Scholarly article on topic 'Age-Based Stratification of Drivers to Evaluate the Effects of Age on Crash Involvement'

Age-Based Stratification of Drivers to Evaluate the Effects of Age on Crash Involvement Academic research paper on "Social and economic geography"

CC BY-NC-ND
0
0
Share paper
Academic journal
Transportation Research Procedia
OECD Field of science
Keywords
{"Crash Analysis" / "Safety Modeling" / "Aging Populations" / "Geographic Information Systems" / "Spatiotemporal Density Analysis"}

Abstract of research paper on Social and economic geography, author of scientific article — Mehmet Baran Ulak, Eren Erman Ozguven, Lisa Spainhour

Abstract: Traffic crashes imperil roadway users and impose a considerable economic burden on society. Previous studies have investigated several aspects of traffic crashes such as severity, frequency, and influential factors, yet the causal and spatial differences between crashes involving drivers from different age groups remain unclear. In this paper, we illustrate the causal, spatial and temporal variations in crash involvement patterns of different age groups, including age 16-49, 50-64, and 65+. Furthermore, we stratified the aging drivers (65+) in order to explore the age-specific differences within this group, considering that the changes brought by aging may differ tremendously at different stages. For this purpose, we implemented an approach with two distinct phases: (a) a geographic information system (GIS)–based spatiotemporal analysis, and (b) a multinomial logistic regression analysis. Results indicate that crashes of different age groups differ not only in terms of influential factors but also spatially and temporally on the roadway network. The findings of this study will be eminently useful for public and/or private agencies to identify and address problematic issues particularly for aging drivers, and thus will enhance the safety and mobility of aging roadway users as well as entire population.

Academic research paper on topic "Age-Based Stratification of Drivers to Evaluate the Effects of Age on Crash Involvement"

Available online at www.sciencedirect.com

Science Direct

Transportation Research Procedia 22 (2017) 551-560

Transportation Research

Procedia

www.elsevier.com/locate/procedia

19th EURO Working Group on Transportation Meeting, EWGT2016, 5-7 September 2016,

Istanbul, Turkey

Age-Based Stratification of Drivers to Evaluate the Effects of Age

on Crash Involvement

Mehmet Baran Ulaka*, Eren Erman Ozguvena, Lisa Spainhoura

a Civil and Environmental Engineering, FAMU-FSU College of Engineering, 2525 Pottsdamer Street, Tallahassee, FL, 32310, USA

Abstract

Traffic crashes imperil roadway users and impose a considerable economic burden on society. Previous studies have investigated several aspects ogtraffic crashes such as severity, frequency, and influential factors, wet the causal and spaiial differences between crasht^ involving drivers from, different age grouns remain unclear. In this paper, we illustraSe the causal, spatial and temporal variations in ce^sh involvement patterns oedifferent age groups, including age 16-49, 50-6)4, and 65+. Furthermore, we stratified the aging drivers (65-+) in odder to explore the age-specific differences within this group, consrd5ridg that the changes brought by aging may PISsi' tremfndouslyet different stages. For this purpose, -we implemented an approach with two distinct phases. (a) a geographic information system (GlS)-based spatiotemporyl analysis, and (b a multinomial logistic regression analysis. Results indicate that crashes of different age groups differ not only in terms of influentM factors bui also spatially and temporally on the roaoway network. Tha fiddidgs ootids study will be eminently uneM for pubfe antner private agencieg to identify and address problematic issues particularly for aging drivers, and thus will enhance the safety and mobility of aging roadway users as well as entire population.

© 2017 The Authers. Pubhshed by B.V

Petr-reviewunder re+sponsiibi^it^ of yhe Scientific Committeeef EWGT2016.

Keywords: Crash Analysis; Safety Modeling; Aging Populations; Geographic Idfofmaeiod Systems; Snaeioefmnofal Dfdtiea Analatit

* Presenting author

E-manl address: inulak@fsu.edu

2214-241X © 2017 The Authors. Published by Elsevier B.V.

Peer-review under responsibility of the Scientific Committee of EWGT2016.

