INTRODUCTION Traffic fatalities account for an astonishing

Background

Traffic fatalities account for an astonishing 1.24 million deaths per year

worldwide (Global Status Report on Road Safety, 2013). In the United States nearly

33 thousand lives were lost in motor vehicle crashes (“Traffic Safety Facts,” 2014).

These alarming statistics are coupled with the economic costs of motor vehicle

crashes, which in 2010 were an estimated $277 billion in the United States

(Economic and Societal Impact of Motor Vehicle Crashes, 2014). These statistics

leave no wonder as to why safety is of the utmost importance in all transportation

sectors. Safety is a complicated issue with many interrelated factors which affect

the safety of a roadway network. With the complete understanding that there are

several factors not explored in this thesis, the focus is mostly on risk-taking

behavior as a construct to fatality rates. Many studies have been performed to

distinguish the factors associated with the variation in safety measures throughout

the United States and the world. The southeastern region of the U.S. stands out for

having some of the highest rates of injury and fatality crashes per capita, which has

led to several studies that try to explain the unsafe anomalies associated with the

southeastern region. The majority of previous studies have focused on physical

design of infrastructure, weather, government policies, and socio-demographic

factors such as gender, age, race and ethnicity; such as: “Stop Sign Violations: The

role of race and ethnicity on fatal crashes,” by Romano, Voas and Tippetts, which

describes the role of race/ethnicity in fatal crashes, “Analyzing Road Safety in the

United States,” by Oster and Strong, which reviews the road safety performance in

theU.S., “Injury Severity of Multivehicle Crash in Rainy Weather,” by Jung, Qin, and

Noyce, which studied effects of rainy weather on injury severity of crashes, and “A

call to Safety,” by Heller, which addresses issues about distracted driving policies in

the U.S. All of the issues from previous studies are important and have significant

impacts on the safety of the roadway network. However, this thesis is different in

that the focus is on risk-taking behaviors, cultural factors, and health factors

associated with the socio-demographic characteristics as possible explanatory

factors in the variation of fatality rates in the southeastern region of the U.S.

There are several definitions of risk-taking behavior available. Risk taking

behavior is defined as, “any consciously, or non-consciously controlled behavior with

a perceived uncertainty about its outcome, and/or about its possible benefits or

costs for the physical, economic, or psycho-social well-being of oneself or others

(Trimpop, 1994).” Another definition of risk-taking behavior is “the tendency to

engage in behaviors that have the potential to be harmful or dangerous, yet at the

same time provide the opportunity for some kind of outcome that can be perceived

as positive (Tull, 2014).” Yet another alternative definition of risk-taking behavior is

the actions of a person when facing the possibility of physical, social, or financial

harm due to a hazard, uncertainty in outcomes, or to seek feelings of excitement

(Rohrmann, 2002). The risk-taking driving behaviors that are being referred to in

this thesis are the unhealthy risk-taking driving behaviors, some of which are

avoidable choices, and some of which are often unavoidable such as driving to work

in adverse weather, or driving home in the dark. The idea that an individual is more

likely to engage in risky driving behaviors if they are willing to engage in other

unhealthy risk-taking behavior in their life decisions comes from the idea that each

type of risk is associated with the individual’s perception of their self-worth, self-

esteem, and confidence (McLeod, S.A., 2012). Examples of other risky behavior in

life decisions are the choice to smoke, drink alcohol, use illicit drugs, drop out of

school, recklessly take out lines of credit that are nearly impossible to repay given

the individuals financial situation, etc.

Some of the risk-taking, cultural, and health factors that were initially

explored in this research include credit scores, seat belt use, education levels,

poverty and homelessness levels, alcohol consumption, drug and tobacco use,

obesity, overall general health, and depression and other mental health issues

causing diminished cognitive capabilities. Road use factors which were considered

include share of tourists and visitors, vehicle occupancy levels, lighting condition of

fatal crashes, average age of vehicles, and land use patterns. Through the

exploration of the data available at both aggregate and disaggregate levels, many

of the above mentioned factors were found to have correlations with fatality rates at

the state and regional levels, suggesting that understanding these different

characteristics of the population by geographical region can help to determine,

explain, and potentially provide insight regarding how to improve the safety of the

roadway network for personal vehicle travel.

The idea in correlating risk-taking behavior with fatality rates comes from the

assumption that individuals who have a higher likelihood of performing risky

behavior in some aspects of their life are also more likely to display risk-taking

behaviors while driving or as pedestrians. Risk-taking behavior while driving is

referring to aggressive driving characteristics such as: speeding, reactionary lane

changing without proper notification or attention, tailgating, hard braking, etc. The

National Highway Traffic Safety Administration (NHTSA) defines the occurrence of

aggressive driving as “an individual commits a combination of moving traffic

offenses so as to endanger other persons or property (NHTSA.gov).” Other risk

taking behavior while driving could include distracted driving, drowsy driving,

impaired driving, lack of use of safety equipment while driving such as safety belt

use, helmet use for motorcycle drivers, or improper or lack of car seat use for

children and infants (Research & Evaluation, 2014). In addition, risk-taking might

include activities such as driving in inclement weather and in poor road conditions,

driving a vehicle in need of maintenance, or allowing oneself to be distracted while

driving, etc. Similarly, behaviors of pedestrians, bicyclists, and motorcyclists can

include risk-taking elements.

1.2 Introductory Note

Vehicular crash rates and fatality rates are consistently higher in the

southeastern U.S. than the national average as shown in the Fatality Analysis

Reporting System. In addition, pedestrian fatality rates are also consistently higher

in the southeastern U.S. than the national average (Dangerous by Design, 2014).

With safety as a top priority in all sectors of transportation, it is imperative that the

safety of the roadway network in the southeastern region of the U.S. is improved.

To improve the safety of the roadway network, the factors associated with the lack of

safety in the region must be identified and addressed. Transportation safety is an

extremely complicated subject matter with an enormous number of interrelated

factors which influence safety as a whole. If transportation network safety were an

easy issue to address, or if safety was a straightforward problem with a singular best

answer, then solutions would be readily available. With the understanding of the

complicated nature of the problem of safety, this thesis focuses on just one aspect,

risk-taking characteristics. In order to identify the possible explanatory factors

involved with the lack of safety on the southeastern U.S. roadway network, risk-

taking characteristics are being explored, which were previously not evaluated as

conventional factors associated with safety.

1.3 Southeastern Transportation Center

The Southeastern Transportation Center (STC) is a consortium focused on

comprehensive transportation safety and is comprised of nine universities in the

southeastern U.S. The University of Tennessee is the lead institution which is home

to the director and co-director. The other universities include: the University of

Central Florida, University of Alabama, University of Alabama Birmingham,

University of Kentucky, North Carolina A&T State University, University of North

Carolina – Chapel Hill, and Clemson University. It is important to note that the

regions referred to are in accordance with U.S. Department of Transportation’s

(DOT) definitions of regions, not the U.S. Census Bureau’s definition of regions. The

DOT defines region four of the U.S. as Alabama, Florida, Georgia, Kentucky,

Mississippi, North Carolina, South Carolina, and Tennessee. In 2010, each of the

states in the southeastern region had higher fatality rates than the national average

both in terms of per capita and per vehicle miles of travel, and half of the

southeastern states ranked in the top ten of overall fatalities per capita, however as

of 2012, the number of southeastern states ranked in the top ten dropped to only

three, but there is still plenty of room for improvement (Southeastern

Transportation Center, 2015).

This thesis is part of a study being carried out by the University of South

Florida in collaboration with the University of Tennessee and the University of

Alabama, each of whom have various tasks exploring the relationship between

risktaking factors and travel safety. This thesis is representative of just one part of

the efforts being carried out by the University of South Florida.

1.4 Objective

The objective of this thesis is to explain the variation of fatality rates in the

southeastern region of the U.S. using different risk-taking characteristics as

explanatory variables in addition to several socio-demographic characteristics

known to effect driving behavior. The hypothesis being tested is that if the average

population in a specific geographical area behaves in a risky manner as a whole in

economic and health impacting characteristics that the same population may also

behave in a risky manner while operating a motor vehicle or using the roadway

network in other manners such as walking or biking. In other words, if an individual

is willing to take a given type of risk, they are inherently more likely to take

multiple risks, not just a singular risk. The hypothesis, restated is that if a person

exhibits risk-taking behavior, that risk-taking behavior is likely not limited to only

one aspect of risk, but is likely to occur in multiple facets of the person’s life.

1.5 Outline

This thesis gives an overall literature review comparing the variation of

fatality rates in the United States (U.S.) and the rest of the world for completeness,

then focuses on the variation of fatality rates in the southeastern region of the U.S.

compared to the rest of the nation. Following the fatality rate review, the different

risk-taking characteristics explored are reviewed. The methodology data collection

and analysis follows the literature review. In the fourth chapter of the thesis, the

data collection and analysis is reviewed, and the proposed regression model to

estimate fatality rates by geography and risk-taking characteristics is introduced.

The final chapter, chapter five, contains the conclusions and the suggestions for

future research regarding the topic of risk-taking variables as explanatory factors in

estimating fatality rates by geography.

CHAPTER 2:

LITERATURE REVIEW

The following literature review covers fatality rates, the variation of fatality rates by

geography, and the risk-taking behaviors that are proposed as explanatory factors

which affect the variation in fatality rates.

2.1 United States Compared to World’s Fatality Rates

Worldwide in 2010, 1.24 million people were killed in road traffic injuries,

which are estimated to be the eighth leading cause of death globally (Global Status

Report on Road Safety, 2013). The World Health Organization (WHO) reported that

road traffic fatality rates vary from 24.1 per 100,000 population in the African

Region, to 10.3 per 100,000 population in the European Region. The U.S.

experienced an 11.4 per 100,000 population traffic fatality rate in 2010. The

statistics show that the middle-income countries have the highest road traffic

fatality rates at an average of 20.1 per 100,000 population, followed by low-income

countries at an average rate of 18.3 traffic fatalities per 100,000 population, and

high-income countries have among the lowest traffic fatality rates at an average of

8.7 per 100,000 population. This variation in fatality rates by income status is

related to the share of vulnerable road users in a country. Vulnerable road users

include pedestrians, bicyclists, motorcyclists, and other users of two- or

threewheeled vehicles. Low- and middle-income countries have a higher proportion

of vulnerable roadway users, due to the lower costs and higher availabilities of

these modes. The U.S. is categorized as a high-income country, however many of

the laws which help to reduce traffic fatality rates are at subnational levels, leaving

plenty of room for improvement in traffic safety. A national urban speed limit

suggested by the WHO is less than or equal to 50 kilometers per hour, which

converts to about 30 miles per hour, however is only implemented in 47% of the

world’s population. The U.S. is one of the only countries with motorcycle helmet

laws and seat-belt laws at subnational levels when compared to the rest of the

world.

Perhaps most alarming is the share of countries with child restraint laws in

place coupled with good law enforcement of these laws. Child restraint law

enforcement is yet another area where the U.S. is categorized at a subnational level

(Global Status Report on Road Safety, 2013). While the U.S. is not among the

highest traffic fatality rates in the world, there is plenty of room for improvement in

the safety of the roadway network in the U.S. With each of the above mentioned

safety measures, speed limit, helmet laws, seat belt laws, child restraint laws, etc.

high levels of enforcement and high perception of enforcement is key to ensuring

new and existing policies are as effective as possible.

Differences in infrastructure conditions among countries also have great

influences on the safety variations of the networks. Road design, engineering,

construction, and maintenance are all influences to the safety of the roadway users

for all modes. According to the WHO, 77 percent of countries already carry out

some type of road safety audit on new road infrastructure, however existing

infrastructure also needs to be inspected with focuses on high risk, high incident

areas. In addition to infrastructure, typical vehicle age also plays a significant role

in the safety of a road network (Global Status Report on Road Safety, 2013).

Technology is improving tremendously with each passing year, and along with the

improved technology, the safety features in vehicles increase as well. With

increasing safety features such as multiple airbags, antilock brakes, and crushable

designs which allow the impact to be absorbed by the vehicle rather than the

occupant, the injury severity of motor vehicle crashes should reduce. This increase

in technology is often seen first in high-income countries where the population can

afford the luxuries of newer vehicles.

Emergency healthcare capacities are also a variable that plays a significant

role in the survivability of a person involved in a sever motor vehicle crash.

According to the WHO, as of 2013 only 59 countries have ambulatory services that

can rapidly transport the majority of patients to a hospital in the event of a crash.

In addition to the ambulatory services that are unavailable in many countries

around the world, the proportion of doctors and nurses that are trained to provide

emergency medical services range from less than 40 percent in the African and

South-East Asia regions to under 90 percent in the Eastern Mediterranean region.

In the region of the Americas, about half of nurses and over 70 percent of doctors

have this type of training (Global Status Report on Road Safety, 2013).

2.2 National Average Compared to Southeastern U.S.

Section 2.2 of this thesis focuses on the variation of characteristics between

the southeastern region of the nation and the rest of the nation, including fatality

rates and other characteristics that may affect fatality rates. The higher fatality

rates that occur in the southeastern region of the U.S. are a great motivation for

this study, and a phenomenon that should be researched thoroughly to improve the

safety of the southeastern U.S. population. The following table displays the fatality

rates in each state in the southeastern region, and the national average fatality

rate, in addition to the lowest and highest state in the nation.

Table 1 Fatality Rates in Southeastern Region

South Carolina

1.70

Mississippi

1.58

Kentucky

1.55

Tennessee

1.40

Alabama

1.35

Florida

1.26

North Carolina

1.23

Georgia

1.12

Total U.S.

1.11

D.C. (Lowest)

0.62

Montana (Highest)

1.74

Data Source: FARS and Census Bureau

2.2.1 Fatality Rates

U.S. roadways have experienced a reduction in fatalities of nearly 25 percent

between 2004 and 2013, however this still equated to almost 33 thousand motor

vehicle traffic crash deaths in 2013 (Traffic Safety Facts, 2014). The national

fatality rate in the U.S. is measured by exposure, thus is measured in fatalities per

100 million vehicle miles traveled (VMT) for most uses. In 2013, the U.S. fatality

rate was 1.10 per 100 million VMT, which constituted a 3.5 percent decrease in just

one year, and also ties for the lowest fatality rate on record for the nation (Traffic

Safety Facts, 2014). Just as recently as 2008, motor vehicle traffic crashes were

the number one cause of death for every age from 13 through 30 (Subramanian,

2012). The CDC reports fatality rates per population in the top 50 Metropolitan

Statistical Areas (MSA), and found that while overall, 74% of MSAs had traffic

fatality rates below the national average, the traffic fatality rates in the southern

states were generally higher. Furthermore, the highest vehicular fatality rates are

concentrated in the southeastern region of the U.S. (Motor Vehicle Crash Deaths in

Metropolitan Areas – United States, 2012). Pedestrian fatalities are another area of

concern, where the southeastern U.S. ranks higher than the national average in

fatalities. In 2012, pedestrians represented nearly 15 percent of all traffic fatalities

in the U.S. (Dangerous by Design, 2014). The share of traffic deaths involving

pedestrians in the U.S. has been an increasing trend from 2004 through 2013

(Dangerous by Design, 2014). Even more alarming is the overrepresentation of the

southeastern region of the U.S. when the top 50 metropolitan areas are ranked by

Pedestrian Danger Index (PDI) shown in Table 2. The PDI gives and approximation

of the likelihood of a person on foot being struck and killed by a vehicle based on

the best available measure of how many people are likely to be walking and the

number of pedestrian fatalities in the past five years of data. The states in the

southeastern U.S. account for eight of the top ten worst metropolitan areas, and

each of the top six are in the southeastern states. Florida is currently the worst

among the PDI rankings, home to the top four most dangerous metropolitan areas

for pedestrians (Dangerous by Design, 2014). These alarming pedestrian fatality

rates, and overall traffic crash fatality rates in the southeastern region of the U.S.

leads to the obvious question of why. Is it the demographics of the southeastern

U.S., the physical design, the governmental policies or lack thereof, the

environmental variations of the southeastern U.S. from the rest of the nation, or

maybe it is all of the above mentioned reasons and more that the southeastern U.S.

has such alarming fatality rates.

