (DR) parameter. DR is simply the difference between the maxima-normalized accident densities for two different age groups being studied. The formula of the DR metric is shown as follows:

$$ D{R}_{ij}=\frac{D_i}{ \max \left({D}_i\right)}-\frac{D_j}{ \max \left({D}_j\right)} $$
(1)

where \( {DR}_{ij} \) is the “density ratio” between the compared maps i and j whereas \( {D}_i \) and \( {D}_j \) are the density values of the corresponding pixels, and \( max\left({D}_i\right) \) and \( max\left({D}_j\right) \) are the maximum pixel density values of the compared maps, respectively.

Using this concept, accident density raster maps of different age groups (16–19, 20–64, and 65+)) are created and normalized according with respect to the maximum density value on each raster map. Based on this normalization process, every density value on the maps is between 0 (minimum density value on the map) and 1 (maximum density value on the map). Next, accident density raster maps for the 16–19 and 20–64 age groups are individually subtracted from the accident density raster map of 65+ drivers. Subtracting the density values of the different maps reveals the key accident locations on the road network. If the final value is close to zero, then that specific location does not reflect a change between different age group-involved crashes.

Using the available data sets, spatial and temporal analyses will be conducted with the ArcGIS 10 software in order to determine the locations and time periods exhibiting high accident risk for aging populations. Based on the evaluation of these techniques, several important conclusions are reached with regards to the spatial and temporal patterns of aging-involved accidents. Results of this analyses will be presented in the next section.

Case Study Application Results

In this section, we present a case study application on the Escambia, Hillsborough and Pinellas counties of Florida where we aim to identify the spatial, temporal and spatio-temporal distributions of aging-involved accidents. This application will also enable us to identify those high risk locations both spatially and temporally in terms of visual illustrations.

Spatial Analysis

Kernel Density Estimation: Evaluation and Comparison

We will first present the results of a comparative spatial analysis based on the ED and RND approaches. These are implemented using ArcGIS 10 where the SANET method (Okabe et al. 2006) is used to conduct the RND approach. In this study, a bandwidth of 250 m was chosen to conduct the RND analysis. For the ED approach, on the other hand, 2000 m bandwidth was selected. Figure 3 shows the comparative results for Escambia, Hillsborough and Pinellas counties, respectively. As observed from Fig. 3, the drawback of the planar approach is as follows: The ED method obtains the density maps based on the planar (Euclidean) distance calculations without using the roadway network itself. That is, the distances from the center of the hotspot are still calculated without considering the existing roadway network structure. The accident density maps also extend over the areas like parks, residential areas, and lakes where there are no roadways and therefore no accidents. In order to overcome these drawbacks, the RND method is introduced to identify the accident hotspots more accurately. Through this method, it is possible to detect the roadways that have a high number of aging-involved accidents where every distance between the accidents is calculated based on the actual roadway (network) distance.

Fig. 3
figure 3figure 3

ED- and RND-based Kernel density maps for a Escambia county, b Hillsborough county, and c Pinellas county

Full size image

Moreover, the ED method identifies all the roadways and intersections that reside in the peak density region as ‘high accident’ risk locations. This is critical since it may cause the following problems: (a) Overestimation: Some roadways that do not actually possess high risk are shown to be risky, (b) Underestimation: Since multiple roadways are shown as critical locations rather than the actual roadways that have high accident risk, one may not give the needed attention to the actual high risk locations. The RND approach, on the other hand, solves these two problems using the actual roadway distances for the kernel density estimation.

For example, the ED approach for Escambia County (Fig. 3.a, left) shows the accident hotspots located in the Northeast and Southeast areas of the largest city of Escambia, namely Pensacola. The RND approach, on the other hand, enables us to identify the exact locations of the most critical hotpot corridors such as the N. Davis Road corridor between the 9th Avenue and Highway 29. This hotspot location in the Northeast Pensacola is basically the one that has the highest number of accidents for those aged 65 and over. However, the RND approach does not show any critical hotspots in the Southeast of Pensacola as identified by the ED approach. Although this may not be entirely visible from the 2-D maps, they are typically identifiable by the high surface peaks in a three-dimensional (3D) view of accidents. This can lead us to a more detailed assessment of aging population-involved accidents via another feature offered by the RND approach through SANET: 3-D visualization of the accidents on the roadway network. This approach allows us to observe high accident risk locations more clearly and accurately.