10.1016/j.trpro.2017.03.044

1. Introduction

In the transportation and traffic safety field, spatial and causal analysis of crashes is one of the most important issues on which researchers focus their efforts. Several studies have investigated the critical aspects of traffic crashes such as severity and frequency considering various methods ranging from basic descriptive analysis to complex regression models (Lord and Mannering, 2010; Mannering and Bhat, 2014; Yasmin et al., 2014). Researchers have also been interested in distinguishing the effects of spatial, temporal, environmental, traffic, roadway and driver-related factors that influence the crashes as well as the correlations between them (Moore et al., 2011; Rifaat et al., 2011). From a traffic safety perspective, this issue becomes even more complex when aging populations are considered, due to their health, behavioral, physical and cognitive limitations. Note that the term "aging," as used in this study, refers to persons 65 years and older. Furthermore, a better examination of aging population-involved crashes compels stratification of older population into subgroups since changes brought by growing old differ tremendously at different stages of aging (Sifrit et al., 2010). Indeed, Braver and Trempel (2004), and Griffin (2004) found statistically significant increase in the fatality probability by increasing age in older drivers. This implies that different age segments of older drivers may have divergent crash involvement characteristics, which are unique to each segment. For instance, a differentiation between 65-69, 70-74, and 75+ seniors would be especially beneficial, since in the early periods of older adulthood (close to the age of 65), deterioration of sight and reflexes is not as evident as it is in the following years (Alzheimer's Disease Facts and Figures, 2009; Dementia: Hope Through Research, 2013). This older age group stratification has not been done in the literature before, which is a significant contribution of this paper. There are many studies that investigated the influence of age and aging on crashes. Bayam et al. (2005) provides an exhaustive review of these studies. Yet, spatiotemporal differences between crash involvement patterns of aging drivers and other age groups, as well as intra-group variations within the aging (65+) drivers are not fully understood.

According to 2014 estimates, aging people are 14% and 18% of the total population in U.S. and Florida respectively (A Profile of Older Americans: 2014; Florida Estimates of Population, 2014). Since this population growth among aging Floridians is expected to continue in the state of Florida (even faster than the overall U.S.), it becomes crucial to investigate the nature of the crashes involving aging populations. In addition, persons 50-64 years old, who can also be referred to as Baby Boomers, are also of critical importance since the 65+ population in the U.S. is expected to increase by 79% due to the aging of the Baby Boom generation in the next two decades (Koffman et al., 2010). As such, the number of aging road users and crashes involving aging drivers on Florida roadways are expected to increase, which makes studies on aging population-involved crashes even more critical. Therefore, it is significant to comprehend the spatiotemporal factors causing these aging population-involved crashes, and identify hazardous situations that jeopardize the well-being of aging drivers, passengers and other roadway users. This paper is an important step towards filling this gap.

2. Methodology

The goal of this study was twofold. First, we aimed to disclose the differences between crashes involving 50-64 and 65+ drivers in terms of involvement characteristics, spatial distribution, and significant factors causing those crashes. Then, we divided aging drivers into three subgroups (65-69, 70-74 and 75+) in order to explore these differences among aging drivers, who are oftentimes evaluated as a homogeneous group. We explored the differences in crash involvement between stratified age groups via two main approaches: (a) spatiotemporal analysis to identify crash hotspots and explore spatial changes in these crash hotspots throughout the day, and (b) regression analysis to delve into the nature of those crashes. We adopted the Comap method (Asgary et al., 2010; Backalic, 2013; Brunsdon et al., 2007; Kilamanua et al., 2011) in order to illustrate the variation in crash hotspots with a focus on the following four time spans: AM peak hour (6:00 AM - 9:00 AM), midday (9:00 AM - 4:00 PM), PM peak hour (4:00 PM - 7:00 PM), and night (7:00 PM - 6:00 AM). Moreover, we incorporated a density ratio method, first proposed by Ulak et al. (2015), with comaps to examine the spatiotemporal variation of crash hotspots with respect to different time spans and age groups. Following the spatiotemporal analysis, multinomial regression analyses were conducted to discover the effects of crash-influencing factors among different age groups.

2.1. Study Area and Data

Figure 1. Selected Urban and Rural Counties in Florida

Ten urban and ten rural counties (Figure 1) were selected for the proposed analysis in the State of Florida, USA:

(a) Urban counties: Alachua, Bay, Broward, Duval, Escambia, Leon, Miami-Dade, Monroe, Orange, and Seminole,

(b) Rural counties: Columbia, DeSoto, Hamilton, Hardee, Jackson, Levy, Madison, Putnam, Taylor, and Walton. These urban and rural counties were documented to have high crash rates with respect to their aging population, and therefore defined as priority counties by the Safe Mobility for Life Coalition in Florida (Sandler et al., 2015). Data for the analysis was obtained from the respective roadway network and crash databases of Florida Department of Transportation (FDOT, 2015). Crash data of the urban and rural counties were merged individually to obtain one urban and one rural data set at the end. Based on this data, we created spatiotemporal crash density maps that illustrated those locations with highest crash risk (hotspots) at different times of day for each age group. Following this investigation, crash related attributes were evaluated using statistical methods in order to identify those factors that influence the occurrence of the aging population-involved crashes. Although crash data has a variety of attributes, there was a need to explore the most prominent ones to be used in the regression analysis, with a specific focus on the effects of these attributes on aging population-involved crashes. For this purpose, we conducted a preliminary study with the following steps: literature review, examination of the crash metadata, and statistical investigations. This resulted in the selection of attributes given in Table 1.