Table 2 Pedestrian Danger Index Ranks in Southeastern Region

Rank

Metro Area

PDI

Orlando-Kissimmee, FL

244.28

Tampa-St. Petersburg Clearwater, FL

190.13

Jacksonville, FL

182.71

Miami-Fort Lauderdale-Pompano Beach, FL

145.33

Memphis, TN-MS-AR

131.26

Birmingham-Hoover, AL

125.60

Atlanta-Sandy Springs-Marietta, GA

119.35

Charlotte-Gastonia-Concord, NC-SC

111.74

Nashville-Davidson Murfreesboro-Franklin, TN

100.79

Raleigh-Cary, NC

100.35

Data Source: Dangerous by Design, 2014

2.2.2 Demographics

Unfortunately, the higher than average fatality rates are not the only

characteristics that are present in the southeastern U.S. When the intergenerational

upward mobility of children was calculated based on the mobility of their parents it

was found that the southeastern region of the U.S. stands out very distinctly from

the rest of the country as being much less upwardly mobile than other regions of

the nation (Chetty, et. al., 2014). According to Chetty, et al. intergenerational

mobility is the degree to which a child’s social and economic opportunities are

dependent upon their parents’ social and economic status. The U.S. Census Bureau

also shows southeastern poverty levels to be among the highest in the nation, with

poverty shares of population ranging from more than 17 percent in Florida to nearly

24 percent in Mississippi (Small Area Income and Poverty Estimates, 2013). The

national share of population below the poverty line is less than 16 percent. This

thesis sets out to determine the varying socio-economic and socio-demographic

characteristics in the region that may help to explain why the fatality rates are

consistently above the national averages. The main argument defined is that

poverty is greatly correlated to propensity for risk taking behavior, and risk-taking

behavior is linked to driving characteristics that make the roadway network in the

southeastern U.S. inherently more dangerous.

The approximate 2013 population of the U.S. was 316 million people, of

which 23.3% were under 18 years of age, and 14.1% were over the age of 65,

leaving 62.6% of the population in the typical working adult age range (Census,

2013). As of 2013, 50.8% of the U.S. population was female. Racially the majority

of the U.S. was White, accounting for a 77.7% share of the population, followed by

13.2% Black or African American, 5.3% Asian, 1.2% American Indian, 0.2% Native

Hawaiian or other Pacific Islander, and 2.4% are two or more races. When the

average was taken over the past 5 years of data, from 2009 to 2013, 86.0% of the

U.S. population over age 25 was a high school graduate or higher, and 28.8% of

that same age group had a Bachelor’s degree or higher. From 2009 to 2013, the

average share of the population below poverty level was 15.4% (Census, 2013).

The approximate 2013 population of the southeastern U.S. was 63 million people,

of which 22.6% were under 18 years of age, and 15.4% were over the age of 65,

leaving 62.0% of the population in the typical working adult age range as shown in

Figure 1 (Census, 2013). As of 2013, 51.2% of the southeastern U.S. population

was female. Racially the majority of the southeastern U.S. population was White,

accounting for a 71.5% share of the population, followed by 21.6% Black or African

American, 2.2% Asian, 0.4% American Indian, less than one tenth of one percent

are Native Hawaiian or other Pacific Islander, and 2.1% were two or more races as

shown in Figure 2. Total shares of race do not add to 100 percent due to errors

associated with rounding. When the average was taken over the past 5 years of

data, from 2009 to 2013, 84.8% of the southeastern U.S. population over age 25

was a high school graduate or higher, and 25.5% of that same age group had a

Bachelor’s degree or higher as shown in Figure 3. From 2009 to 2013, the average

share of the southeastern population below poverty level was 17.7% (Census,

2013). In general, the southeastern U.S. has less working age adults, more young

people, more elderly, and a larger share of Black or African American population

than the rest of the nation. The southeastern states are also disproportionately less

educated, and have a larger share of the population below poverty than the

national average. Each of these demographic characteristics has an impact on the

general safety of the roadway network in the southeastern U.S.

Figure 1 Age Group Comparisons. Data Source: U.S. Census Bureau

Figure 2 Race Comparisons. Data Source: U.S. Census Bureau

23.3

22.6

62.6

62.0

14.1

15.4

100

Total U.S.

Southeastern U.S.

and older

Under 18

77.7

71.5

13.2

21.6

5.3

2.2

100

Total U.S.

Southeastern U.S.

or more races

Pacific Islander

American Indian

Asian

African American

White

Figure 3 Education Comparisons. Data Source: U.S. Census Bureau

According to Polzin, demographics, more specifically race and ethnicity, have

been found to have strong distinctions in mode choice, and some mode choices are

distinctly safer than others. Hispanics are twice as likely to carpool as White

NonHispanics, and almost three times as likely to use transit to commute. Transit

commute share is the greatest among the Black population. Hispanics have the

highest commute share of bicycle travel, and Asians have the highest share of walk

as a commute mode choice (Polzin, et al., 2013). While this thesis focuses primarily

on fatality rates of the overall roadway network, it is important to understand the

connection between share of alternative mode users and the safety of a roadway

network. Vulnerable road users such as pedestrians and bicyclists experience the

highest risk when traveling on the roadway network. Non-motorists have the

inherent exposure risk, without the shell of a vehicle surrounding them. In addition,

they are traveling at slower speeds than vehicular modes, and thus cannot react as

quickly to avoid dangerous situations as one with a motor vehicle could. On the

contrary, transit use is inherently a safer mode choice than personal vehicle travel.

Transit travel injury or fatality rate is about one-tenth the personal vehicle travel

14.0

15.2

57.2

59.3

28.8

25.5

100

Total U.S.

Southeastern U.S.

Bachelor's Degree or

Higher

High School to

Associate's Degree

Less than High School

rate, and transit oriented communities have about one-fifth of the per capita fatality

rates as automobile oriented communities (Litman, 2014). The safety increase with

transit is not only associated with the larger size of the vehicle which can absorb

more of the shock of an impact should an incident occur, but also with the

improvements of physical design of the infrastructure which often corresponds with

transit-oriented development (TOD).

2.2.3 Physical Design

TODs increase the safety for communities as a whole, most notably by giving

pedestrians and transit users “defensible spaces” and providing settings with fewer

hazards for non-motorists (TCRP Report 102, 2004). The change in physical design

of a network that implements TODs includes traffic-calming characteristics such as

speed humps, streetscapes with wide grassy medians and reduced width lanes to

further reduce the speed of vehicles. In addition other mixed-use land-use patterns

that are incorporated increase the overall safety of the network for all users.

Communities in the southern Sunbelt that were established and grew

dominantly in the post-war period were designed to accommodate automobile travel

and sprawling low density neighborhoods (Dangerous by Design, 2014). These

designs are not only more dangerous for vulnerable road users, but are also

inherently more dangerous for all modes of travel because of the tendencies to have

higher speed limits, wider lanes, and accommodate larger capacities. The arterial

focused designs are concentrated on the purpose of serving quick automobile travel.

Although the arterials are commonly populated with shopping centers, office parks,

and residential access, people or communities are too often the secondary

concerns, or non-concerns in the design process of the arterials (Dangerous by

Design, 2014).

The physical design however, is not only considering the safety of the

vulnerable road users, but also the effect of physical design on the fatalities of

vehicular road users. “Since 1979, the proportion of traffic fatalities involving the

collision with a fixed object on the roadside has fluctuated between nineteen and

twenty-three percent (Delgado, 2011).” In the southeastern region, it was

calculated by Wang (2006) that roadside hazards accounted for at least one death

in forty percent of all crashes in which a vehicle occupant was fatally injured. With a

ten percent higher share in the southeastern region, it is obvious that roadside

hazards play a role in the increase in fatality rates in the southeastern region.

2.2.4 Governmental Policies

The federal, state, and local laws and policies put in place have significant

impacts on the transportation system, with growing awareness on the emphasis on

quality of life and health, in addition to the efficient movement of people and goods

(CDC, 2010). The CDC has made several recommendations to increase the safety of

the transportation network. One recommendation includes providing the individual

states with incentives to implement and strengthen laws associated with: seat-belt

use, booster-seat use, helmet use for motorcyclists and bicyclists, impaired and

distracted driving, reducing speed limits, comprehensive graduated driver licensing,

improved roadway design, increased education on safe driving, bicycling, and

walking, and community designs that promote reduced speeds in neighborhoods.

The CDC also recommends increasing support for new technologies to improve the

safety of motor vehicles and enable vehicles to withstand crashes with lower risk of

injury to the occupants and non-occupants, and also new technologies to prevent

alcohol impaired driving (CDC, 2010).

2.2.4.1 Seat Belt Policies

Wearing a seat belt reduces the likelihood of occupant death by roughly sixty

percent, when involved in a motor vehicle accident (Levitt and Porter, 1999). When

states pass laws to require the use of front-seat occupants to wear seat belts, the

average seat belt usage rates increase about seventeen percent (Cohen and Einav,

2001). Table 3 displays the southeastern states and their primary seatbelt policies

in addition to some national comparisons. Some states currently have secondary

seat belt laws, which require the driver to be pulled over for an unrelated offense,

and do not allow law enforcement to stop a vehicle for the sole purpose of not

wearing a seat belt. In other states, seat belt laws are primary offenses, meaning

law enforcement can stop a vehicle for the sole purpose of not wearing a seat belt.

It was estimated that if all states which had secondary seatbelt enforcement laws in

2001 changed their laws to primary, the national seat belt usage rate would have

increased from 68% to 77% and in turn would have saved approximately 500 lives

per year (Cohen and Einav, 2001). When comparing seat belt usage between states

with primary seat belt laws and states with secondary or without seatbelt laws, in

2013 the share of seat belt usage varied by 11% (Traffic Safety Facts, 2014). The

trends from 1995 to 2013 show a reduction of Daytime percent unrestrained personal

vehicle occupant fatalities associated with an increase in overall safety belt usage

(Traffic Safety Facts, 2014). Booster seat laws have similar safety influences as seat

belt laws. Booster seat laws were associated with a 17% reduction in the per capita

rate of children who sustained fatal or incapacitating injuries from motor vehicle

accidents (Eichelberger, A.H., Chouinard, A.O., Jermakain, J.S., 2012).

Table 3 Primary Seat Belt Policies

Alabama

Front Seat only

Florida

Front seat and minors in rear seats

Georgia

Front seat and minors in rear seats

Kentucky

All

Mississippi

Front Seat only

North Carolina

Front (secondary for rear seat)

South Carolina

All

Tennessee

Front Seat only

Nationally

33 states have primary front seat laws (17 primary all seats)

Data Source: GHSA

2.2.4.2 Helmet Policies

Motorcycle helmet laws vary among the states, with some states having

universal helmet laws, some states having partial helmet laws, and a few states

having no helmet laws at all as of February 2015. A universal helmet law is

inclusive of all motorcycle riders regardless of age, experience, or insurance

coverage. Many states also extend the helmet laws to cover riders of low-powered

cycles which are not classified as motorcycles, such as mopeds and scooters, while

some states specify the extension to low-powered cycles with specific minimum

engine displacement, braking, and/or top speed specifications. A partial helmet law

is generally age dependent, with some states having additional dependent criteria

as well. As of February 2015, There are currently three states in the U.S. which

have no helmet laws at all; Illinois, Iowa, and New Hampshire. Currently 19 states

and the District of Columbia have universal helmet laws, and 28 states have partial

helmet laws. The majority of the states in the southeastern region of the U.S. have

universal helmet laws with the exception of Florida and South Carolina. The partial

helmet laws in Florida are inclusive of all riders under the age of 21 without

exception, and riders over the age of 21 who do not carry a minimum of $10,000 in

medical insurance. From 1967 to 2000 all motorcycle riders in Florida were required

to wear helmets; however the law was changed to a partial law in July of 2000. In

South Carolina, all riders under the age of 21 are required to wear a helmet, but

once the rider reaches the age of 21, there are no applicable helmet laws. From

1967 to 1980 there was a universal helmet law in effect in South Carolina, at which

time the law was reduced to a partial law covering only riders under the age of 21

(Motorcycles, 2015) . Motorcycles require important consideration when trying to

reduce overall traffic fatality rates because motorcyclists account for a

disproportionate share of traffic fatalities. In 2011 motorcyclists accounted for 14

percent of all traffic fatalities while only accounting for less than one percent of all

vehicle miles of travel (Traffic Safety Facts, 2013). Helmet laws are closely

correlated with helmet use and with motorcycle fatalities. In states where a

universal helmet law was repealed, the average helmet use decreased 41% and the

total number of motorcycle fatalities increased 38% (Use of Motorcycle Helmets:

Universal Helmet Laws, 2014). On the contrary, states that changed from partial or

no helmet laws to universal helmet laws experienced an increase in helmet use of

56% and a decrease in motorcycle fatalities of 37% (Use of Motorcycle Helmets:

Universal Helmet Laws, 2014).

2.2.4.3 Distracted Driving Policies

While the term distracted driving is defined by the National Highway Traffic

Safety Administration (NHTSA) as inattention that occurs when a driver diverts their

attention away from the task of driving to focus on another activity (Driver

Distraction Program, 2010), distracted driving laws generally only pertain to the use

of electronic devices (Distracted Driving Laws, 2015). Alarmingly, no state in the

U.S. bans all cell phone use while driving, and only twenty states prohibit cell phone

use for school bus drivers, and thirty-eight states ban cell phone use for novice

drivers. The definition of novice driver varies by state and is usually determined by

the level of licensure and/or by the driver’s age. Currently forty-four states have

texting while driving bans in place, and in the southeastern region, that is inclusive

of all states except Mississippi, which only prohibits texting while driving for school

bus drivers and drivers with learners’ permits. Distracted driving accounts for

approximately ten percent of all fatal crashes in the U.S. and nearly one in every

five (17%) injury crashes (Traffic Safety Facts, 2013). It has been demonstrated

that high visibility of law enforcement is an effective way of reducing cellphone/text

message use while driving in certain situations. This measure is reliant on increased

law enforcement presence in addition to increased media coverage about the

increased enforcement, which in turn increases the perceived risk of being caught

violating the law. This measure is of course, only effective where laws are in place

to deter drivers from distractions. While this measure has only been implemented in

a few states, Hartford, Connecticut saw a hand-held cell phone use decrease of

57% (Countermeasures That Work, 2013).