Age-Focused Visual Comparison Analysis

Figure 4 shows the resultant comparison maps based on the density ratio approach, where the numerical value (or percentage value) represents the difference of the normalized crash densities with respect to the different age group maps. Figure 4 also provides the 65+ population density in the background in order to facilitate a comparative analysis between the accident locations involving aging people and the aging populations living in the vicinity (with darker areas showing more 65+ populations). This approach allows us to also focus on those areas that have larger age 65+ populations, and investigate if there are more aging people-involved crashes in the vicinity. We note that the density ratio-based comparison provides visual results for exploration rather than testing for statistical significance. Locations with higher accident risk are presented in more detail for each county.

Fig. 4
figure 4

65+ vs. 16–19 and 20–64 accident density comparison maps for a Escambia county, b Hillsborough county, and c Pinellas county (The visual legend at the bottom right corner provides the locations of the maps in the county)

Full size image

Figure 4 shows that the accident high risk locations are substantially different for each age group. This difference is not equally explicit in all networks because of the unique population and characteristics of these networks. However, age-specific crash patterns are still observable on all three networks. For example, in Escambia County, aging-involved crashes are relatively higher in the southwest, whereas other age group crashes occur closer to Downtown Pensacola. In Hillsborough County, we observe higher aging-involved crashes at those east and south locations with higher aging population densities (Two maps in the bottom – Fig. 4b). Closer to the University of South Florida, the distinction is very clear (shown with the entirely green roadways in the area) indicating that the accidents with respect to the 16–19 populations happen more than those involving aging populations in that area. For Pinellas County, we observe less crashes in the southeast (Downtown Saint Petersburg area) where the 65+ population density is lower than other areas. Aging-involved crashes occur more in the north (closer to Clearwater) where more aging populations are living, as clearly shown by the 2-D population density maps. Visual inspection of these crash density maps is critical in order to recognize the geo-spatial patterns of aging-involved crash clusters. The next section will focus on 3-D maps in order to identify the accident hotspots more clearly with a focus on aging populations.

3-D Visualization

The RND based kernel approach available through SANET provides a means of visualizing the accident distributions over the whole roadway network in 3-D. Figures 5, 6 and 7 show the 3-D accident distribution maps for Escambia, Hillsborough and Pinellas counties, respectively. The peaks in these maps represent those locations that have higher number of accidents than others. In order to determine the hotspot locations, we first focus on each whole county. Next, based on the network distance-based kernel density estimates, we determine the high risk regions, and finally the hotspot locations. Figures 5, 6 and 7 also clearly show that aging-involved accidents mostly occur at the intersections for all the counties studied. This indicates that intersections appear to be significantly problematic for the aging populations. Please note that intersections can also be problematic for other ages; however, the locations of those risky intersections may differ for the aging and other adult age groups.

Fig. 5
figure 5

3-D Visualization of aging-involved roadway accidents for Escambia county a The whole county, b High accident roadways, c Hotspot corridor: N. Davis highway

Full size image
Fig. 6
figure 6

3-D visualization of aging-involved roadway accidents for Hillsborough county a The whole county, b High accident roadways, c Hotspot corridor: I-4 roadway

Full size image
Fig. 7
figure 7

3-D visualization of aging-involved roadway accidents for Pinellas county a The whole county, b High accident roadways, c Hotspot corridor: U.S. highway 19 and 66th street

Full size image

The drawbacks of the ED approach are more visible from these 3-D plots. For example, Sun City Center Boulevard of Hillsborough County, which is shown as a moderate hotspot in the ED-based maps of Fig. 3, has the highest peak (highest accident risk) in the 3-D view based on the RND approach (Fig. 6). Similarly, in Pinellas County, the Saint Petersburg area is shown as a hotspot location based on the ED approach (Fig. 3); however, the whole area is not actually a hotspot but rather certain roadways have higher accident rates than others in that area, which is more precisely shown through using the RND approach (Fig. 7).