Table 1. Variables used in regression analysis

Binary Variables "Day of Week" (1:Weekend, 0:Weekday), "At Peak Hour" (1:Yes, 0:No), "Alcohol-Drug Abuse" (1:Yes, 0:No), "Intersection Presence" (1:Yes, 0:No), "Traffic Control Unit Presence" (1:Yes, 0:No), "Work Zone Presence" (1:Yes, 0:No), "Light Condition" (1:Night, 0:Else), "Road Condition" (1: Defected, 0:Good), "Road Surface Condition" (1:Slippery, 0:Dry), "Visibility"(1:Bad, 0:Clear) and "Lane Departure Action" (1:Yes, 0:No)

Continuous Variables "AADT" and "Speed Limit"

2.2. Spatiotemporal Crash Density Analysis

Spatiotemporal age group-specific crash density maps were created by using the network distance-based Kernel Density Estimation (KDE) analysis, implemented via the Spatial Analysis along Networks (SANET) tool (Okabe et al., 2006) in ArcGIS. This led to a more detailed exploration of crash hotspots pertaining to four time spans (AM peak hour (6:00 AM - 9:00 AM), midday (9:00 AM - 4:00 PM), PM peak hour (4:00 PM - 7:00 PM), and night (7:00 PM - 6:00 AM)). Note that AM and PM peak hours as well as midday and night time spans were grouped in pairs so as to obtain two main temporal groups namely; "AM & PM Peak" and "Midday & Night". In this study, we implemented the network distance-based KDE instead of planar KDE because it was desirable to detect the high crash risk locations where distances between the crashes were calculated based on the actual roadway distance. This issue was studied by many researchers where network distance-based KDE was found to be a better method for studying roadway crashes (Dai et al., 2010; Okabe et al., 2009, 2006; Steenberghen et al., 2010; Xie and Yan, 2013; Yamada and Thill, 2004). Spatiotemporal crash density maps were very helpful to illustrate those high crash risk locations on the roadway network. Yet, our main goal was to investigate the crash variation at these locations throughout the day. Hence, another approach was necessary in order to reveal the differences between crash maps, and detect age group-specific spatiotemporal variations. For this purpose, we modified the "Density Ratio" measure,

developed by Ulak et al. (2015), and integrated it within the Comap method. Consequently, the resultant "Spatiotemporal Density Ratio" (STDR) measure represents the difference between the maxima-normalized crash densities for different age groups at a specific time span. For more information on the "Density Ratio" method, please refer to Ulak et al. (2015). The formula of STDR is shown in Equation (1).

Dit Dit

STDRn =--^-r--'j-T (1)

max[pit) max[Pjt)

where STDR is the "Spatiotemporal Density Ratio" between the compared maps i and j at time span t, Di,t and Dj,t are the density values of the corresponding roadway links, and max(Dlt) and max(Dj,t) are the maximum link density values of the compared maps, respectively.

2.3. Regression Analysis

In this study, we utilized a multinomial (ordinal) logistic regression method in order to explore the effects and variation of the significant factors that cause traffic crashes involving different age groups. Multinomial logistic regression is an appropriate approach for this type of analysis since it allows us to create a categorical response variable (i.e., crashes involving 50-64, 65+ and other age group drivers), and explore the relative effects of the predetermined predictors for these different age group categories. This was achieved in two phases: first, three age categories were defined, including 16-49 years old, 50-64 years old, and 65 years and older drivers, respectively. Next, this was followed by a further categorization in the second phase in order to stratify the aging (65+) age group as follows: 65-69, 70-74, and 75+ drivers. As a result, two multinomial logistic models were created in order to evaluate the crash involvement with respect to different age groups. This older age group stratification has not been done in the literature before.