2.2.4.4 Drunk Driving Policies

Driving while drunk or intoxicated is a crime in all states in the U.S.; however

the definition of the legal limit of blood alcohol concentration (BAC) varies greatly

across states. In all U.S. states the minimum BAC to receive a penalty for driving

under the influence (DUI) is 0.08. Other states have additional penalties for

excessive BAC limits which vary from 0.10 in Rhode Island to 0.20 in Florida and

Tennessee (GHSA, 2015). Many countermeasures to deter drivers from driving

under the influence have been proven to be effective such as; revocation or

suspension of license for refusal to submit to a BAC test, publicized sobriety

checkpoints, saturation patrol of areas prone to DUI’s during specific times when

DUI’s are at the highest, generally associated with the closing of the local alcoholic

establishments (Countermeasures That Work, 2013). Mothers against Drunk Driving

(MADD) was established in 1980 when a mother lost her daughter in an accident

caused by a drunk driver, and the goal of MADD was to make the drunk driving laws

stricter (Ronis, 2014).The effectiveness of strengthening DUI laws was reflected in

the reduction of fatalities associated with drunk drivers. Between 1982 and 2005

the share of drivers in fatal crashes with BAC of 0.08 or higher decreased from 35

percent to just 20 percent (Statistical Analysis of Alcohol-Related Driving

Trends, 2008).

2.2.4.5 Speed Policies

Speed is greatly associated with the severity of a collision, especially when a

pedestrian is involved. Figure 4 shows how the percentage of pedestrian fatalities

increases with speed increments of just 10 miles per hour (mph). “Studies suggest

that a fatal pedestrian accident is 6 times less likely to happen if the vehicle speed

is 23 mph as opposed to 28 mph” (Photo Enforcement Program – Historical

Perspective, 2006). Photo enforcement programs installed to enforce speed limits in

school zones have shown to be effective in reducing the number and severity of

speeding drivers, thus making the network safer for all users (Photo Enforcement

Program – Historical Perspective, 2006). The southeastern region, highest and

lowest top speeds are shown in Table 4.

Figure 4 Pedestrian Fatality Share by Vehicle Speed. Amended from: Photo

Enforcement Program – Historical Perspective, 2006. City of Renton, WA.

Table 4 Top Speeds

Top Speed by St

ate

All SE States

Lowest (Alaska)

Highest (Texas)

Data Source: motorists.org

100

mph

Vehicle Speed

2.2.5 Environmental Variations

There are several factors to be considered when trying to distinguish the

reasoning behind the higher than average fatality rates in the southeastern region,

and one factor which cannot be over looked is the environmental differences. The

southeastern region of the U.S. has much milder winters than the rest of the

nation. If we compare the average high temperatures in the historically coldest

months of the year, December through February, in Tampa, Florida the average high

temperature is in the 70’s, in Montgomery, Alabama the average high temperature

is in the 60’s, and in Memphis, Tennessee the average high temperature is in the

50’s. However when those average high temperatures are compared to some states

in the northern parts of the U.S. a much colder trend is apparent. The average high

temperature in those same months in St. Paul, Minnesota is in the 20’s, which is

similar to Bismarck, North Dakota. And the states in the Northeastern region of the

state average highs are in the 30’s and low 40’s (U.S. Climate Data, 2015). In

addition to the milder temperatures of the winter, the higher temperatures in the

summer are often reasoning to wait until the sun goes down to participate in

outdoor activities in the warmer months. For instance, one may choose to go before

sunrise or wait until sunset to go for a jog outside because those are the times

when the temperatures are milder during the summer months.

This overrepresentation of pedestrians and bicyclists in the dark may be an

explanatory factor in the increased number of pedestrian fatalities in the

southeastern region.

In general, one state in the southeastern region stands out from the rest in

visitor attractiveness, Florida, also known as the “Sunshine State”. Orlando is the

leader of all cities in the U.S. at attracting visitors, with over 48 million attracted in

2009 (Forbes, 2010). With the warm weather, tons of amusement parks, over

thirteen hundred miles of coastline accompanied by breathtaking beaches and the

largest cruise market in the world, Florida is major attractor of visitors, both

domestically and internationally (Census, 2010). In 2013 Florida received nearly 94

million visitors (VisitFlorida, 2015). With this amount of tourists visiting the state

each year, there is a disproportionate share of unfamiliar drivers on the roadway

network in the Sunshine State. It has been found that the volume of tourists or

visitors who are unfamiliar drivers are proportionate to the number of traffic

fatalities in the region, meaning as the number of unfamiliar drivers increases on

the roadway network, so too does the number of traffic fatalities

(Somasundaraswaran, 2010). The combination of outdoor inviting weather, and

increased share of unfamiliar drivers on the roadway play a significant role in the

increased number of pedestrian and bicycle fatalities in the state of Florida

(Dangerous by Design, 2014). This same reasoning could also be extended to

motorcycle fatality rates in the state, with weather conditions which make it inviting

to ride a motorcycle year round, and a greater proportion of unfamiliar drivers on

the network, it is a recipe for increased fatalities (Reyes, 2013).

2.3 Risk-Taking Characteristics

Risk-taking characteristics are descriptive of behavior which can be perceived

as risk-taking behavior. For this thesis, the adopted definition of risk-taking behavior

is defined as “the tendency to engage in behaviors that have the potential to be

harmful or dangerous, yet at the same time provide the opportunity for some kind

of outcome that can be perceived as positive(Tull, 2014).” Risk-taking

characteristics may not only be tied to an individual’s specific driving behavior, but

also to the propensity to procrastinate on general upkeep of the vehicle, such as

changing tires, brake pads, and oil on a routine basis. As the upkeep of a vehicle is

neglected, the chances of vehicle failure of some kind increase, and the risk of

incident also increases. In this thesis the types of risk-taking characteristics will be

limited to the following; credit score, smoking and tobacco use, alcohol use, drug

use, mental health, obesity, general health, and educational attainment. A credit

score is a risk-taking characteristic because it gives an idea of the financial

responsibility of the credit holder, and thus a lower credit score is often tied to an

individual who takes more financial risks than an individual with a higher credit

score. Tobacco use, alcohol, and drug use are all associated with adverse health

characteristics and each present different levels of risk users are exposing

themselves to. Mental health, obesity, and overall general health may not be as

straight forward as the previous risk taking characteristics; however they are

associated with general risk. For example, someone who suffers from some form of

mental instability may be more likely to engage in risky behavior, which they may

not be able to control, and while the mental disability is not a risk they chose, it is

still a contributing factor to their life (Guiang, 2014). Finally education levels are

generally correlated with income levels, specifically of the family. It has been shown

that youth from low income families are less likely to graduate from high school, get

a degree, and are more likely to engage in risky behaviors (Kent, 2009).

2.3.1 Credit Score

Beginning with perhaps the most controversial of all risk taking or cultural

behavior traits, credit score is already used in 47 states within the U.S. as a factor

to predict the consumer’s likelihood and amount of insurance losses, with the

exception of Massachusetts, Hawaii, and California. To clarify credit-based scoring is

not the sole factor in determining the amount of an individual’s automobile

premium, it is just one of several factors used in the calculation of risk, which varies

by insurance company, explaining why the same person may pay a much different

premium for the same automobile depending on which insurance agency their

automobile insurance is issued through. According to the National Association of

Insurance Commissioners (NAIC) credit-based scores were introduced in the 1990’s

and 95% of auto insurers that are able to use credit-based insurance scores do so,

citing that research shows a correlation between credit scores and insurance losses.

A credit-based insurance score is not exactly the same as a general credit score, as

it only uses specific aspects of an individual’s credit score to rate them, such as;

payment history, outstanding debt, credit history length, pursuit of new credit, and

credit mix (NAIC, 2012). The Insurance Information Institute (III) states, “Studies

have shown a 99 percent correlation between insurance scores and loss ratio,”

meaning individuals with low credit-based insurance scores, account for a higher

share of the claim dollars paid. In July of 2007 the Federal Trade Commission (FTC)

released a report to Congress on the impacts of credit based scores on consumer

automobile insurance, which determined that score based predictors were much

more accurate than using race and ethnicity. The report also states while there is no

definitive reasoning as to why the low credit scores, on average, lead to higher

amounts and rates of claims, the report did give some reasoning as to why strong

credit history likely leads to reduced claims. The FTC report states:

“A strong credit history, however, might indicate that a consumer has taken

care in managing his or her financial affairs by avoiding loans that might be

difficult to repay, avoiding high balances on credit cards, making sure that

bills are not misplaced and are paid on time, etc. A consumer who is prudent

in financial matters may also be cautious in other matters related to

insurance, such as being more likely to put time, effort, and money into

things like car and home maintenance, cautious driving habits, etc. An

overall inclination to be prudent may lead a consumer both to have a strong

credit history and file fewer insurance claims.”

It should be noted that this credit-based insurance scoring reasoning is based on an

average, and is not completely explanatory for each individual circumstance;

however it is the best way currently available to predict losses, or more precisely,

loss ratios. It should also be noted through this same FTC study, that 59 percent of

consumers are predicted to have lower premiums when credit-based scoring is used

instead of previous methods of calculating premiums without using credit as a

factor.

2.3.2 Smoking and Tobacco Use

Other risk-taking behavior includes one’s choice to use tobacco, alcohol, or

illicit drugs. These decisions can be greatly tied to cultural effects, as the probability

of an individual to choose to partake in these specific behaviors is often much

greater when the individual was exposed to these behaviors throughout their

childhood. The Center for Disease Control and Prevention (CDC) reports that the

probability of use of alcohol and other drugs is correlated to the probability of

tobacco use. In addition, 99 percent of tobacco users report trying tobacco before

their 26th birthday, indicating that it is extremely rare for an adult who never tried

tobacco to pick up the habit after their 25th birthday. This is also indicates that

teens and young adults are generally more impressionable, have less understanding

or care of the consequences related to their behavioral choices, and thus more

likely to take risks in general. Many teens who are aware of the consequences

associated with smoking, drug use, or alcohol use think that the consequences do

not apply to them, which can also be mirrored in their choice to behave in a risky

manner while driving, because again, they think that they are the exception; that

crashes happen to everyone else, but not to them. Social groups such as peers and

family members have a great influence on the probability of youth to use tobacco,

because they tend to see it as acceptable or normal. Other influences demonstrated

to affect tobacco use include low socioeconomic status, low education levels, low

self-esteem, and exposure to tobacco advertising. Smoking and other use of

tobacco is categorized as risk-taking behavior because it is known to have adverse

effects on health ranging from bad breath and foul-smelling clothes to diseases

such as cancer and even death. The CDC reports that smoking harms nearly every

organ of the body and is tied to nearly one in every five deaths in the U.S.

Individuals who are willing to take the risks associated with the use of tobacco are

also likely to take risks in other aspects of their life, including the way they drive. It

is arguable that if an individual is willing to smoke, they have less regard for the

human life, and thus are more likely to drive erratically, aggressively, while

distracted, or in other circumstances that would otherwise be considered risky.

2.3.3 Drug Use

Drug use, while highly correlated with tobacco use. The U.S. Federal

Department of Health and Human Services has an agency called SAMHSA, which

stands for Substance Abuse and Mental Health Services Administration, whose

mission is to “reduce the impact of substance abuse and mental illness on America’s

communities.” Drug use is known to have adverse health effects on the user, and

thus when an individual chooses to use any drug, they are in turn willing to put

their body at risk. This general idea has inspired several studies, many of which

compare adolescent substance use with risky sexual behavior. It has been found

that substance using adolescent females had higher rates of sexually transmitted

diseases, and more pregnancies than non-abusing adolescent females (Tapert, et.

Al., 2001). To focus back to the risk of interest of this thesis, driving and fatal crash

risk, it has been found that the “use of drugs while driving tends to have a larger

effect on the risk of fatal and serious injury accidents than on the risk of less

serious accidents” (Rune, 2013). One aspect of risk-taking is sensation seeking, or

trying to attain a feeling of euphoria or excitement. Sensation seeking as a

construct predicts alcohol use, smoking, sexual behavior, substance use, and

driving under the influence (Ames, et al., 2002). Several studies have associated

sensation seeking with not only substance use/abuse, specifically marijuana, such

as Satinder and Black (1984), but also associated sensation seeking with erratic

and aggressive driving such as Jonah (1997). With current ever-changing laws

associated with marijuana use at the present time, it is important to note that “the

more prevalent and socially acceptable a drug is in a particular population, the less

likely it is to find a relationship between the use of that drug and sensation-seeking

orientation (Satinder and Black, 1984).” This is particularly important when

referring to marijuana use in the U.S., because as of 2015, twenty-three states and

the District of Columbia have legalized some type of marijuana use, whether

medical or recreational (governing.com, 2015).

2.3.4 Mental Health

Adults with symptoms of depression, just one type of mental disability, were

significantly more likely to engage in certain risk behaviors than persons without

depression (Mental Health in the United States, 2005). A study by Prochaska et al.

studied 693 people with serious mental illness for the number of risk-taking

behaviors they participated in. The average number of risk-taking behaviors the

sample of serious mentally ill participated in was 5.2 with 100 percent using

tobacco, 68 percent eating high fat diets, 46 percent using marijuana, and 26

percent binge drinking. The seriously mentally ill population is more vulnerable and

thus more impressionable into making risky behavior choices such as smoking

(Prochaska, et al., 2014) Chwastiak et al. conducted a report on veterans with

serious mental illness and found there is an increased risk for mentally ill patients

to have multiple health risk behaviors. The NHTSA held a symposium in 1999 where

the Executive Director of the Institute for Mental Health Initiatives, Mr. Michael

Benjamin, addressed the topic of aggressive driving. Mr. Benjamin mentioned that

aggressive driving is a characteristic of high-risk-behavior individuals often with

uncontrolled anger and violence, requesting that aggressive driving improvement

strategies be incorporated into driving education and anger management classes

(NHTSA, 1999).

2.3.5 Obesity

Obesity is defined as having an excessive amount of body fat, measured as a

body mass index (BMI) of 30.0 or higher, calculation is shown in Equation 1 (Mayo

Clinic, 2014).

Equation 1

According to the researchers at the Mayo Clinic, obesity is usually a result of a

combination of genetics, lifestyle, inactivity and unhealthy dieting, although even

when ramped within a family, diet and exercise can mitigate and control onset of

obesity. Obesity is often a result of choices made by an individual, whether

conscious or subconscious, and is related to several other quality-of-life altering

characteristics such as depression, shame, social isolation, lower work achievement,

etc. (Mayo Clinic, 2014). The mentioned mental characteristics are known to affect

ones increased propensity to engage in risky behavior, thus obesity can also be

associated with an individual’s increased propensity to behave in a risky manner.

In 2003, Arbabi S, et al. wrote “The Cushion Effect” in which they stated

that, “In motor vehicle accidents, obese individuals are less likely than

normalweight individuals to suffer abdominal and pelvic injuries.” However a

nonlinear relationship was discovered in the study, and overweight people tended to

have lower fatality risk than normal weight and obese patients, with an implication

that the cushion effect is optimal when the patient has not yet reached the level of

obesity, where other factors affect the health of the individual (Arbabi, 2003). Since

that publication another study (Cramer C. et al.) indicates, “More protective soft

tissue leads to changing absorption and distribution of energy, thereby serving as

protection for abdominal organs.” This could be a possible indication that obesity,

while leading to more risk-taking behavior may not have as large effect as initially

considered in overall average vehicular fatality rates by geography.