The 3-D maps reveal other interesting trends. Accidents where aging populations are involved do not necessarily cluster the same areas where there are clusters for other age groups such as the downtown areas or other typically congested regions. This is an interesting pattern for aging road users which suggests that they may not necessarily prefer driving on congested or busy roadways as other age groups may do. Many aging drivers may actually prefer to use familiar nearby streets while shopping, volunteering or socializing (Ball et al. 1998; Charness and Schaie 2003; Staplin et al. 2012), which could be a prominent reason behind how aging-involved accident spatial patterns differ from other adult age groups. This finding is also correlated with temporal patterns of aging-involved accidents, which will be discussed later in the Temporal Analysis section. The next section will focus on a more detailed analysis of the hotspot locations.

Hotspot Corridor Evaluation

From the 3-D RND maps produced by SANET, three hotspot corridor locations that have the highest peaks are identified for each county: N. Davis Road between the 9th Avenue and Highway 29 in Escambia County, N. Dale Mabry Highway in Hillsborough County, and U.S. Highway 19 between the 144th Ave N. and 49th St in Pinellas County. These locations have the highest kernel density values obtained from the RND approach. These high accident risk locations (locations that have higher accident rates) are presented in Table 1 for Escambia, Hillsborough and Pinellas counties, respectively. In Table 1, results are presented based on different age groups in order to facilitate a comparison. The italicized portion of Table 1 represents the accident statistics for the whole county whereas the remaining part of Table 1 shows the number of accidents and intersection-only accidents at the selected hotspot locations. Please note that intersection-only accidents include those that occur at the intersections and the ones that are influenced by the intersections, as defined by the Florida Department of Transportation (FDOT).

Table 1 Hotspot corridor characteristics for Escambia, Hillsborough and Pinellas (2008–2012)
Full size table

Table 1 shows that the percentage of aging-involved accidents at the hotspot locations divided by the total aging-involved accidents in the county is always high compared to other age groups. This indicates that the hotspots identified based on their high rate of aging-involved accidents may not be that risky for other adult age groups. Similarly, when we compare the hotspot accidents with respect to their age group population (selected based on a 5-mile buffer around the hotspot location), the percentages of aging-involved accidents divided by the aging population are also relatively higher for the three counties compared to other locations. The 5-mile buffer zone radius is selected as a reasonable proxy for a typical trip length traveling by driving a personal car. For example, it may reflect the length of a trip to visit a nearby store along a familiar route (Ball et al. 1998; Charness and Schaie 2003; Staplin et al. 2012). Moreover, a 5-mile radius is roughly consistent with the assumption of a 15 minute driving distance from the hotspot location accounting for the effects of traffic signals and possible delays on roadways (depending on the roadway and traffic characteristics). For these reasons, a 5-mile radius is selected as a representative measure for all the hotspot locations in the three studied counties.

Results also indicate that intersections appear to be the high risk locations for aging people. For all hotspot locations, accidents at intersections are more than 50 % of the total number of accidents involving aging populations. For Escambia and Hillsborough counties, this percentage (55.67 % for Escambia and 65.82 % for Hillsborough) is higher than other age groups, especially for the 50- population (51.17 % for Escambia and 43.75 % for Hillsborough). For Pinellas, the percentage is 64.36 %, which is close to the 50–64 age group (66.46 %), and still more than 50- people (61.93 %). The 50–64 age group is given extra attention in Table 1 since they represent the ‘Baby Boom’ generation, and are expected to drive a 79 % increase in the 65+ population over the next 20 years (Koffman et al. 2010), making future planning all the more challenging.

Further analysis is needed to identify the factors that might affect aging-involved accidents, such as the geography of people living in the vicinity of the accidents, or the aggregate number of vehicles on the roadway. Such analysis would reveal the prominent reasons behind the occurrence of these accidents, which is a very promising future research direction.

Temporal Analysis

In this section, temporal analysis is conducted through the use of spider plots for the same three selected counties (Fig. 8). For this purpose, temporal units of hours, days, and months are examined to reveal the trends in the accident data. This analysis reveals an interesting hourly pattern of accidents. Most of the aging-involved accidents occur during the mid-hours of the day, not during the peak travel hours, with the ‘peak’ being usually before the evening rush hours. That is, we do not observe a peak in the morning and evening rush hours as we would expect for the working age group accidents. This makes the aging-involved accidents unique in the sense that they do not follow the expected pattern. This may be due to the fact that most aging Floridians are retirees and/or they do not prefer to drive in the rush hours. This specific aging driver behavior is also clearly presented in several other studies including Ball et al. (1998), Charness and Schaie (2003), and Staplin et al. (2012).