For the proposed analysis, we considered spatial, temporal, environmental, traffic, roadway and driver-related factors such as Annual Average Daily Traffic (AADT), speed limit, time of day, day of week, light and weather conditions. Since dependent variable is multinomial, we adapted a special type of logistic regression method, namely multinomial (ordinal) regression, in order to be able to analyze those variables with more than two categories. For example, age group category was defined as a multinomial variable with the following categories: "1/younger" for 16-49 years old, "2/middle" for 50-64 years old, and "3/older" for 65 years and older (65+). Using this multinomial regression approach made it possible to analyze the crash occurrences with respect to different age group drivers. The regression model for three analyzed categories can be illustrated as follows (Maddala, 1986):

where n1, n2, and n3 are P(y=1), P(y=2), and P(y=3), respectively, fis are the variable coefficients, and j is the variable index. The ratio within the ln(.) function is the so called odds ratio (OR). Note that raising the absolute value of a factor (X) increases the probability of either the numerator or the denominator in the LHS of the Equation (2), depending on the sign of the coefficient of the factor. Positive coefficients (fi>0) increase the probability given in the numerator, whereas negative coefficients (fi<0) increase the probability given in the denominator. Likewise, if odds ratio, OR is greater than 1, the odds for the numerator are higher than the odds for the denominator.

3. Results

3.1. Spatiotemporal Crash Density Analysis Results

Although we have studied all the ten urban and rural counties aforementioned, we will only provide the visual results for age 16-49, 50-64 and 65+ drivers in the Duval County in this paper due to space limitations (Figure 2). Analysis results illustrate that crashes involving stratified age groups differ spatially and naturally from other population-involved crashes (i.e. 16-49 years old). Findings indicate that each age group has a unique crash hotspot pattern that distinguishes them from each other. For example, "AM & PM peak hour" crash density maps show that high crash risk clusters of 65+ age group are considerably different than of 65- groups. For instance, Figure 2 illustrates that crashes involving 16-49 and 50-64 years old drivers are more uniformly distributed on the roadway network whereas aging population-involved crashes are more clustered at specific location. Furthermore, Figure 3 reveals that crashes involving 16-49 and 50-64 years old drivers are highly clustered in the downtown Jacksonville area whereas aging population-involved crashes are less frequent at this particular location. This is particularly evident in AM & PM Peak comparisons (Figure 3), which clearly indicate the temporal differences between age groups. It is arguably an expected result since 65- drivers are likely still working as opposed to 65+ drivers. Results indicate temporal variations within each age group.

Figure 2. Spatiotemporal crash density maps of Duval County. Rows top to bottom: age 65+, age 50-64, and age 16-49. Columns right to left: AM & PM Peak, and Midday & Night.

Figure 3. Spatiotemporal age group-based crash density ratio comparison maps. Rows top to bottom: age 16-49 vs. 65+, and age 50-64 vs. 65+. Columns right to left: AM & PM Peak, and Midday & Night.

Figure 4. Spatiotemporal time group-based crash density ratio comparison maps. Columns right to left: age 65+, age 50-64, and age 16-49.

The ratios between maximum crash densities of AM & PM Peak and Midday & Night temporal groups are 1.07, 1.63, and 1.96 for age groups 65+, 50-64, and 16-49, respectively (Figure 2). This implies that 50-64 and 16-49 drivers are more likely to have a crash during AM and PM peak hours. In contrast, we do not observe such an increase in crashes involving 65+ drivers. For example, crashes involving 50-64 year old drivers are substantially clustered in the downtown Jacksonville during AM and PM rush hours (Figure 4). However, we do not observe those crashes during midday and night hours. Spatiotemporal maps obtained using the Comap method provide information on the variation in the crash densities throughout the day. Note that crash density, as an indicator of the number of crashes at a certain roadway section or intersection, is an important metric provided to the transportation agencies towards solving the problems related to the crash frequency and severity. Results for Duval County indicate that aging population-involved crash densities do not show such a variation during different time spans. On the other hand, there is a notable increase in the crash densities for other age groups (16-49 and 50-64) during AM and PM peak hours, which is even clearer for the 16-49 years old age group. Exploring these spatiotemporal variations with respect to different crash hotspot patterns can lead the agencies to provide more effective preventive measures during those specific time spans that have higher crash rates involving different age groups.

3.2. Multinomial Regression Analysis Results

Results show that factors affecting the crashes involving 50-64 age group drivers are more similar to those of aging population-involved crashes rather than those involving 16-49 years old drivers both in urban and rural counties. For example, 65+ and 50-64 years old drivers appear to be less prone to being involved in crashes under alcohol and/or drug influence (Driving Under Influence - DUI crashes) compared to younger drivers (16-49 years old). Similarly, presence of an intersection and/or traffic control unit increase the crash involvement risk of both 50-64 and 65+ age group drivers. However, the negative effects of these factors are more evident for 65+ drivers. These findings confirm and support results of previous studies (Bayam et al., 2005; McGwin and Brown, 1999).