2.3.6 General Health

General overall health, as assessed in the Behavioral Risk Factor Surveillance

System (BRFSS) is self-determined, however is surveyed along with several other

behavioral health questions, thus is assumed to be relatively related to chronic

diseases, healthy behaviors, and access to preventative health-care services. The

top five causes of death in the U.S. according to the CDC are heart disease, cancer,

lower respiratory diseases, accidents, and stroke; at least four of which are effected

by risk-taking health decisions made by an individual. Accidents’, meaning any

unintentional injury, is the only one that is not necessarily effected by health related

risk-taking behavior (CDC, 2015). The most common known causes of heart

disease, cancer, respiratory diseases and stroke are unhealthy diet, lack of

exercise, being overweight, and smoking (Mayo Clinic, 2015), all of which are due

to risk-taking choices made by an individual, with some added risk from air

pollution, allergens, and occupational agents (WHO, nd).

2.3.7 Educational Attainment

Obtaining an education is not a risk taking behavior, but rather the lack of

educational attainment is a risk taking behavior. Referring back to the established

definition of risk taking presented earlier in the introduction of this thesis, an action

that can be dangerous or harmful with the perception of some type of positive

outcome; dropping out of school may be viewed as a positive outcome by a young

individual who experiences less responsibility and more free time, however their

future success can be harmed by the action of dropping out of school. “People with

more education are likely to live longer, to experience better health outcomes, and

to practice health-promoting behaviors such as exercising regularly, refraining from

smoking, and obtaining timely healthcare check-ups and screenings (Education

Matters for Health, 2009).” Educational attainment and poverty are interrelated,

with the Federal Educational Testing Service (ETS) reporting that adults who grew

up in poverty are likely to achieve a lower level of educational attainment, have

lower earnings when they enter the work force, pay less in taxes, and “exhibit

negative behaviors and health outcomes that add a burden to the nation’s

economy” (ETS, 2013). In other words, the less education a person completes the

more likely that person is to act in a risky manner, including breaking laws and

endangering their health. Education has also been linked to sense of control,

indicating that as an individual’s education levels increase they gain a greater

perception of how their actions influence their circumstances (Education Matters for

Health, 2009). Thus, the individual will be less likely to act in a risky manner, given

a better understanding of the possible outcomes associated with the risky behavior.

As education levels increase, the healthy behavior levels such as healthy diet,

exercise, and lack of smoking, increase, and life expectancy increases (Maralani,

2014).

CHAPTER 3:

DATA COLLECTION AND METHODOLOGY

The following sections of chapter three review the data collected from each

source, along with the methodologies used in the collection of each set of data. The

large range of data sources resulted in varying methodologies. The final sections of

chapter three describe the methodology used in the data analysis performed.

3.1 Fatality Rates

Fatality rates can be defined as fatalities per one hundred million vehicle

miles of travel (VMT), to account for exposure. Fatality rates can also be defined as

fatalities per population, to give an overall risk per person. In a previously

mentioned report, Dangerous by Design, mode specific fatality rates are calculated

by approximate exposure. For example, pedestrian volumes are not adequately

reported, so often an estimation of propensity to walk as a mode choice is

estimated along with historical pedestrian fatalities to calculate a pedestrian fatality

rate. To account for regression to mean, the fatality rates used in this thesis are

always an average of three years of data. According to De Pauw, et al. regression to

mean is referring to the phenomena that high crash frequency periods are often

followed by low crash frequency periods which occurs due to the randomness

associated with vehicular crashes. Regression to mean uses multiple years of data

to account for fluctuations which over time generally represent a mean value.

Without using multiple years of data to account for regression to mean, the data

may be under-representative or over-representative of the actual fatality rates

(DePauw, et al., 2014.) The following sections describe the data which was collected

and the methodology of each source of data that was used in the determination of

fatality rates.

3.1.1 Fatality Analysis Reporting System

The Fatality Analysis Reporting System (FARS) was the data source used to

gather and examine all of the fatality rates, and all of the specific characteristics of

the fatalities. The FARS (2013) reports fatality information and characteristics in

eighteen data files, three of which were used to evaluate the given data, the crash

level file, the vehicle level file and the person level file. The levels of data are shown

in Table 5.

Table 5 Levels of Data in FARS

Name

Variables Includes

Crash

crash level

Vehicle

vehicle level

Person

individual specific

Parkwork

parked and working vehicles

cevent

sequence of all qualifying events, e.g. crossed center line

vevent

sequence of events for each vehicle

vsoe

sequence of events for each in-transport vehicle

distract

driver distractions

factor

vehicle circumstances

drimpair

physical impairment of driver od motor vehicle

nmimpair

physical impairment of non-motorist, e.g. pedestrian

maneuver

actions taken by driver to avoid something or someone

nmprior

what non-motorists are doing prior to crash

Table 5 Continued

safetyeq

safety equipment used by non-motorists

violatn

violations charged to drivers

vision

circumstances which may have obscured the driver’s vision

damage

all areas of vehicle damaged in the collision

Data Source: Fatality Analysis Reporting System, 2013

There are several characteristics which are present in several levels of reported data

files used such as; the state the crash occurred in, the date, time, and location of

the crash, and the roadway function where the crash occurred, as shown in Table

Table 6 Data Level Inclusivity

Data Level

Entries

Includes

Crash

1 per crash

number of: vehicles involved, occupants, fatalities and

weather, lighting conditions, manner of collision, first

harmful event, related factors, work zone details,

school bus, rail crossing, time of EMS arrival, if driver

was drunk

Vehicle

1 per vehicle

VIN number, weight class, body type of vehicle,

number of occupants, travel speed, initial contact

point, roll over, fire, fatalities, registration state,

license state of driver, type of license, driver height

and weight, previous convictions

Person

1 per person

injury level, age, sex, race, person type, injury

severity, seating position, restraint/helmet use or

misuse, airbag deployment, ejection and ejection path,

alcohol and drug tests, time of death, other related

factors

Data Source: Fatality Analysis Reporting System 2013

The FARS is produced through the National Highway Traffic Safety

Administration (NHTSA), and is a census compilation of fatal motor vehicle crashes

from all 50 states, the District of Columbia, and Puerto Rico since 1975. In order for

an incident to be reported as a FARS case, it must have occurred on a roadway that

is customarily open to the public, and must have resulted in at least one related

fatality within 30 days of the crash. Non-fatal crashes are not reported in the

database, nor are crashes that occur on private property or crashes where the

injured individual passes away more than 30 days after the collision. FARS data is

gathered by representatives in each state from police reports, death certificates and

coroner reports, vehicle registration and driver licensing files, hospital records, EMS

reports, vital statistics, and other state records; however no personal identifying

information is recorded to comply with the Privacy Act and allow all data files to be

made available to the public. The number of entries per state is dependent on the

aforementioned requirements to qualify as a fatal crash.

Fatality rates were calculated using the number of fatalities over a three year

period, to account for regression to mean in fatality levels, divided by the number

of vehicle miles of travel (VMT) for the same three years, obtained from the FHWA

Highway Statistics Series by state as shown in Equation 2.

Equation 2

The southeastern region calculations used the sum of the fatalities in the

eight southeastern states divided by the sum of the VMT for the corresponding

states, again using the three year totals to account for regression to mean

variations, as shown in Equation 3.

Equation 3

where i=1:8 is representative of each southeastern state.

In addition to fatalities, several other characteristics such as age and gender

of driver involved in fatal crash, lighting condition of fatal crash, and vehicle

occupancy of fatal crashes which are reported in the above mentioned files were

used to determine if correlations existed, and the levels of the correlations between

fatality rates by geography. This was done using a file that was created by

combining the same years, 2010 through 2012, that were used to calculate the

fatality rates for consistency. These files were then manipulated using Excel and

SPSS to find similarities, differences and correlations by state, national, and

southeastern regional levels.

3.1.2 Census Bureau Population Data

The U.S. Census Bureau conducts annual American Community Surveys

(ACS) data collection in all 50 states, the District of Columbia, and Puerto Rico. The

ACS uses a series of monthly samples to produce annual estimates at the state

level. The ACS is inclusive of not only the housing unit population but also, as of

2006, the group quarter population as well. In 2010, approximately 2.9 million

housing unit addresses were selected; in 2011 the sample size was increased to

3.54 million addresses per year. The sampling of the data is collected in two phases,

and the first phase consists of two stages. The first stage of the first phase consists

of dividing each county into five sub-frames, which are each a representative

sample of addresses in the county. The addresses are sorted and assigned to one of

the five sub-frames, to ensure that no household is sampled more than once in a

five-year period. The second stage of the first phase assigns the selected addresses

into a month within the year of the data collection. The sampling rate varies by the

size of the blocks in the sampling entity, with the blocks in the smallest sampling

entities at a rate of 15 percent, blocks in small sampling entities at a rate of

10percent and blocks in medium sampling entities at a rate of 7 percent. Blocks in

large sampling entities with predicted levels of completed interviews greater than

60 percent can have a sampling rate as small as 0.5 percent. The second phase of

sample selection gathers non-mailable (mail is directed to a post office box, or

address is incomplete) and non-responding addresses with computer assisted

telephone interviews and computer assisted personal interviews. In 2012, a total of

2.2 million of the estimated 132.5 million housing units completed the interview,

yielding a sample population of approximately 1.7 percent which covers about 93

percent of the population

(Census, 2014).

Once the sample is determined and the data is collected, “weights are used

to bring the characteristics of the sample more into agreement with those of the full

population by compensating for differences in sampling rates across areas”(Census,

2014). The ACS uses ratio estimation to make more precise estimations of

population by age, sex, race, etc. Adjustments are made to account for incomplete

interviews and vacant housing units. Multiyear estimation methodology involves

reweighing data over a three or five year period by pooling the data and using a

derivation of multiyear controls (Census, 2014).

The U.S. Census bureau was used for all population estimates and

calculations. This includes share of population calculations for specific age groups,

race, sex, and for weighing percentages of rates in the southeastern region. When

the population data was used from the U.S. Census Bureau a three-year estimate

from 2012 was used, to properly account for the corresponding data and fatality

rates which also used three years of data for all calculations.

3.1.3 Federal Highway Administration

The U.S. Department of Transportation has a department called the Federal

Highway Administration (FHWA) which publishes data and statistics at the national

and state level. One publication in particular which was used extensively for the

collection of data for analysis, is the annual Highway Statistics Series which

contains statistical information on bridges, highway infrastructure and travel,

travelers, vehicles, motor fuel, revenue and expenditures, conditions and safety,

and performance indicators (Highway Statistics Series, 2013). The majority of the

data reported in the Highway Statistics Series is submitted by individual states and

is analyzed for consistency against previous year’s data and against national data,

and is then prepared and processed through the Office of Highway Policy

Information within the FHWA.

The highway travel statistics reported in the Highway Statistics Series collects

annual average daily traffic (AADT) and centerline mileage in accordance with the

Highway Performance Monitoring System Field Manual (HPMS) (Highway Statistics

Series, 2013). In accordance with HPMS, AADT of all sections of interstate, National

Highway System and principal arterials are collected on a 100 percent basis;

however for minor arterials and collectors, sample AADT’s are collected and

expanded in accordance with HPMS. The per capita VMT in the

Highway Statistics Series is based on information from three sources, the

Nationwide Personal Transportation Survey, the Truck Inventory and Use Survey,

and the National Transportation Statistics report. Data from the Truck Weight Study

are collected in each state via weigh-in-motion systems.

In this thesis, the FHWA’s publications of the Highway Statistics Series were

used to calculate the total VMT by state, and were summed to calculate

southeastern region and national totals. The VMT data was subsequently used as a

denominator in the calculations of the fatality rates by state, southeastern region,

and at a national level (Highway Statistics Series, 2013).

3.2 Descriptive Data

Descriptive data collection was an imperative process in the analysis

performed to construct this thesis. The fatality rates that were calculated using the

above mentioned methodology were set as the dependent variable and the

descriptive characteristics that were collected and calculated were used as

independent variables to model the variation in fatality rates by geography. The

characteristics and the methodologies for each data source are below.

3.2.1 Federal Highway Administration

In addition to the VMT statistics used to calculate the dependent variable,

fatality rates, FHWA was also used to collect state level descriptive statistics such as

licensed drivers by age group, registered vehicles, and some ratios of these

statistics were calculated such as registered vehicles per licensed driver, which were

considered for use as descriptive independent variables in the fatality rate linear

regression model, presented later in this thesis, but were found to not be correlated

at an integrated level.

The licensed drivers’ statistics reported in the Highway Statistics Series

collects data from individual state departments of motor vehicles who issue drivers

licenses. There are some specifics about this data which should be noted. The ratio

of licensed drivers to total population should never be 1 due to the population who

choose not to become licensed or have mental or physical disabilities that limit their

ability to obtain a driver’s license. However the ratio may be higher than expected

due to some extenuating circumstances. There is a chance that some drivers may

be doubly counted if, for instance, they move from one state to another and get

licensed in their new state of residence; they will be double counted until their

original state license expires or until they relinquish that license. Other factors that

may not be represented in the statistics are drivers who drive without obtaining a

license, drivers who fraudulently obtain multiple licenses, and deceased individuals

whose licenses will usually be counted until expiration (Highway Statistics Series,

2013).

Vehicle registrations are another variable which are reported to the FHWA by

individual states, and states practices vary widely throughout the U.S. All states use

a vehicle registration renewal system which distributes the renewal workload

throughout all months of the year, with the exception of some vehicles in some

states which are renewed on a calendar year basis. For example, trucks in certain

weight classes may be registered by calendar year, while most personal vehicles are

generally registered annually in the birth month of the vehicle owner. To attempt

uniformity between states to the extent possible, information is reported as nearly

as possible on a calendar-year basis, with extra efforts put forth to exclude title

transfers and re-registrations which may occur in the pre-registration grace period.

It is possible that vehicles registered more than once in the same state, are

reported as multiple vehicles, although contrary to reporting protocol (Highway

Statistics Series, 2013).

3.2.2 Census Bureau Descriptive Demographic Data

In addition to the above mentioned population data collected from the

American Community Survey (ACS) through the U.S. Census Bureau, several other

descriptive characteristics statistics were also collected from the ACS (Census,

2014). Educational attainment, specifically the share of population over the age of

25 years who graduated from high school or obtained a high school equivalent

general education diploma, and the share of the same age group that obtained a

Bachelor’s degree or higher by state, southeastern region, and at the national level

were obtained. The split of sex of the population by geography was also explored as

a possible explanatory variable in the fatality rate variations. The share of elderly

population, those 65 years of age or older, was also determined by geography using

ACS data to try to distinguish which states had higher than national rates of aging

population. The share of residents born in their state of residency was gathered via

ACS data to give an idea of the share of population who were at least at one time

an unfamiliar driver on the roadway network, and also who might contribute to the

share of out of state visitors who would also be unfamiliar drivers on the roadway

network. Finally, the share of population below poverty level was explored as a

possible explanatory independent variable, given the poverty stricken population is

more likely to use vulnerable means of travel given the lower costs associated with

those means, such as walking or riding a bicycle.

Each of these above mentioned descriptive characteristics obtained using the

ACS data have allocation rates which are a measure of nonresponse. These

allocation rates vary by state and by variable, and range from about five percent to

about fifteen percent. The ACS methodology states that when the nonresponse rate

for a specific characteristic is high, bias can be introduced if the non-responses

differ substantially from given responses. Allocation rates are calculated by dividing

the number of nonresponses allocated for a specific characteristic in a specific state

divided by the total number of responses for that characteristic in that state times

one hundred to give a share, as shown in Equation 4.