Fig. 8
figure 8

Temporal analysis of aging-involved accidents: hourly, daily and monthly variations for a Escambia county, b Hillsborough county, and c Pinellas county

Full size image

From the daily variation plots, we also observe higher accident rates during the weekdays as compared to the weekends. This may reveal another specific preference for some aging people: driving during the weekdays more than weekends. Monthly variation plots, on the other hand, do not really show a significant change in the accident rates between months, except a substantial increase during the winter months for the Pinellas County. This may be due to the seasonal migration (in-flow) of retired adults from elsewhere in the northern U.S. during the colder seasons. Rose and Kingma (1998) state that the Pinellas County has the highest amount of nonpermanent residents of 65 and older in Florida. Hillsborough and Pinellas counties show a slight decrease in aging-involved accidents during the summer months. However, there is not a clear picture on the accident variation like the hourly and weekly plots. It is evident however that this temporal analysis can provide us with a better understanding of the variation in aging-involved accidents between hours of the day, days of the week, and months of the year, which again, can inform future mitigation strategies.

Spatio-Temporal Analysis

In order to further discern the spatial and temporal patterns of aging-involved accidents, an analysis that makes use of both dimensions is needed, such as the comap approach. The comap method is used to conduct the spatio-temporal analysis shown in Fig. 9, where maps are constructed for four different time periods of the day: AM peak (6–9 AM), midday (9 AM-4 PM), PM peak (4–7 PM) and night (7 PM- 9 AM) periods. For each time period, the spatial distribution of the aging-involved accidents are presented for each county, which leads us to being able to identify the temporal hotspots, especially those during the AM and PM rush hours. For example, Escambia County has two hotspots in the morning, namely the downtown Pensacola and Brent areas; however, as time progresses, a new hotspot is detected near the Northwest and Northeast of Pensacola (away from the downtown). For Hillsborough, the hotspots are detected around New Tampa, Brandon and Sun City Center regions, and they mostly remain high risk locations as time passes. In Pinellas County, on the other hand, we do not observe any change on the hotspot locations; however, the intensity of the accidents around Palm Harbor, Largo and Tarpon Springs regions clearly increases. Therefore, this time-based variety for the hotspots can be determined using the comap method rather efficiently. Identifying accident distribution patterns both spatially and temporally can provide better guidance to the planners on where and when the accident prevention remedies and strategies should be applied. Other spatio-temporal methods such as the SatScan method, developed by Kulldorff (1997), can also be implemented and compared with the results presented in this paper. However, in this paper, the comap method is still a more viable option since the general clustering behavior identified by the comap method is what is of interest rather than the identification of specific clusters, which can be accomplished successfully using the SatScan method.

Fig. 9
figure 9

Comap of aging-involved accidents for a Escambia, b Hillsborough, c Pinellas

Full size image

Discussion

This exploratory study is conducted to fill the knowledge void related to aging-involved accidents. A diverse suite of methods is applied in order to identify the accident hotspots and time periods in three urban Florida counties with high accident rates involving aging road users. To this end, spatial, temporal and spatio-temporal methods are implemented based on accident locations found in the Florida Department of Transportation accident database recorded for the years 2008 and 2012. There are several important findings that can assist planners and policy makers for better management of traffic operations with a focus on aging populations:

Intersection locations are more adversely related to the 65+ populations compared to other adult age groups. This is critical since popular places like grocery stores and pharmacies for aging people are often located in commercial areas that often contain numerous intersections for access purposes. This problem can become even more difficult when multiple intersections in close proximity, complex signalizations and unexpected design features are present. Since redesigning a roadway intersection would be very costly to transportation agencies, it can be more appropriate to maintain and operate problematic intersections in a smarter way, especially in regions that have high aging populations such as the counties studied in this paper. These strategies include better signalization, signing and communication through IT-based systems such as Intelligent Transportation Systems (Hanowski and Dingus 2000; Charness and Schaie 2003). This will help with the timely communication of vital roadway/traffic information to the 65+ population, which can be critical for their safety.