AADT increases the probability of having a crash involving both 50+ and 65+ drivers in urban counties; however, small coefficients (and Odds Ratios - ORs) imply that this effect is very limited (Table 2 and Table 3). In rural counties, on the other hand, we observe an opposite effect (Table 4). This could be arguably due to aging drivers' avoidance of congested traffic conditions as shown by Baker et al. (2003). Another interesting result is that the probability of involving in a crash is decreasing during weekends for 50+ years old drivers as well as 65+ drivers. Yet, we do not have any significant evidence showing a difference among 65+ drivers as verified in the second phase II (Table 3 and Table 5). "At Peak Hour" factor, however, indicates that the tendency of involving in a crash during AM (morning) and PM (evening) rush hours decreases by aging, even after the age-65 threshold. This result supports and adds to the findings of Collia et al. (2003), who did not delve into subgroups of 65+ aged drivers. Similar to "At Peak Hour", likelihood of involving in a crash while driving impaired reduces continuously by aging. Furthermore, results reveal that negative effects of the presence of an intersection and traffic control units become more tangible as drivers get older. This is critical because important facilities like grocery stores and pharmacies are oftentimes built close to busy intersections with many traffic signs and signals. ORs of "Light Condition", "Road Surface Condition", and "Lane Departure Action" imply that older drivers may prefer to avoid or do not prefer to drive at night, on roads with slippery surfaces, and recklessly. These findings are consistent with results of previous studies (Baker et al., 2003; Bayam et al., 2005).

In general, we obtained comparable results for urban and rural counties, despite several slight differences between. For instance, when weather is dark and effects of other factors are fixed, the odds of involving in a crash for younger (16-64) drivers are 113% higher than the odds for older (65+) drivers in urban counties, whereas it is 149% in rural counties. This may show that aging drivers are much more reluctant to drive at night in rural counties than in urban counties. It is possible to observe a similar pattern for the effect "Lane Departure Action" factor, which arguably indicates more aggressive driving behaviour for younger drivers and/or more risk-averse behaviour of older drivers in rural counties than in urban ones. Last but not the least, there is a considerable difference in impaired crash involvement among the 65+ drivers. Drivers aged 65-69 are much more prone to be involved in a DUI crash than 70+ drivers while driving in rural counties (OR: 2.03), compared to driving in urban counties (OR: 1.34).

Table 2. Multinomial logistic regression results of urban counties for phase I

Urban Counties (Phase -1: 16-49, 50-64, 65+)

K1 Vs. (X2 + ns) (ni + K2) vs. ns

Regressors Coeffs. OR St. Err. p val. Coeffs. OR St. Err. p val.

Intercept 0.07 1.07 0.01 ~ 0 1.45 4.24 0.01 ~ 0

Day of Week 0.17 1.18 0.01 ~ 0 0.11 1.12 0.01 ~ 0

At Peak Hour 0.08 1.08 0.01 ~ 0 0.28 1.33 0.01 ~ 0

AADT/10,000 0.00 1.00 0.00 ~ 0 0.01 1.01 0.00 ~ 0

Speed Limit 0.00 1.00 0.00 0.60 0.03 1.03 0.00 ~ 0

Alcohol-Drug Abuse 0.24 1.27 0.02 ~ 0 0.63 1.87 0.03 ~ 0

Intersection Presence -0.18 0.84 0.01 ~ 0 -0.22 0.81 0.01 ~ 0

Traffic Control Unit Presence -0.04 0.96 0.01 ~ 0 -0.06 0.94 0.01 ~ 0

Work Zone Presence 0.01 1.01 0.01 0.70 0.07 1.07 0.02 ~ 0

Light Condition 0.63 1.87 0.01 ~ 0 0.76 2.13 0.01 ~ 0

Road Condition -0.01 0.99 0.01 0.36 0.00 1.00 0.02 0.92

Road Surface Condition 0.16 1.17 0.01 ~ 0 0.21 1.23 0.01 ~ 0

Visibility 0.22 1.25 0.01 ~ 0 0.20 1.22 0.02 ~ 0

Lane Departure Action 0.38 1.46 0.01 ~ 0 0.25 1.28 0.01 ~ 0

ni : P(y=1), K2 P(y=2), ns P(y=3) for age categories: "1 " 16-49, "2": 50-64, "3" ■ 65+

Table 3. Multinomial logistic regression results of urban counties for phase II

Urban Counties (Phase - II : 65-69, 70-74, 75+)

ni vs. (K2 + ns) (ni + K2) vs. ns

Regressors Coeffs. OR St. Err. p val. Coeffs. OR St. Err. p val.