Equation 4

3.2.3 Fair Isaac Company (FICO)

While the literature review from a previous section in this thesis states that

insurance companies use only certain characteristics of an individual’s credit score

to determine their credit based insurance score, the concept behind the correlation

of a credit score and likelihood to behave in a risky manner in other aspects in life

should be cohesive using a general credit score as well. The FICO credit score is

comprised of varying importance of five factors; payment history (35%), amounts

owed (30%), length of credit history (15%), new credit (10%) and types of credit

used (10%) as shown in Figure 5 (MyFICO, 2015). The Equal Credit Opportunity Act

(ECOA) was enacted in 1974 and makes it unlawful for a creditor to use race, color,

religion, origin, sex, or marital status or age as a basis to determine an individual’s

credit score.

Figure 5 Credit Score Factors. Data Source: MyFICO.com

The credit trends, average credit score by state, were collected through

Credit Karma, which is an internet source that provides credit scores to their

members, in addition to providing the average credit score trends by state, age and

domain. This is done by taking their members and performing crosstab analyses

with credit scores and the perspective characteristics. The result yields that the

state with the lowest average credit score is Mississippi with a score of 612, and the

state with the highest average credit score is Hawaii with a score of 659 (Credit

Payment

History

Amounts

Owed

Length of

Credit

History

New Credit

Types of

Credit Used

Karma, 2013).

3.2.4 Federal Department of Health and Human Services The

Federal Department of Health and Human Services has a department called the

Substance Abuse and Mental Health Services Administration (SAMHSA) which

conducts an annual National Survey on Drug Use and Health (NSDUH) to

population age 12 years and older. The NSDUH generates estimates on a national

state and sub-state level to describe patterns of prevalence of alcohol, drug and

tobacco use and mental health disorders in the U.S. The 2012 survey data used

in the analysis of this thesis was a compilation of data collected from 136,147

respondents in all 50 states and the District of Columbia, with an overall response

rates ranging from 45.8 percent to 75.2 percent with a median response rate of

61.5 percent (SAMHSA, 2013).

The survey state small area estimation obtained using the survey-weighted

hierarchical Bayes estimation methodology models aggregate data at the state and

national level. The Bayesian confidence interval for the NSDUH is a 95 percent

confidence interval, which is asymmetric meaning that the size difference between

the lower limit and the estimate is not equal to the size difference between the

upper limit and the estimate. The asymmetry is appropriate when dealing with

small percentages, such as those in the NSDUH, and help to adjust for

overestimating or underestimating the actual precision levels of the data. It should

also be noted that if two states have overlapping confidence intervals that does not

indicate equivalency of the state estimates. It is also noteworthy that correlated

substance use cannot be properly extracted at a small area level, but can be done

at a national level. For example, subtracting cigarette use from tobacco use at a

state level could give negative results, which would be misleading and should be

avoided, however making that same substitution at the national level is valid and

would indicate proper non-cigarette tobacco use (SAMHSA, 2013). For this

thesis, the NSDUH was used to collect and compare state and national data by age

group for marijuana use in the past year, illicit drug, alcohol, binge alcohol,

tobacco product, and cigarette use in the past month, in addition to serious mental

illness, any mental illness, thought of suicide, and depressive episodes in the past

year. All of the shares were tabulated and the correlations to fatality rates by state

were determined.

3.2.5 Center for Disease Control and Prevention (CDC)

The CDC is also under the umbrella of the Federal Department of Health and

Human Services. The CDC produces a uniform questionnaire called the Behavioral

Risk Factor Surveillance System (BRFSS) which was first conducted in the early

1980’s, and became a nationwide surveillance system in 1993. The BRFSS

conducted more than 500,000 interviews nationally in 2011, making it the largest

phone survey in the world. The data used in this thesis was from data collected in

2011 and 2012, which used more a sophisticated methodology than the previous

years. The new methodology uses iterative proportional fitting, with weights applied

through several strata including age, gender, race/ethnicity, educational attainment,

marital status, property ownership, and telephone ownership (CDC,

2013). In addition, cellular telephone interview standards have been accepted as of

2011; and due to the multiple methodology changes which occurred in 2011, any

pre-2011 data should not be directly compared to 2011 and subsequent data.

“Response rates for BRFSS are calculated using standards set by the

American Association of Public Opinion Research Response Rate Formula #4.

The response rate is the number of respondents who completed the survey

as a proportion of all eligible and likely-eligible persons. The mean survey

response rate for all states, territories and Washington, DC, in 2013 was 46.4

and ranged from 29.0 to 60.3. Response rates for states included in this

analysis had a median of 9.1% and ranged from 4.9% to 21.2%. For detailed

information see the BRFSS Summary Data Quality Report (CDC, 2013).”

The response rate Formula 4 defined in the American Association of Public

Opinion Research includes partial interviews as respondents and also allocates cases

where the eligibility is unknown. The formula used the summation of complete and

partial interviews as the numerator, divided by a denominator which is the

summation of complete and partial interviews plus the summation of the refused

non-contact, and other interviews plus the estimated proportion of cases of

unknown eligibility that are eligible multiplied by the sum of unknown if household

is occupied and unknown other interviews as shown in Equation 5 (AAPOR, 2011).

Equation 5

where RR = response rate

I = complete interview

P = partial interview

R = refusal and break-off

NC = non-contact

O = other

UH = unknown if household is occupied

UO = unknown other.

From this survey, the data collected for comparison to fatality rates by

geography included Alcohol use in the past 30 days, body mass index (BMI),

specifically BMI classifications such as underweight, normal weight, overweight, and

obese, and self-described overall general health classification, specifically

selfdescribed excellent and poor health conditions.

3.3 Methodology of Data Analysis

To initially determine the likelihood of correlation between the different

risktaking and other socio-demographic and socio-economic characteristics and

fatality rates by geography, a very broad based collection of data from multiple

sources occurred over approximately a nine month time period. The data was

collected from all of the above mentioned sources. In addition, some additional

sources were initially explored but not included in this thesis because they were

found to have insignificant correlations early in the data collection stages. For the

exploratory data collection the first correlations were calculated using Pearson

Correlation

Coefficients comparing the array of data of average fatality rates by state over the

three year period to each of the above mentioned independent variables, and if the

correlation coefficient was of a magnitude smaller than 0.30, positive or negative, it

was disregarded and considered to be of little significance to the research. Following

the collection of data, and the calculations of correlation coefficients, several

characteristics were obtained and used to complete bivariate correlation analyses

and a regression model was eventually formed to estimate the fatality rates by

specific correlated characteristics. The intention is to understand the strength of the

relationship between the respective variables and fatality rates.

3.3.1 Bivariate Pearson Correlation Analyses

Correlation coefficients are indicators of how closely plotted data fits a

straight line. The closer the correlation coefficient is to a magnitude of one, the

better the fit of the data to a straight line. Whether the correlation coefficient is

positive or negative indicates the sign of correlation. For instance, if a correlation

coefficient is calculated to be -0.9 there is a strong negative correlation between

the variables indicating as one variable increases, the opposite variable decreases

proportionately. On the other hand, if the correlation coefficient is calculated to be

0.9, then the relationship between the two variables is indicated to be a strong

positive correlation, meaning as one characteristic increases, the other

characteristic also increases proportionately. Using a 95 percent two-tailed

confidence interval a correlation coefficient is said to be of moderate statistical

significance if the magnitude of the Pearson Correlation Coefficient is larger than

0.3, positive or negative (Explorable.com). Alternatively, any Pearson Correlation

Coefficient that is less than 0.3 in magnitude is statistically insignificant at the 95

percent confidence interval. The confidence interval indicates the probability of an

estimate being representative of the individual sample 95 percent of the time. The

Pearson Correlation Coefficient is calculated as shown in Equation 6.

Equation 6

where r = Pearson Correlation Coefficient

X = fatality rate

Y = each independent variable individually

= the average fatality rate

= the average of each independent variable.

A Bivariate Pearson Correlation analysis was calculated in SPSS with each of

the above mentioned characteristics entered by state. This correlation analysis was

used to determine the characteristics which had significant correlations not only to

the fatality rates by state, but also between each characteristic. For instance, it is

known from the literature review previously performed that education levels and

poverty levels are greatly negatively correlated, with higher education levels

indicating a much lower probability of living in poverty. Given the strength of the

correlation between the two characteristics, only one should be used in the linear

regression model to estimate fatality rates by risk-taking characteristics, with the

choice of which variable to use depending on the interactions with other variables

used in the model. This type of comprehensive analysis was performed throughout

all previously listed characteristics to determine which characteristics best estimates

fatality rates by geography at a state level.

It is noteworthy that the correlation between two variables does not indicate

causation. For example, while credit score is correlated with the safety of an

automobile, the crash is not caused because an individual has a poor credit score.

Thus correlation and causation have two different meanings and should not be used

interchangeably. On the other hand, this does not mean that fatality rates cannot

be affected by the implementation of specific goals to improve specific

characteristics. For instance, risk-taking behavior reduces with an increase in

education levels, therefore it is arguable that if more of the population becomes

more educated and is counseled on the consequences of risk-taking behavior, the

risk-taking behavior may decrease, thus decreasing the probability of a fatal crash

occurring. More reasoning behind this will be presented in the data analysis and

conclusion portions of this thesis.

3.3.2 Regression Model

A linear regression model is used to describe the relationship between a

dependent variable and one or more independent variables. Linear regression

models use the same concept introduced in algebra as an equation of a line, as

shown in Equation 7. Linear regression was used with the assumption that all

variables were linearly correlated with the fatality rates at a state level. Nonlinear

correlations were initially calculated but resulted in extremely similar correlation

coefficients there for a linear regression was used.

Equation 7

where Y = the dependent variable, in this case fatality rate

a = a constant each = independent variable each =

coefficient for independent variable .

The regression models calculated for this thesis were estimated using a 95

percent confidence interval to determine statistical significance. The Pearson

Correlation Coefficients mentioned in the previous section 3.3.1, were used to

determine which variables should be considered for input into the linear regression

model. Once the original regression model was estimated the magnitude of the t

statistic score was compared to a critical t statistic score of 1.28 which is associated

with a one-tailed confidence interval of 90 percent. This 90 percent confidence

interval indicates that the probability of each parameter in the regression model is

statistically significant 90 percent of the time. The coefficient of each of the

variables in the linear regression model represents the change in fatality rates per

each increment in the given variable, given all else remains constant.

When using a linear regression model to describe a dependent variable, a

null hypothesis is inherently being tested. The null hypothesis is a statement that all

else remaining the same, that specific independent variable does not influence the

value of the dependent variable. If the magnitude of the t statistic score is greater

than or equal to the critical t statistic score associated with the given confidence

interval, then the null hypothesis can be rejected. A null hypothesis is never

accepted, but rather can fail to be rejected. If there is a failure to reject the null

hypothesis at a given confidence interval, that variable is inherently insignificant in

influencing the dependent variable. The specific calculated linear regression model

used to estimate the fatality rates by geography will be given in the following

chapter.

CHAPTER 4:

DATA ANALYSIS

The variables that were collected and analyzed were mentioned in the

previous chapter on data collection and methodology. This chapter will focus on the

specific data analyses that were performed, and the significance or insignificance

that each variable was found to have. It is important to note that while some

variables have significant Pearson Correlation Coefficients in the initial stages of

comparison, when all significant variables were considered for the regression model,

many were found to be statistically insignificant at an overall level. Therefore, while

twenty-five characteristics were calculated and compared to fatality rates, only six

variables were used to construct the linear regression model, three of which can be

considered to be classified as risk-taking behaviors, and three which are descriptive

or demographic characteristics not necessarily associated with risk-taking

behaviors. These non-risk-taking behaviors were still important to include in the

regression model to include as many variables as possible, which increases the

accuracy of the estimate model and reduces unobserved component captured in the

error term.

4.1 Fatality Variation across Geography

Although the NHTSA reports fatality rates, not only per VMT, but also per

population, they do not report fatality rates for the southeastern region specifically.

The rates are also yearly rates, and for the purposes of accounting for the

regression to mean, the fatality rates for three years were calculated. A simple

average of fatality rates would not be precisely representative of the three year

average fatality rate, thus the number of fatalities was recorded from the NHTSA

FARS files, and the VMT was recorded from the FHWA Highway Statistics Series for

each state over each year of analysis from 2010 to 2012. For the calculation of the

southeastern region fatality rate the sums of fatalities over three years in each of

the eight southeastern states were summed. The same procedure was done to the

VMT rates for each of the eight states, and the summation of the fatalities was

divided by the summation of the VMTs. The eight states in the southeastern region

include Alabama, Florida, Georgia, Kentucky, Mississippi, North Carolina, South

Carolina, and Tennessee. These averaged fatality rates were used as the dependent

variable in all analyses regarding this thesis. The fatality rates by state, national

average and southeastern region calculated from the FARS database are all listed in

Table 7. It should be noted that every state in the southeastern region of the U.S.

has a higher fatality rate than the national average fatality rate, and southeastern

states represent three of the top six states in the highest fatality rates. These are

unfortunate statistics, and leave a massive amount of room for improvement on the

safety of the roadway network in the southeastern U.S.

Table 7 Fatality Rates by State

State

Fatality

Rate

State

Fatality

Rate

State

Fatality

Rate

District of Columbia

0.62

Oregon

0.98

Texas

1.34

Massachusetts

0.65

Indiana

0.99

Arizona

1.34

Minnesota

0.69

Wisconsin

1.00

Kansas

1.35

Washington

0.79

Nebraska

1.01

Alabama

1.35

New Jersey

0.80

Maine

1.07

New Mexico

1.39

Rhode Island

0.82

Nevada

1.08

Tennessee

1.40

Connecticut

0.83

Total U.S.

1.11

South Dakota

1.42

New Hampshire

0.84

Georgia

1.12

Oklahoma

1.45

California

0.87

Hawaii

1.13

Wyoming

1.48

Maryland

0.88

Delaware

1.16

Louisiana

1.53

Illinois

0.89

Idaho

1.17

North Dakota

1.54

Utah

0.90

Missouri

1.17

Kentucky

1.55

New York

0.91

Iowa

1.18

Mississippi

1.58

Virginia

0.93

North Carolina

1.23

Arkansas

1.67

Vermont

0.94

Florida

1.26

South Carolina

1.70

Ohio

0.96

SE region

1.31

West Virginia

1.73

Michigan

0.97

Pennsylvania

1.31

Montana

1.74

Colorado

0.98

Alaska

1.32

Note: Yellow indicates the states in the southeastern region, orange indicates the

southeastern average, and green indicates the national average.

Data sources: Fatality Analysis Reporting System, 2010-2012 for data on fatalities;

FHWA’s annual Highway Statistics for VMT.