Aging-involved accidents tend to occur during mid-day rather than the peak hours, which is not a similar pattern to other adult age groups, especially compared with the working populations. This is again related to an aging population-specific behavior since they may want to visit those places like grocery stores and pharmacies in the least congested time periods, with the most familiar routes, and with the least traffic present on the roadways (Ball et al. 1998; Charness and Schaie 2003; Staplin et al. 2012). Since the 65+ Floridians are mostly comprised of retirees, it is possible for them to avoid the rush hours and complete their daily obligations within the workday. Weekday accidents that involve aging people are also more than those of weekend accidents. This can also be due to the fact that aging populations prefer to drive and/or complete their daily activities like shopping mostly on weekdays rather than on weekends. Transportation officials, especially those that maintain and operate the roadways close to senior living communities, should be aware of the consequences of this high accident risk for aging populations during the mid-day hours of the weekdays.

Kernel density analysis based on the ED approach is the most widely used geo-spatial analysis method in the literature and practice in order to determine accident clusters (Paez and Scott 2004). Results indicate that this may not be the most appropriate approach since accidents actually occur on the roadway network where distances between two points are not necessarily Euclidean. ED approach also overestimates the problem by presenting all the roadways located in the peak density regions as risky. Therefore, it is appropriate to suggest the RND approach, which makes use of the network distances, and can create more accurate density maps.

Crash density comparison maps based on the density ratio approach clearly indicate that aging-involved crash maps have a different geo-spatial pattern than the density maps with respect to other age groups. Therefore, transportation officials can use the proposed approach to analyze the aging-specific spatial and temporal characteristics of accidents, which can help with identifying the possible reasons behind the high risk associated with the hotspot locations. Using this aging-focused application, better accident prevention and reduction strategies can be adopted.

Conclusions and Future Research

This study evaluates and analyzes aging-involved roadway accidents via spatial and temporal techniques. Following a review of the spatial, temporal and spatio-temporal methods, a case study is presented on the roadway networks of three urban counties in Florida, namely Escambia, Hillsborough and Pinellas, based on three distinct GIS-based approaches. Results of this research not only highlights the risky accident locations and time periods, but can also contribute to the development of more reliable aging-focused transportation plans and policies.

Results of this analysis indicate the following: (a) a substantial amount of aging-involved accidents happen at the intersections, and (b) they happen during the mid-hours of the day rather than the peak hours. This temporal result is an unusual pattern for accidents, which basically separates aging-involved accidents from other adult age groups. This knowledge can allow planners and engineers to focus on those high risk areas and time periods for safety-focused intervention efforts to assist aging drivers. This can also help increasing the efficiency of the traffic operations towards obtaining better plans that focus on aging populations.

Returning to the policy front, exploratory analyses such as the present effort can provide key information that could be used by planners and decision makers to formulate educational agendas aimed at promoting safety and well-being for aging populations. Specifically, the types of data, maps, analyses, and insights presented in this paper could be a part of efforts to raise awareness of safety issues, and make people more cognizant of locations near them that might require further care in navigating. With regards to aging populations, there are many community-oriented organizations charged with assisting aging people meet their daily transportation and other life needs. The types of insights produced in this study may have potential to assist them in their efforts to help aging people remain mobile and successfully interact with society. In terms of prioritizing efforts to make intersections safer, results such as these could also support plans for infrastructure spending aimed at improving ‘risky’ conditions at critical intersections for aging populations. Of course, such actions would be a part of larger studies, including cost-benefit analysis of particular infrastructure improvements, the calculation of which is beyond the scope of this study. Clearly, however, our results would be useful relative to such broader planning efforts.

The results of this research can be extended in the future via statistical models to identify the significant factors that affect the aging-involved accidents. This can help researchers to better understand the prominent reasons behind the occurrence of these accidents, focus deeper on aging driver behavior, and pinpoint which road characteristics are unsuitable for aging drivers. However, such results may also be location specific. Therefore, another interesting area of research is to expand this research to other counties of Florida, especially to other priority locations, identified by Florida’s Safe Mobility for Life Coalition as having high aging populations (Sandler et al. 2015). Florida notwithstanding, aging populations are growing all over the U.S. and world, and further analysis of the factors influencing their accident rates is an important area of future research.