Intercept -0.73 0.48 0.02 ~ 0 0.29 1.34 0.02 ~ 0

Day of Week -0.02 0.98 0.02 0.23 -0.02 0.98 0.02 0.41

At Peak Hour 0.18 1.20 0.02 ~ 0 0.19 1.21 0.02 ~ 0

AADT/10,000 0.02 1.02 0.00 ~ 0 0.02 1.02 0.00 ~ 0

Speed Limit 0.04 1.04 0.01 ~ 0 0.03 1.03 0.01 ~ 0

Alcohol-Drug Abuse 0.29 1.34 0.06 ~ 0 0.54 1.72 0.07 ~ 0

Intersection Presence -0.08 0.92 0.02 ~ 0 -0.09 0.92 0.02 ~ 0

Traffic Control Unit Presence -0.03 0.97 0.02 0.04 -0.04 0.97 0.02 0.03

Work Zone Presence 0.06 1.06 0.04 0.16 0.11 1.12 0.04 0.01

Light Condition 0.24 1.27 0.02 ~ 0 0.28 1.32 0.02 ~ 0

Road Condition 0.06 1.06 0.04 0.13 0.10 1.11 0.04 0.02

Road Surface Condition 0.08 1.08 0.02 ~ 0 0.14 1.15 0.02 ~ 0

Visibility 0.09 1.09 0.03 0.01 0.05 1.06 0.03 0.09

Lane Departure Action -0.02 0.98 0.02 0.26 -0.07 0.93 0.02 ~ 0

rn: P(y=1), nr. P(y=2), ns: P(y=3) for age categories: "1": 65-69, "2": 70-74, "3": 75+

Table 4. Multinomial logistic regression results of rural counties for phase I

Rural Counties (Phase -1: 16-49, 50-64, 65+)

ni vs. (n2 + ns) (ni + n2) vs. ns

Regressors Coeffs. OR St. Err. p val. Coeffs. OR St. Err. p val.

Intercept -0.09 0.91 0.06 0.10 1.11 3.05 0.07 ~ 0

Day of Week 0.19 1.21 0.04 ~ 0 0.13 1.14 0.05 ~ 0

At Peak Hour 0.18 1.19 0.03 ~ 0 0.31 1.36 0.04 ~ 0

AADT/10,000 -0.09 0.92 0.01 ~ 0 -0.07 0.94 0.02 ~ 0

Speed Limit -0.01 0.99 0.01 0.44 0.03 1.03 0.01 0.03

Alcohol-Drug Abuse 0.15 1.16 0.06 0.02 0.54 1.72 0.10 ~ 0

Intersection Presence -0.30 0.74 0.03 ~ 0 -0.40 0.67 0.04 ~ 0

Traffic Control Unit Presence -0.06 0.94 0.03 0.05 -0.09 0.92 0.04 0.04

Work Zone Presence -0.22 0.81 0.09 0.01 -0.20 0.82 0.11 0.05

Light Condition 0.77 2.16 0.04 ~ 0 0.91 2.49 0.05 ~ 0

Road Condition 0.03 1.03 0.07 0.71 0.21 1.24 0.10 0.04

Road Surface Condition 0.34 1.40 0.04 ~ 0 0.35 1.42 0.06 ~ 0

Visibility -0.07 0.93 0.05 0.17 -0.03 0.97 0.07 0.67

Lane Departure Action 0.62 1.87 0.03 ~ 0 0.48 1.62 0.05 ~ 0

ni : P(y=i), n2 P(y=2), ns P(y=3) for age categories: "1 " 16-49, "2": 50-64, "3" ■ 65+

Table 5. Multinomial logistic regression results of rural counties for phase II

Rural Counties (Phase - II: 65-69, 70-74, 75+)

ni vs. (K2 + K3) (ni + n2) vs. K3

Regressors Coeffs. OR St. Err. p val. Coeffs. OR St. Err. p val.