4.2 Risk-Taking Variations across Geography

Unfortunately for the southeastern region of the U.S. fatality rates are not

the only characteristics that are chart toppers. The following three risk-taking

characteristics are all characteristics which every state in the southeastern region

ranks worse than the national average, and each of the characteristics are

presented in their respective tables in ascending order; therefore for a characteristic

such as credit score, each of the southeastern states have values lower than the

national average, whereas for poor overall health, each of the southeastern states

have shares higher than the national average. The states in the southeastern U.S.

also have some of the worst credit scores in the nation, taking all three top spots

for lowest credit scores in the nation starting from the lowest, Mississippi with an

average FICO credit score of 612, South Carolina is next to lowest with an average

score of 621, and Alabama comes in third with an average of 623. In addition, all

eight southeastern states rank within the lowest nineteen states for credit scores,

the highest of which is Florida at 637, which is still below the national average of

640, as shown in Table 8. This is an indication that each of the southeastern states

has average FICO credit scores below the national average credit score (FICO,

2013).

Another factor where the southeastern states rank worse than the nation is

overall poor health as indicated by the BRFSS. All eight states in the southeastern

region have higher rates of poor health than the national average of 4.4 percent as

shown in Table 9. The southeastern average share of adult population with poor

health is 6.1 percent. Bachelor’s degree or higher educational attainment is yet

another characteristic where every state in the southeastern region ranks worse

than the national average share. For the adult population 25 years of age or older,

the national share of population that has obtained at least a Bachelor’s degree is

28.8 percent. For that same age group in the southeastern states, the average

share is 25.5 percent, and the state with the highest share in the southeastern

region is Georgia, with 28.0 percent of the population with a Bachelor’s degree or

higher as shown in Table 10.

Table 8 Credit Scores by State

State

Credit

Score

State

Credit

Score

State

Credit

Score

Mississippi

612

Michigan

636

South Dakota

646

South Carolina

621

Florida

637

Vermont

646

Alabama

623

Nevada

637

Oregon

647

Arkansas

624

Maryland

638

Rhode Island

647

Louisiana

624

Idaho

639

District of Columbia

648

Kentucky

627

Wisconsin

639

North Dakota

648

Texas

627

Wyoming

639

Alaska

650

West Virginia

628

Total U.S.

640

New Hampshire

650

Oklahoma

629

Arizona

640

California

651

Tennessee

629

Illinois

640

Connecticut

651

Georgia

630

Kansas

640

Washington

651

SE region

630

Pennsylvania

640

New Jersey

653

Indiana

631

Iowa

641

New York

654

Missouri

632

Virginia

643

Minnesota

655

New Mexico

632

Montana

645

Maine

658

North Carolina

632

Utah

645

Massachusetts

658

Ohio

634

Colorado

646

Hawaii

659

Delaware

635

Nebraska

646

Note: Yellow indicates the states in the southeastern region, orange indicates the

southeastern average, and green indicates the national average. Data Source:

CreditKarma.com, 2012

Table 9 Overall Poor Health by State

State

Poor

Health

State

Poor

Health

State

Poor

Health

Minnesota

2.9

New Jersey

4.0

Texas

5.0

District of Columbia

3.0

North Dakota

4.0

Oregon

5.2

South Dakota

3.0

Virginia

4.0

Florida

5.3

Colorado

3.1

Wisconsin

4.1

Missouri

5.4

Utah

3.1

Washington

4.2

New Mexico

5.4

Connecticut

3.2

Wyoming

4.2

North Carolina

5.4

Maryland

3.2

Ohio

4.3

Indiana

5.5

Hawaii

3.3

Arizona

4.4

South Carolina

5.9

Idaho

3.4

Delaware

4.4

SE region

6.1

Iowa

3.4

Nevada

4.4

Oklahoma

6.3

Massachusetts

3.4

Total U.S.

4.4

Louisiana

6.5

Table 9 Continued

New Hampshire

3.4

Alaska

4.5

Alabama

7.3

Vermont

3.4

California

4.6

Kentucky

8.0

Nebraska

3.5

Georgia

4.6

Arkansas

8.2

Kansas

3.8

Montana

4.7

Mississippi

8.3

Rhode Island

3.8

Pennsylvania

4.7

Tennessee

8.5

Illinois

4.0

Michigan

4.8

West Virginia

8.8

Maine

4.0

New York

4.8

Note: Yellow indicates the states in the southeastern region, orange indicates the

southeastern average, and green indicates the national average.

Data Source: BRFSS, 2011-2012

Table 10 Bachelor’s Degree or Higher by State

State

Higher

Degree

State

Higher

Degree

State

Higher

Degree

West Virginia

18.3

Missouri

26.2

Kansas

30.3

Arkansas

20.1

South Dakota

26.2

Utah

30.3

Mississippi

20.1

Florida

26.4

California

30.7

Kentucky

21.5

Texas

26.7

Rhode Island

31.3

Louisiana

21.8

Wisconsin

26.8

Illinois

31.4

Nevada

22.4

Arizona

26.9

Washington

31.9

Alabama

22.6

North Dakota

27.2

Minnesota

32.6

Indiana

23.2

North Carolina

27.3

New York

33.2

Oklahoma

23.5

Alaska

27.5

New Hampshire

33.7

Tennessee

23.8

Pennsylvania

27.5

Vermont

34.8

Wyoming

24.7

Maine

27.9

Virginia

35.2

Idaho

25.1

Georgia

28.0

New Jersey

35.8

South Carolina

25.1

Nebraska

28.5

Connecticut

36.5

Ohio

25.2

Montana

28.7

Maryland

36.8

SE region

25.5

Total U.S.

28.8

Colorado

37.0

Iowa

25.7

Delaware

28.9

Massachusetts

39.4

New Mexico

25.8

Oregon

29.7

District of Columbia

52.4

Michigan

25.9

Hawaii

30.1

Note: Yellow indicates the states in the southeastern region, orange indicates the

southeastern average, and green indicates the national average.

Data Source: U.S. Census Bureau, 2010-2012

When it comes to the number of states in the southeast that have a larger

share of adult smokers than the national average, the region is not quite as bad as

the fatality rate rankings and the credit score rankings. However, only one

southeastern state, Florida (16.8 percent), has a smaller share of adult smokers

than the national average (18.2 percent). In the southeastern states the average

share of adult population that smokes is 20.0 percent as shown in Table 11 (BRFSS,

2012). Trends of alcohol use in the past 30 days are a bit more promising. The

average share of the adult population in the southeastern states that reported

drinking alcohol in the past month was 46.6 percent, which is lower than the

national average of 54.5 percent. As a matter of fact, only one southeastern state,

Florida (54.7 percent), had a higher share of alcohol drinkers than the national

average as shown in Table 12 (BRFSS, 2012).

Table 11 Adult Tobacco Users by State

State

Tobacco

Users

State

Tobacco

Users

State

Tobacco

Users

Utah

10.3

Illinois

18.0

Wyoming

20.6

California

12.5

Minnesota

18.0

Pennsylvania

21.0

Hawaii

13.3

Total U.S.

18.2

North Dakota

21.2

Connecticut

15.5

Nebraska

18.5

Michigan

21.4

New Jersey

15.7

Wisconsin

18.7

Alabama

21.5

Texas

15.9

District of Columbia

18.8

Indiana

21.9

Washington

16.1

Georgia

18.8

South Carolina

22.0

New Hampshire

16.2

Montana

19.0

Missouri

22.1

Arizona

16.3

Virginia

19.0

Alaska

22.6

Maryland

16.4

New Mexico

19.1

Ohio

23.4

Massachusetts

16.6

Nevada

19.4

Louisiana

23.5

New York

16.6

Iowa

19.5

Oklahoma

23.7

Vermont

16.6

Delaware

19.6

Tennessee

24.3

Florida

16.8

South Dakota

19.6

Mississippi

24.8

Idaho

17.2

Kansas

20.0

Arkansas

25.9

Oregon

17.3

Maine

20.2

Kentucky

26.5

Rhode Island

17.4

SE region

20.2

West Virginia

27.3

Colorado

17.7

North Carolina

20.3

Note: Yellow indicates the states in the southeastern region, orange indicates the

southeastern average, and green indicates the national average.

Data Source: BRFSS, 2012

Table 12 Adult Alcohol Drinkers by State

State

Alcohol

Users

State

Alcohol

Users

State

Alcohol

Users

Utah

30.7

Idaho

49.6

Nebraska

57.5

West Virginia

34.0

Arizona

51.5

South Dakota

57.8

Tennessee

37.5

Kansas

51.5

Maine

58.6

Kentucky

38.2

Virginia

52.0

New Jersey

58.6

Mississippi

38.4

Wyoming

53.0

Washington

58.7

Arkansas

39.8

Ohio

53.3

Montana

58.8

Alabama

40.3

Nevada

54.1

North Dakota

61.0

Oklahoma

42.0

Pennsylvania

54.4

Oregon

61.0

North Carolina

44.3

Maryland

54.5

Colorado

61.9

SE region

46.6

Total U.S.

54.5

Connecticut

62.6

Georgia

47.1

Florida

54.7

Rhode Island

62.6

New Mexico

48.3

Alaska

55.4

Vermont

62.9

Indiana

48.5

New York

55.4

Massachusetts

63.6

Missouri

48.5

California

55.5

Minnesota

63.6

South Carolina

48.5

Michigan

56.6

New Hampshire

63.7

Louisiana

49.0

Delaware

57.1

District of Columbia

64.3

Hawaii

49.3

Illinois

57.2

Wisconsin

64.6

Texas

49.3

Iowa

57.2

Note: Yellow indicates the states in the southeastern region, orange indicates the

southeastern average, and green indicates the national average. Data Source:

BRFSS, 2012

Serious mental illness is another characteristic that was explored as being

associated with increased risk-taking behaviors, thus possibly associated with traffic

fatality rates. For adults 26 years of age and older, the average share of the

population that had a serious mental illness was higher in the southeastern region

than the national average share, however three of the southeastern states had

lower shares of mental illness in that age group, Alabama, Florida, and Georgia, as

shown in Table 13.

Table 13 Adults (26+) with Serious Mental Illness by State

State

Mental

Illness

State

Mental

Illness

State

Mental

Illness

Maryland

3.16

Delaware

4.10

Montana

4.58

New Jersey

3.16

Total U.S.

4.14

Ohio

4.62

Illinois

3.24

Minnesota

4.16

Missouri

4.66

Connecticut

3.28

Texas

4.16

Tennessee

4.66

Hawaii

3.32

Wisconsin

4.16

Idaho

4.69

South Dakota

3.51

Nebraska

4.18

Oregon

4.81

New York

3.71

Wyoming

4.19

North Carolina

4.84

Alabama

3.79

Massachusetts

4.21

Mississippi

4.86

New Hampshire

3.79

Arizona

4.26

Washington

4.90

Colorado

3.83

Alaska

4.27

Maine

4.96

California

3.87

New Mexico

4.30

Rhode Island

4.99

Virginia

3.90

SE region

4.32

Kentucky

5.03

District of Columbia

3.95

Louisiana

4.33

Oklahoma

5.14

Florida

3.98

Iowa

4.37

West Virginia

5.27

Pennsylvania

4.01

Kansas

4.40

Utah

5.36

Georgia

4.02

South Carolina

4.41

Vermont

5.36

Nevada

4.05

Indiana

4.44

Arkansas

5.41

North Dakota

4.05

Michigan

4.58

Note: Yellow indicates the states in the southeastern region, orange indicates the

southeastern average, and green indicates the national average. Data Source:

NSDUH, 2012

4.3 Correlations with Fatality Rates

Pearson Correlation Coefficients were calculated for each explored

characteristic by state to get an idea of how much the characteristic related not

only to fatality rates, but also to other characteristics. This is an important step in

the analysis because a regression model should not include two highly correlated

variables as independent variables. Including two highly correlated variables in a

regression model will skew the results of the model and make the output of the

model less reliable. While all correlations that were above 0.3 are considered

statistically significant at a 95 percent confidence interval, not all statistically

significant characteristics remained significant when multiple characteristics were

entered into the regression model. Therefore, some of the variables that are shown

below to have individual statistical significance to the fatality rates by state are not

included in the final regression model used to estimate the fatality rates. All

correlations between fatality rates and characteristics are presented in Table A1

located in the appendix.

4.3.1 Credit Score

The average credit score for each state and the nation as a whole was

collected and the weighted average for the southeastern region was calculated

using a population based weighted average over the eight southeastern states

(CreditKarma, 2015). The Pearson Correlation Coefficient was calculated between

the average credit scores and the average fatality rates. As noted previously, the

closer the magnitude of the coefficient is to one, the stronger the correlation

between the two arrays of values. The Pearson Correlation Coefficient between

credit scores and fatality rates was calculated to be -0.643 which is statistically

significant at the 95 percent confidence interval. This is indicative of a strong

negative correlation. This means that with 95 percent confidence, average credit

scores are related to fatality rates at the state level. Credit score is one

characteristic that may help to explain the increased levels of fatality rates in the

southeastern region when compared to the rest of the U.S. because the credit

scores are significantly lower in the southeastern states also.

4.3.2 Smoking

The average share of the adult population that has used any tobacco product

in the past thirty days (of the survey) was calculated using the BRFSS data and the

NSDUH data, although the results were nearly identical between the two data

sources, the BRFSS data was used because the methodology of the survey

contained a larger sample size. The Pearson Correlation Coefficient was calculated

comparing tobacco use among adults to fatality rates by state, and was found to be

0.641, indicating that there is a moderate statistically significant correlation at a 95

percent confidence interval. This correlation can be interpreted as meaning states

with a higher share of adult smoking population also have higher fatality rates. This

is a correlation and not causation, meaning that while smoking does not cause

motor vehicle crashes, smoking is considered a risk-taking behavior. As

hypothesized, as the share of the adult population that uses tobacco increases, the

share of adults who are likely to behave in a risky manner also increases, thus the

number of fatal crashes increases.

4.3.3 Alcohol Consumption

The average share of the adult population that has consumed an alcoholic

beverage in the past month was calculated using the BRFSS data collected over a

two year period. This is self-reported alcohol use, so the share may be under

representative of the actual share of the adult population that has had a drink,

because some individuals that are classified as adults but are not of legal drinking

age may be less apt to tell the truth in a voluntary survey. In addition, several

counties in the southeastern region of the nation are dry counties which prohibit the

sale, distribution, and consumption of alcohol, which may also lead to inaccuracies

when referring to self-reporting in a survey. Any self-reported survey data should

be used with caution with an understanding that error biases may potentially be

higher than other forms of data collection. A southeastern region weighted average

was calculated to compare with the given national average, and it was found that

the southeastern states have a smaller share (46.6 percent) of adult population

that consumed alcohol in the past month (of the survey) than the national average

(54.5 percent) (BRFSS, 2012). The Pearson Correlation Coefficient that was

calculated comparing the alcohol consumption by state to fatality rates by state was

-0.577, indicating that there is a moderate negative statistically significant

correlation between the two variables at a 95 percent confidence interval. It is of

importance to note that simply consuming an alcoholic beverage in the past month

is not in itself a risk-taking behavior, but rather the act of driving while intoxicated

would be considered a risk-taking behavior. Therefore the correlation coefficients

found in this section of the thesis may not be of much significance when referring to

fatality rate variations by risk-taking behavior characteristics. The same limitations

are present in the other types of drinking used in this thesis, as all the data relating

to alcohol consumption was retrieved from the same source. Binge drinking and

heavy drinking are again from self-reported survey collected data, and is found to

be less of an actual indicator of risk-taking behavior than originally hypothesized.

Binge drinking is defined by the BRFSS as more than five drinks at one time for a

male and more than four drinks at one time for a female. Another variable that was

compared to fatality rates is average share of the adult population that admitted to

binge drinking in the last 30 days. Binge drinking share also had a statistically

significant Pearson Correlation Coefficient at a value of -0.366, with a confidence

interval of 95 percent. While this is a statistically significant correlation, the

correlation is defined as moderate, not strong, and is less correlated than any

alcohol consumption is to fatality rates. The share of binge drinkers in the adult

population is less in the southeastern region than the national level (BRFSS, 2012).

This negative correlation is not what was expected, as the original hypothesis was

that as alcohol use, a potentially risky behavior, increases, fatality rates would also

increase. Yet as previously mentioned, the act of consuming an alcoholic beverage

is not necessarily a risky behavior in itself, but the risk associated with drinking

alcohol is introduced if driving occurs subsequent to the alcohol consumption. Given

the above mentioned limitations of the data in addition to the aggregate level that

the data was collected, the negative correlations are of little significance to this

thesis. Of possible greater significance, is the sales volume of alcohol, and the

correlations between alcohol sales and fatality rates. Gruenewald and Ponicki

(1995) report finding a positive correlation between alcohol sales and nighttime

fatal crashes using data from 38 states over a 12 year time period

(Gruenewald,P.J., Ponicki, W.R., 1995). McMillan and Lapham (2006) found a 42

percent increase in alcohol related crash fatalities on Sunday correlated to Sunday

packaged alcohol sales ban being lifted in New Mexico (McMillan,G.P., Lapham,S.,

2006).

4.3.4 Drug Use

Using the NSDUH given through the SAMHSA, drug use is categorized into

three categories in addition to five age groups, some of which are overlapping. The

categorization of drug use is defined as illicit drug use in the past month, marijuana

use in the past month, and illicit drugs other than marijuana in the past month. At

the national level it is theoretically sound to add the marijuana use and illicit drug

use other than marijuana to get total drug use, however because of the

methodology mentioned in section three of this thesis; it is not sound to do that

same iteration at the state level. The age groups that are included in the survey are

shown in Table 14.

Table 14 Age Groups of NSDUH

Age Group (years)

Represents

12 +

All respondents

12 - 17

Minor respondents

18 - 25

Young adult respondents

18 +

Adult respondents

26 +

Older adult respondents

Data source: NSDUH, 2012

For the all illicit drug use category, the Pearson Correlation Coefficient was

calculates to be inversely correlated (-0.53) in the 18 to 25 year old age range

when compared with total fatality rates. However, when the correlation to fatality

rates of only the 18 to 25 year old age group was calculated, all illicit drug use was

found to have a strong statistically significant correlation of 0.95 (NSDUH, 2012).

This is an indication that it is not sound to correlate independent characteristics of a

specific age group to a dependent characteristics that is inclusive of all age groups,

meaning the share of young adult illicit drug users is not directly correlated with the

overall fatality rate, but is strongly correlated to the fatality rate of young adults.

This presents a clearer representation of what one would expect, given drug use is

an inherently risky behavior that also causes an individual to behave in a risky

manner. The same trends are present in marijuana only use, and in illicit drugs

other than marijuana, however are not found to be of statistical significance at any

reasonable confidence interval (NSDUH, 2012). It should be noted that as

mentioned in the previous methodology, the sample size for the NSDUH is not

nearly the size of other similar population surveys, thus the results should be used

with caution.

4.3.5 Physical Health

The BRFSS reports shares of adults in body mass index (BMI) ranges from

underweight to obese, in addition to overall general health from poor to excellent.

For the purposes of this thesis, three physical health characteristics were chosen to

compare with fatality rates by state; obese BMI (30.0 or higher), excellent overall

health, and poor overall health. The national share of the adult population that is

classified as having an obese BMI is 29.4 percent, which is less than one percent

lower than the southeastern region average share of adult population with an obese

BMI (30 percent). Florida is the only state in the southeastern region that has a

lower share of obese adult population than the national average. The Pearson

Correlation Coefficient was calculated comparing the share of adult population that

is obese with the fatality rates by state was found to be 0.616, which is statistically

significant at a 95 percent confidence interval. It is worth noting again that

correlation coefficients do not indicate causation, but correlation, and it is

imperative that the two are not confused. When the correlation between general

health and obesity is calculated it was found that obesity has a statistically

significant strong direct correlation with poor health (0.714) and a statistically

significant strong inverse correlation with excellent health (-0.821). This is an

indication that as the share of adults that are obese increase, the share of adults

that are overall generally unhealthy increases, and the share of generally excellent

healthy population decreases. This is also an indication that both overall health and

obesity should not be included in a regression model to estimate fatality rates, as

they are more correlated with each other than they are correlated with fatality

rates.

In the case of overall health, Pearson Correlation Coefficients were calculated

between excellent overall health and fatality rates, between poor overall health and

fatality rates, and between excellent overall health and poor overall health. All

correlations were found to have moderate statistical significant correlations at a 95

percent confidence interval. Excellent overall health and fatality rates have a

moderate inverse correlation of -0.615. The healthier the population, the more

likely they are to survive a collision that an otherwise unhealthy person may not

have survived. Just as intuitive, the correlation between overall poor health and

fatality rates is a direct correlation of 0.666. If an unhealthy person is involved in a

traffic collision, their chances of survival is less than that of a healthy person. For

completeness, the overall poor health and overall excellent health were also

compared, and were found to have a strong inverse statistically significant

correlation of -0.771, indicating that they should not both be used in a regression

model to estimate the fatality rates by state.

4.3.6 Mental Health

Mental health was another area of exploration for the study of this thesis, in

the idea that decreased levels of cognitive mental capabilities may be indicative of

increased probability of behaving in risky manners. While the theoretical

relationship between risk-taking behavior increases with increased mental illness

made sense, especially throughout the literature review, where studies revealed a

higher propensity for risky sexual behavior among those with decreased mental

cognitive capabilities, the data in this thesis found no statistical significance in a

relationship between mental disabilities and fatality rates at a state level, with the

slight exception of one age group in one mental illness variable. The mental illness

variables that were explored were again from the NSDUH data; serious mental

illness in the past year, any mental illness in the past year, at least one major

depressive episode in the past year, and thoughts of suicide in the past year. In

general the southeastern region was fairly close to the national level in all mental

health characteristics. The only characteristic that was found to be slightly

statistically significant with fatality rates at the state level was the 26 and older age

group of serious mental illness in the past year, with a Pearson Correlation

Coefficient of 0.326. In order to be statistically significant at the 95 percent

confidence interval the Pearson Correlation Coefficient must have a magnitude

greater than or equal to 0.3.

4.3.7 Educational Attainment

Two categories of educational attainment were explored for comparison

between educational levels and fatality rates. Both categories were defined by the

U.S. Census Bureau as population 25 and older who has a high school diploma or

equivalent, and population 25 and older with a Bachelor’s degree or higher. The

Pearson Correlation Coefficient calculated was the highest between the shares of

population over 25 with a Bachelor’s degree or higher and fatality rates, at a value

of -0.742, indicating a strong inverse statistically significant correlation at a 95

percent confidence interval. This is an expected correlation, because in general, as

a person’s education level increases, their self-worth also increases, and their

propensity to partake in risky behavior that may jeopardize their hard work in

obtaining their education is greatly reduced. There is also a statistically significant

inverse correlation of -0.345 between the shares of population that obtained their

high school diploma or equivalent, however that correlation is not as strong as the

higher education correlation. When compared to share of the population below the

poverty level , the share of population who obtained their high school diploma is

greatly inversely correlated (-0.774). This means that as the share of population

that has completed high school increases, the share of population below poverty

level decreases, which is an expected relationship. The last educational correlation

coefficient that was calculated was the correlation between the share of population

who completed their high school education, and the share of population that

obtained a Bachelor’s degree or higher. This direct correlation was statistically

significant at a value of 0.407. With all the correlation coefficients taken into

consideration, it was concluded that the characteristic that should be included in the

regression model to estimate fatality rates is the share of population with a

Bachelor’s degree or higher at the age of 25.

4.4 Other Demographics Used for Regression

In order to have as reduce the unobserved or random component accounted

for in the error term of a regression model, a couple other characteristics aside

from the risk-taking characteristics being tested are included in the model.

Characteristics that are known to affect the fatality rates, from previously reviewed

literature, should be included so that the risk-taking characteristics are not being

misrepresented in the fatality rate estimation. Transportation safety is a complex

problem with numerous interconnected variables all having some type of

contribution overall. Unfortunately, it is unrealistic to include every influencing

factor in a thesis to estimate approximate fatality rates given the variation of

independent variable characteristics by geography. This thesis used characteristics

known to affect transportation safety in conjunction with several risk-taking

characteristics to estimate the amount of the risk-taking characteristics have on the

overall fatality rate at a state level. The demographic characteristics will also help to

estimate more accurate marginal effect coefficients to the risk-taking characteristics

in the linear regression model. For this model, the demographic characteristics

included are share of the population born in state of residence, and unemployed

share of population by state.

The share of population born in state of residence seems to be a moderately

decent indicator of the rate of unfamiliar drivers on the roadway network. This is

not implying that because you move to another state you are always an unfamiliar

driver, however, visitors from the state of origin will be unfamiliar drivers, and the

larger the share of residents born out of their state of residence, the larger the

share of friends and family that will visit the current state of residence. Inversely if

an individual lives in the state they were born in, the chances of them being

unfamiliar of the roadway network decreases. Familiarity of the roadway network

includes, but is not limited to, knowing where crosswalks and rail crossings are

located, knowing posted speed limits, school zones, senior zones, etc. and which

exit to take off the interstate. The share of unemployed population is considered as

one of the descriptive independent variables because the unemployed population

generally drives much less than the share of the population that is employed. The

reduction in travel contributes to the overall reduction in fatality rates because if an

individual is not traveling, their chances of being fatally injured while traveling are

significantly decreased via simple exposure rates. A report produced through the

NHTSA in 2010 revealed that the Metropolitan Statistical Areas with the highest

increase in unemployment between 2007 and 2008 also experienced the largest

decrease in fatalities, and the Metropolitan Statistical Areas with the smallest

increase in unemployment rates experienced the smallest decline in fatalities during

the same time period, citing great reductions in crashes involving young drivers,

multiple vehicle crashes and crashes which occurred on the weekends when most

discretionary driving occurs (NHTSA, 2010).

Other studies such as “The Effects of Mandatory Seat Belt Laws on Driving

Behavior and Traffic Fatalities,” by Cohen and Einav used independent descriptive

characteristics such as median income, mean age, share of black and Hispanic

population, rural density, violent crimes, speed limit, and fuel tax to explore the

differences in fatality rates by geography. “Gasoline Prices and Motor Vehicle

Fatalities,” by Grabowski and Morrisey used gasoline prices, license revocation, seat

belt laws, speed limit, unemployment rate, and income per capita to estimate

fatality rates by state. The results of the estimation are often dependent on the

geography the data is aggregated to. According to Clark and Avery (2010) due to

aggregation bias “it is incorrect to assume that relationships existing at one level of

analysis will necessarily demonstrate the same strength at another level’ (Clark and

Avery, 2010). For instance a variable that is known to have significant correlation at

a Census Tract aggregate level may not display the same significance or even the

same direction of relationship at a state level. Some other characteristics that are

commonly used as independent variables to describe a dependent variable of

fatality rates, but were not used in the regression analysis of this thesis include:

race, income, traffic density, property crimes, fuel taxes, fuel consumption, average

speed, vehicle expenditures, speed limit laws, gross national product, retail sales,

etc. Several of these mentioned independent variables are inter-correlated,

meaning it would not be meaningful nor correct to include all listed variables in one

regression analysis. As mentioned previously, safety is a complex problem with

multifaceted interrelated solutions which are not obvious nor easily solved, which is

why fatalities still occur on the roadway network.

4.5 Linear Regression

It is of extreme importance to note that the linear regression formed for this

thesis was formed at an aggregate state level. Aggregate level data is not ideal, as

it is not as accurate nor is it always as representative as disaggregate data. For

instance the average credit score of a specific city, Phoenix, Arizona is 737;

however the average credit score for the state of Arizona is only 640, indicating

that if an average state credit score is used to represent each city, the average

credit score for the city of Phoenix would be greatly underestimated.

Unfortunately, for the variables being used for consideration in this thesis, such as

credit score, smoking population, educational attainment, etc., consistent

disaggregate data is not available. Aggregate data has the potential to have

aggregation bias, where the correlations may appear to be statistically significant at

an aggregate level but not at a disaggregate level, or the regression analysis may

indicate a significantly different relationship between the dependent and

independent variables than would be apparent if the data were evaluated at a

disaggregate level (Clark and Avery, 2010). With the understanding that the

risktaking characteristics are at an aggregate state level, all calculations and

conclusions should be used with caution.

There are always unknowns present in regression analyses, some are known

unknowns and some are unapparent unknowns. In this thesis there are several

characteristics which were not considered because the scope of this thesis was

focused on risk-taking behaviors, and could not reasonably consider every impactful

possible characteristic. Some unknowns which are known to affect fatality rates

which were not considered in this thesis include types of vehicles driven, travel

speeds, state traffic laws, emergency response capabilities, weather, topography,

intersection density, time of day which the crash occurs, driving under the

influence, rural and urban share of crashes, etc. The error term in the regression

model attempts to account for all the characteristics which were not explored.

An estimated regression model of fatality rate as a function of average credit

score, share of population over 25 with bachelor’s degree or higher, share of

population with overall poor health, share of population who reports always using a

seat belt, share of population born in state of residency, and unemployed share of

population is shown in Equation 8.

Equation 8

where = fatality rate

= average credit score

= share of population over 25 with bachelor’s degree or higher

= share of population with overall poor health

= share of population that reports always using a seat belt

= share of population born in state of residency

= unemployed share of population

= error term to account for unknowns.

When all data was entered and calculated, the coefficients were determined and

shown in Equation 9.

Equation 9

where the terms represent the same variables as indicated in Equation 8.

It is important to understand that the above calculated linear regression

equation is using a limited number of variables, and as mentioned previously,

safety, specifically fatality rates, are extremely complex and impossible to

accurately measure with a limited number independent variables. The correlations,

both known and unknown, make this problem too complex to solve with one linear

regression equation. This thesis is an attempt at using the best available data to

distinguish the amount of effect risk-taking independent variables have on fatality

rates while also accounting for some demographic independent variables which are

known to effect fatality rates but are not too strongly correlated to the risk-taking

variables being used.

The sign of the coefficient of each independent variable determines whether

or not the dependent variable is directly or inversely related. If the sign of the

coefficient of the independent variable is positive, then the independent variable is

directly related to the dependent variable, meaning as the independent variable

increases the dependent variable also increases marginally at a proportion equal to

the magnitude of the coefficient, given all other variables are held constant. For

instance, for each percent share increase of the adult population with poor health,

the fatality rate of that geography should experience an increase of 0.079, given all

other independent variables remain the same. Contrarily, if the sign of the

independent variable is negative then the independent variable and the dependent

variable have an inverse relationship, meaning as the independent variable

increases, the dependent variable should decrease proportionately at the magnitude

of the independent variable coefficient given all other characteristics remain the

same. For instance, for each percent share increase of the population over 25 years

of age that obtains a bachelor’s degree or higher, the fatality rate should decrease

0.018; given all other independent variables remain constant.