Intercept -0.73 0.48 0.13 ~ 0 0.30 1.35 0.13 0.02

Day of Week 0.04 1.04 0.09 0.67 0.07 1.07 0.09 0.44

At Peak Hour 0.02 1.02 0.08 0.79 0.03 1.03 0.08 0.68

AADT/10,000 0.02 1.02 0.03 0.50 0.04 1.04 0.03 0.20

Speed Limit 0.04 1.04 0.03 0.15 0.02 1.03 0.02 0.30

Alcohol-Drug Abuse 0.68 1.98 0.19 ~ 0 0.71 2.03 0.22 ~ 0

Intersection Presence -0.15 0.86 0.08 0.07 -0.20 0.82 0.08 0.01

Traffic Control Unit Presence -0.12 0.89 0.08 0.14 -0.07 0.94 0.08 0.41

Work Zone Presence -0.08 0.92 0.20 0.68 0.07 1.08 0.19 0.70

Light Condition 0.20 1.22 0.10 0.05 0.38 1.46 0.10 ~ 0

Road Condition -0.14 0.87 0.20 0.48 -0.17 0.84 0.19 0.37

Road Surface Condition -0.17 0.84 0.12 0.15 0.03 1.04 0.11 0.76

Visibility 0.12 1.13 0.13 0.33 0.19 1.21 0.13 0.12

Lane Departure Action 0.04 1.04 0.09 0.65 0.04 1.04 0.08 0.64

nr. P(y=1), m: P(y=2), K3: P(y=3) for age categories: "1". 65-69, "2": 70-74, "3" 75+

4. Conclusions

This study presents a GIS-based methodology to evaluate and analyze the nature of crashes involving aging drivers in ten urban and ten rural counties of Florida via following approaches: (a) a spatiotemporal analysis through the Comap method and density factor approach, and (b) a multinomial logistic regression analysis. For this purpose, we stratified the driver age groups in two phases: (a) First phase: 16-49, 50-64, and 65+ drivers, and (b) Second Phase: 65-69, 70-74, and 75+ drivers, with a more detailed focus on the aging (65+) drivers. The differentiation between 65-69, 70-74 and 75+ seniors was eminently worthy, as aging drivers have often been evaluated as a homogeneous group in the literature, which is not the case. For example, there is still a decreasing tendency of crash involvement even after the age-65 threshold during AM (morning) and PM (evening) rush hours and driving under the influence. Likewise, negative effects of the presence of an intersection and traffic control units become more tangible as drivers get older. Note that, in the early periods of older adulthood (closer to the age of 65), a substantial amount of people still keep working rather than becoming retired. Moreover, the deterioration of physical capacity, sight, and reflexes is less evident during the early older adulthood than it is in the following years. Findings verify that there are significant differences in the effect of significant causal and spatiotemporal factors on the crash involvement not only between 65- and 65+ drivers, but also between stratified age groups of 65+ drivers.

As a result, we identified different crash involvement patterns for different age groups and time periods. Findings of this study confirmed and added to previous studies investigated effects of aging on drivers and crashes. In addition, we showed that the subgroups of aging drivers also differ from each other in terms of crash involvement characteristics. Finally, the comparison of findings of urban and rural counties present a few but important differences that could be quite helpful to enhance safety and mobility of aging and other drivers. Therefore, the knowledge gained from the results of this analysis will not only highlight critical crash hotspots to the public and/or private agencies, but will also contribute to the development of more reliable transportation plans and models with a focus on aging populations.

Acknowledgements

This project was supported by United States Department of Transportation grant DTRT13-G-UTC42, and administered by the Center for Accessibility and Safety for an Aging Population (ASAP) at the Florida State University (FSU), Florida A&M University (FAMU), and University of North Florida (UNF). We thank the Florida Department of Transportation for providing the data. The opinions, results, and findings expressed in this manuscript are those of the authors and do not necessarily represent the views of the United States Department of Transportation, The Florida Department of Transportation, The Center for Accessibility and Safety for an Aging Population, the Florida State University, the Florida A&M University, or the University of North Florida.

References

Administration on Aging, 2014. A Profile of Older Americans: 2014 [WWW Document]. U.S. Department of Health and Human Services. (URL http://www.aoa.acl.gov/aging_statistics/Profile/2014/docs/2014-Profile.pdf.) (accessed 1.10.16).

Alzheimer's Disease Facts and Figures, 2009. Alzheimer's Association, Alzheimer's and Dementia.

Asgary, A., Ghaffari, A., Levy, J., 2010. Spatial and temporal analyses of structural fire incidents and their causes: A case of Toronto, Canada. Fire Safety Journal 45 (1), 44-57.

Backalic, S., 2013. Temporal analysis of the traffic accidents occurence in province of vojvodina. Transport Problems 8 (1), 87-93.

Baker, T.K., Falb, T., Voas, R., Lacey, J., 2003. Older women drivers: Fatal crashes in good conditions. Journal of Safety Research 34 (4), 399405.

Bayam, E., Liebowitz, J., Agresti, W., 2005. Older drivers and accidents: A meta analysis and data mining application on traffic accident data. Expert Systems with Applications 29 (3), 598-629.

Braver, E.R., Trempel, R.E., 2004. Are older drivers actually at higher risk of involvement in collisions resulting in deaths or non-fatal injuries among their passengers and other road users? Injury Prevention 10, 27-32.