It is also important to mention the error term which is present in the

regression model. The error term helps to explain the variability in fatality rates

when all independent characteristics are equal. Error terms are present in all

regression models and are assumed to represent a normal distribution with a mean

value of zero. The error term is considered the random or unobserved component in

the regression analysis. Once the regression analysis is estimated, the error term is

no longer included and the effects of the variables that are not included in the

model, or the unobserved characteristics, are captured in the constant term. Table

15 displays the parameter estimates, standard error of the estimate, and t statistic

for each of the explanatory variables, along with goodness of fit measures.

Table 15 Linear Regression Variable Estimates and Goodness of Fit Measures

Explanatory variable

Parameter

estimate

Standard

error (S.E.)

of the

estimate

statistic

Credit Score

-0.007

0.003

-2.169

≥25 yrs with Bachelor’s or higher

-0.018

0.006

-3.206

Overall Poor Health

0.079

0.024

3.264

Seat Belt Use

-0.005

0.004

-1.370

Residents born in state

-0.005

0.002

-2.702

Unemployed

-0.046

0.015

-3.140

Goodness of fit measures:

Regression Sums of Squares (SSR)

3.381

Residual Sums of Squares (SSE)

1.107

Total Sums of Squares (SST)

4.488

0.753

Adjusted R2

0.721

Number of cases

Credit score was calculated to have an inverse effect on fatality rates at the

aggregate state level. This result was expected given the previously performed

literature review which revealed that as a person’s fiscal responsibility increases,

their credit score also increases. This increased credit score and increased fiscal

responsibility is accompanied by a decreased likelihood of risk-taking behavior. As

an individual’s likelihood of risk-taking behavior in financial matters decreases, their

likelihood of driving in a risky manner should also decrease. In addition to driving in

a safer manner, a fiscally responsible person is also more likely to keep their

vehicle properly maintained, such as getting regularly scheduled oil changes,

replacing tires prior to balding, replacing brake-pads at regular intervals, etc. This

regularly scheduled proper maintenance will reduce the likelihood of vehicle failure,

and thus reduces the chances of a collision due to vehicle failure. While fatality

rates are inclusive of all person types such as drivers, passengers, and pedestrians,

etc. decreasing the propensity of risk-taking behavior would have a positive effect

on all roadway network users.

The share of population with a Bachelor’s degree or higher was calculated to

have an inverse effect on fatality rates at the aggregate state level. This result was

expected given the previously performed literature review which revealed that as an

individual’s education levels increase their understanding of consequences which

accompany their actions also increases. Accompanying a better understanding of

consequences for actions is a reduced likelihood of driving while intoxicated,

distracted, at unsafe speeds, or without a seat belt. In addition, people with higher

education levels are more likely exhibit healthy behaviors such as eating a healthy

diet, exercising, refraining from smoking, all of which are examples of decreased

risk-taking behaviors which also lead to increased life expectancy.

The share of population with overall poor health was calculated to have a

direct relationship with fatality rates at the aggregate state level. This result was

expected given the previous literature review which assumes general health to be

related to chronic diseases, healthy behaviors, and access to preventative

healthcare services. Overall poor health can be caused by several factors which are

also considered to be risk-taking behaviors such as eating an unhealthy diet, lack of

exercise, being overweight, and smoking. Thus as risk taking behavior increases,

the chance of being classified as overall unhealthy also increases. This supports the

original hypothesis that as risk-taking behaviors increase, fatality rates also

increase.

The share of population which always wears a seat belt while driving was

calculated to have an inverse relationship with fatality rates at an aggregate state

level. Seat belt use, or rather lack of seat belt use is considered a risk-taking

behavior which is inherently different than the other risk-taking behaviors being

examined in this thesis. Seat belt use has been proven to reduce the severity of a

motor vehicle crash when used properly, and is the only risk-taking behavior

analyzed which has a direct effect on the injury severity of a vehicular crash. When

considering the other risk-taking characteristics which were examined in the

regression analysis, changing the proportionate share of population would not have

a direct effect on fatality rates. For instance, increasing the average credit score of

an individual does not inherently make them behave in a less risky manner thus

would not directly cause them to drive more carefully. Unlike the other risk-taking

characteristics reviewed however, increasing the share of the population that always

wears their seat belt can directly impact the fatality rates by reducing the injury

severity of collisions which occur.

The last two independent variables presented in the regression model were

not risk-taking characteristics, instead were demographic characteristics used to

help explain the variation in fatality rates. The share of population born in state of

residency was calculated to have an inverse correlation to the fatality rate by state.

This is an intuitive relationship because as previously stated, if a person resides in

the state in which they were born they are less likely to be unfamiliar drivers on the

roadway network. As the share of unfamiliar drivers on the roadway network

decreases, the variation in speed on that network should also decrease, and a

reduction in variation of speed is associated with a lower chance of collision

occurrence. In addition, a person born in their state of residence will likely know the

location of high risk of collision areas such as crosswalks, rail at-grade crossings,

etc. Finally, the share of unemployed population was calculated to an inverse

relationship to fatality rates by state. This is an intuitive correlation because an

unemployed individual will travel less than an employed individual, thus reducing

their exposure risk. Trip reduction of the unemployed population would not only be

due to less home-based work trips, but the lack of income associated with

unemployment would likely also lead to a reduction in all types of trips. This

reduction in trips equates to a decreased exposure rate, thus reducing the chances

of being involved in a fatal collision. This increased unemployment has shown to

lead to decreased crashes involving young drivers, multiple vehicle crashes, and

weekend crashes, all leading to reduced fatalities (NHTSA, 2010).

CHAPTER 5:

CONCLUSIONS

Fatality rates in each of the eight states in the southeastern region of the

U.S. have consistently been higher than the national average fatality rate in terms

of fatalities per VMT, and fatalities per population. As of 2012, three of the eight

southeastern states were ranked in the top ten of overall fatalities per capita in the

nation. Safety is an incredibly complex problem affecting every aspect of the

transportation sector. Each of the influencing characteristics associated with the

safety of a transportation network are interconnected in such a complicated way

which make a complete unanimous solution an impossible feat. With the complete

understanding that no one solution is going to have the ability of eliminating all

transportation fatalities which occur on the network, hypotheses are made to try to

understand why some regions have higher fatality rates than other regions. While

the southeastern region does have socio-economic and socio-demographic traits

such as higher levels of poverty, higher share of elderly population, higher minority

shares of population, and more vulnerable road users, which partially help to

estimate why fatality rates are higher in this region of the nation, this thesis

focused mostly non-conventional risk-taking characteristics as explanatory

variables. Many risk-taking characteristics were found to have significant

correlations to conventional characteristics used in previous literature to estimate

fatality rates at specific levels of geography, such as median income and share of

population with a Bachelor’s degree or higher. When two independent variables

exhibit strong correlations, using both the correlated independent variables would

not be of sound logic, and thus for the purposes of this thesis, median income was

replaced with share of population with a Bachelor’s degree or higher.

The hypothesis tested implies that if a person behaves in a risky manner in

economic and health aspects of their life, risk is likely present is several aspects of

their life, therefore that person is also expected to drive in a risky manner, and

delay maintenance of their vehicle which, would in turn equate to more collisions

and higher fatality rates. This is under the assumption that if a person displays

risktaking characteristics of any type, that person likely behaves in a risky manner

in other aspects of their life. This assumption of risk-taking behavior is used as a

great starting point to help explain variation in fatality rates in specific regions of

the nation where risk-taking behaviors and fatality rates are both higher than

national averages.

A breadth of risk-taking characteristics such as credit score, educational

attainment, seat belt use, tobacco, alcohol and drug use, obesity, and overall

health, were shown to have statistically significant bivariate correlations with

fatality rates, as shown in Table A1 in the appendix. Four risk-taking characteristics

were found to influence fatality rates at an integrated level when a linear regression

model was formed; average credit score, educational attainment, overall health,

and seat belt use. While one of the mentioned risk-taking characteristics -

propensity to use a seat belt - is directly related to fatality rates, the other

risktaking behaviors are indirectly related to fatality rates, such as credit score,

educational attainment, and overall health.

Seat belt use has been shown to reduce the injury severity of a vehicular

passenger when involved in a collision; therefore increasing the share of seat belt

using population in a region such as the southeastern region of the U.S. would

directly impact the fatality rates of the region. In the context of other risk-taking

characteristics considered in this thesis, the relationship with fatality rates is not

necessarily causal. Correlation is not causation; two variables that are statistically

significantly correlated does not imply that changing one variable would cause a

change in the other variable. For example, increasing the share of population that

has a higher education may not directly lead to a reduction of fatality rates unless

that increase in educational attainment is coupled with a reduction in overall

sensation-seeking or risk-taking behavior. The correlation between fatality rates and

educational attainment is due to the typical characteristics of an individual with a

higher educational attainment, not the education level itself. As stated in the

literature review of this thesis, a person with a higher education level is more likely

to consider the consequences of their actions, have a higher self-worth, and have a

higher household income, all of which are characteristics associated with decreased

levels of risk-taking behavior. Increasing education levels in itself will not

necessarily increase these traits associated with decreased risk-taking, however it is

likely that as education levels increase awareness would also increase, decreasing

overall risk0-taking behavior. Still, all interpretations of the calculated regression

model from chapter four must be considered with caution.

Credit score is similar to educational attainment, in that there is not a direct

relationship between credit score and fatality rates, however the literature review

revealed that credit score increases are associated with likelihood of vehicle

maintenance and cautious driving habits, which have the ability to influence fatality

rates. As revealed in the literature review, a consumer who is prudent in their own

financial matters is also expected to be more prudent in vehicle maintenance and

display more cautious driving habits. It is not the increased credit score in itself that

reduces the fatality rates; it is the underlying responsible characteristic of an

individual that leads to a high credit score and reduced propensity to engage in risk-

taking behavior. This is an indication that simply increasing the average credit score

in a particular geography would not reduce the fatality rates in that geography,

unless the increased credit score is coupled with changes in the typical traits that

reduce risk-taking behavior.

Poor overall health is perhaps the most indirect characteristic which is related

to risk-taking behaviors and correlated to fatality rates, at least at a state level

geography. The literature review revealed that overall poor health is related to

chronic diseases, unhealthy behaviors and lack of preventative health-care services

such as regular check-ups, with the most common factors to overall poor health

being specific risk-taking behaviors such as smoking, consuming an unhealthy diet,

lack of exercise, and being overweight. Again, the distinction between correlation

and causation must be fully understood. This correlation between poor overall

health and fatality rates does not indicate that simply increasing the overall health

levels of a population would directly decrease vehicular fatality rates. The increased

health of the overall population would necessarily be coupled with an overall

decrease of risk-taking behaviors in order to have any impact on the fatality rates in

the region. For instance, just because someone eats a healthy diet and exercises,

thus increasing their overall health, does not mean that individual will not drive in a

risky manner, nor does this indicate that person is more likely to survive a collision

than a person who eats an unhealthy diet and does not exercise, however a healthy

individual may be more attentive or less fatigued while driving, thus reducing their

risk of collision while driving. The correlation should be interpreted as the likelihood

of risk-taking behaviors increases, the chances of classification of overall poor

health also increases, in addition, the probability of collision increases due to the

increased risk-taking behavior.

It is important to reiterate that the linear regression model was formed using

state level aggregate data, and therefore the chance of aggregate bias is

introduced. Unfortunately, there is not much risk-taking characteristic disaggregate

data available for analyses, leading to the requirement of using data aggregated to

the state level. When data is aggregated at the state level in the U.S. the data

points are then limited to 52 cases when the District of Columbia and the national

average are also included as data points. To improve the usefulness of the linear

regression model, more data points and more variables could be introduced;

however this adds a level of difficulty beyond the scope of study for this thesis.

To reduce the driving related risk-taking behavior, measures which have

previously been shown to be effective include making laws more comprehensive by

changing non-existent or secondary laws to primary laws which are inclusive of all

roadway users, coupled with increased enforcement, and increased perception of

enforcement. Increasing enforcement is most effective when media is also

emphasizing the enforcement to increase the perception of enforcement as well.

The increased enforcement should be placed on policies such as safety belt and

helmet use, booster-seat use, speeding, and distracted driving such as texting while

driving. Reducing these directly related risk-taking characteristics should produce

the greatest associated decrease in fatality rates in the southeastern U.S.

In addition to reducing the risk-taking behaviors that have a direct effect on

fatality rates, efforts should also be made to implement education and counseling to

the share of population that is known to behave in risky manners. Education and

counseling will not eliminate the share of population that behaves in a risky manner

while driving; however it may help to reduce the risk-taking behaviors of those

individuals who are not aware of the risk-taking characteristics they possess. Some

individuals display risk-taking characteristics because they don’t care, nor do they

desire to reduce their amount of risk-taking behaviors. The implementation of

counseling and educating services will not have much impact on individuals who do

not want to change their habits, however there is likely a share of the population

who displays risk-taking characteristics because they do not know any better, or are

unaware of the consequences that may accompany their behavior. For the share of

population who would behave in a less risky manner if they knew better, education

and counseling could be beneficial. The counseling and education referred to should

emphasize the reduction of risk-taking characteristics directly related to fatality

rates such as increasing safety belt and helmet use, reducing distractions while

driving, reducing driving under the influence, reducing speed, and driving

defensively rather than aggressively. If education and counseling implementations

reduce overall risk-taking behavior, then the fatality rates would also decrease,

increasing the total safety of the roadway network as a whole.

Additional efforts focusing on the addition of more variables could

significantly enhance the usefulness of the efforts put forth in this thesis. Increasing

the number of variables included in the regression analysis would help to reduce the

error associated with the unobserved characteristics and increase the usefulness of

the results by including more suggestions of policy implications. In addition to

increasing the number of variables, decreasing the size of data aggregation would

help to reduce aggregation bias and possibly infer better focused policy

implications. In addition, remediation actions are easier and less expensive at small

geographies, rather than at the state level, and the aggregation of the data used in

the analyses of this thesis does not allow for small geography associations and the

subsequent targeted actions.

Carrying out the analyses in this thesis on sub-segments of the overall

fatality data may enhance the usefulness of the linear regression estimation. For

instance, comparing the independent risk-taking variables with only alcohol related

fatalities may help to clarify the characteristics specifically associated to driving

while intoxicated. Similarly sub segments such as urban versus rural or freeway

versus non-freeway may help to distinguish what types of areas are most influenced

by specific characteristics within specific designs. For instance most freeways are

relatively constant over geographies, therefore the variation of fatality rates on

freeways by state may show a better indication of the influence specific risk-taking

traits of the population in that area have on fatality rates. A further exploration of

daytime versus nighttime and vulnerable versus vehicular road user fatalities may

help to distinguish why the southeastern states have a higher pedestrian fatality

rate when compared to the other states in the U.S. and what risk-taking

characteristics are playing a significant role in the fatality rate differences.

This thesis can be used as a starting point for additional analysis that can

assist states in moving closer to a zero traffic fatality goal. As previously mentioned,

safety is a complex problem and numerous strategies will need to be deployed

together to find solutions which work to reduce fatalities in each area. Reducing

risk-taking behaviors -traits that are disproportionately common in the southeastern

states - of drivers, pedestrians and bicyclists are a component of the myriad of

activities to improve the transportation vehicles, infrastructure, enforcement and

education efforts directed toward improving travel safety.

INTRODUCTION Traffic fatalities account for an astonishing

Students also viewed