Brunsdon, C., Corcoran, J., Higgs, G., 2007. Visualising space and time in crime patterns: A comparison of methods. Computers, Environment and Urban Systems 31 (1), 52-75.

Bureau of Economic and Business Research, 2014. Florida Estimates of Population 2014 [WWW Document]. University of Florida. (URL http://edr.state.fl.us/Content/population-demographics/data/PopulationEstimates2014.pdf.) (accessed 1.10.16).

Collia, D. V., Sharp, J., Giesbrecht, L., 2003. The 2001 National Household Travel Survey: A look into the travel patterns of older Americans. Journal of Safety Research 34 (4), 461-470.

Dai, D., Taquechel, E., Steward, J., Strasser, S., 2010. The impact of built environment on pedestrian crashes and the identification of crash clusters on an urban university campus. The Western Journal of Emergency Medicine 11 (3), 294-301.

Dementia: Hope Through Research, 2013. National Institutes of Health.

FDOT, 2015. Traffic Data. Florida Department of Transportation. (URL http://www.dot.state.fl.us/planning/statistics/trafficdata/.) (accessed 1.15.15).

Griffin, L.I.I., 2004. Older driver involvement in injury crashes in Texas, 1975-1999. AAA, Foundation for Traffic Safety, Washington, D.C.

Kilamanua, W., Xiaa, J., Caulfieldb, C., 2011. Analysis of spatial and temporal distribution of single and multiple vehicle crash in Western Australia: a comparison study. In: 19th International Congress on Modelling and Simulation. Perth, Australia, 683-689.

Koffman, D., Weiner, R., Pfeiffer, A., Chapman, S., 2010. Funding the Public Transportation Needs of an Aging Population, American Public Transportation Association.

Lord, D., Mannering, F., 2010. The statistical analysis of crash-frequency data: A review and assessment of methodological alternatives. Transportation Research Part A: Policy and Practice 44 (5), 291-305.

Maddala, G.S., 1986. Limited-Dependent and Qualitative Variables in Econometrics. Cambridge University Press, Cambridge, United Kingdom, pp. 401.

Mannering, F.L., Bhat, C.R., 2014. Analytic methods in accident research: Methodological frontier and future directions. Analytic Methods in Accident Research 1, 1-22.

McGwin, G., Brown, D.B., 1999. Characteristics of traffic crashes among young, middle-aged, and older drivers. Accident Analysis and Prevention 31 (3), 181-198.

Moore, D.N., Schneider IV, W.H., Savolainen, P.T., Farzaneh, M., 2011. Mixed logit analysis of bicyclist injury severity resulting from motor vehicle crashes at intersection and non-intersection locations. Accident Analysis and Prevention 43 (3), 621-630.

Okabe, A., Okunuki, K.I., Shiode, S., 2006. The SANET toolbox: New methods for network spatial analysis. Transactions in GIS 10 (4), 535550.

Okabe, A., Satoh, T., Sugihara, K., 2009. A Kernel Density Estimation Method for Networks, Its Computational Method and a GIS-Based Tool. International Journal of Geographical Information Science 23 (1), 7-32.

Rifaat, S.M., Tay, R., De Barros, A., 2011. Effect of street pattern on the severity of crashes involving vulnerable road users. Accident Analysis and Prevention 43 (1), 276-283.

Sandler, M., Nichols, T., Tarburton, R., Allan, B., 2015. Health Care Providers and Older Adult Service Organizations to Assist in the Prevention and Early Recognition of Florida's At -Risk Drivers.

Sifrit, K.J., Stutts, J., Staplin, L., Martell, C., 2010. Intersection Crashes among Drivers in their 60s, 70s and 80s. In: Proceedings of the Human Factors and Ergonomics Society Annual Meeting. 2057-2061.

Steenberghen, T., Aerts, K., Thomas, I., 2010. Spatial clustering of events on a network. Journal of Transport Geography 18 (3), 411-418.

Ulak, M.B., Ozguven, E.E., Kocatepe, A., 2015. GIS-Based Investigation of Aging Involved Crashes : A Case Study in Northwest Florida. In: The Road Safety and Simulation International Conference. Orlando, Florida.

Xie, Z., Yan, J., 2013. Detecting traffic accident clusters with network kernel density estimation and local spatial statistics: an integrated approach. Journal of Transport Geography 31, 64-71.

Yamada, I., Thill, J.-C., 2004. Comparison of planar and network K-functions in traffic accident analysis. Journal of Transport Geography 12 (2), 149-158.

Yasmin, S., Eluru, N., Pinjari, A.R., Tay, R., 2014. Examining driver injury severity in two vehicle crashes - A copula based approach. Accident Analysis and Prevention 66, 120-135.