Effectiveness of enforcement levels of speed limit and drink driving ...

j o u r n a l o f t r a f fi c a n d t r a n s p o r t a t i o n e n g i n e e r i n g ( e n g l i s h e d i t i o n ) 2 0 1 7 ; x ( x ) : 1 e8

Available online at www.sciencedirect.com

ScienceDirect journal homepage: www.elsevier.com/locate/jtte

Original Research Paper

Effectiveness of enforcement levels of speed limit and drink driving laws and associated factors e Exploratory empirical analysis using a bivariate ordered probit model Behram Wali a,b, Anwaar Ahmed c,*, Shahid Iqbal c, Arshad Hussain a a National Institute of Transportation, School of Civil and Environmental Engineering, National University of Sciences and Technology, Islamabad 44000, Pakistan b Department of Civil and Environmental Engineering, The University of Tennessee, Knoxville, TN 37996, USA c Military College of Engineering, National University of Sciences and Technology, Risalpur 24080, Pakistan

highlights Simultaneous association between drink driving and speeding among fatally injured drivers has been quantified. Bivariate ordered probit model is statistically superior compared to univariate counterpart. Socioeconomic factors, fatalities and highway agency road safety policies are simultaneously associated with enforcement levels of speed limit and drink driving laws.

article info

abstract

Article history:

The contemporary traffic safety research comprises little information on quantifying the

Available online xxx

simultaneous association between drink driving and speeding among fatally injured

Keywords:

substantial methodological concern which needs investigation. This study therefore

Speed limit law

focused on investigating the simultaneous impact of socioeconomic factors, fatalities,

Drink driving law

vehicle ownership, health services and highway agency road safety policies on enforce-

Enforcement

ment levels of speed limit and drink driving laws. The effectiveness of enforcement levels

Road traffic crashes

of speed limit and drink driving laws has been investigated through development of

drivers. Potential correlation between driver's drink driving and speeding behavior poses a

bivariate ordered probit model using data extricated from WHO's global status report on road safety in 2013. The consistent and intuitive parameter estimates along with statistically significant correlation between response outcomes validates the statistical supremacy of bivariate ordered probit model. The results revealed that fatalities per thousand registered vehicles, hospital beds per hundred thousand population and road safety policies are associated with a likely medium or high effectiveness of enforcement levels of speed limit and drink driving laws, respectively. Also, the model encapsulates the effect of several other agency related variables and socio-economic status on the response

* Corresponding author. Tel.: þ92 346 8123478. E-mail addresses: [email protected] (B. Wali), [email protected] (A. Ahmed), [email protected] (S. Iqbal), arshad_ [email protected] (A. Hussain). Peer review under responsibility of Periodical Offices of Chang'an University. http://dx.doi.org/10.1016/j.jtte.2017.04.001 2095-7564/© 2017 Periodical Offices of Chang'an University. Publishing services by Elsevier B.V. on behalf of Owner. This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/).

Please cite this article in press as: Wali, B., et al., Effectiveness of enforcement levels of speed limit and drink driving laws and associated factors e Exploratory empirical analysis using a bivariate ordered probit model, Journal of Traffic and Transportation Engineering (English Edition) (2017), http://dx.doi.org/10.1016/j.jtte.2017.04.001

2

J. Traffic Transp. Eng. (Engl. Ed.) 2017; x (x): 1e8

outcomes. Marginal effects are reported for analyzing the impact of such factors on intermediate categories of response outcomes. The results of this study are expected to provide necessary insights to elemental enforcement programs. Also, marginal effects of explanatory variables may provide useful directions for formulating effective policy countermeasures for overcoming driver's speeding and drink driving behavior. © 2017 Periodical Offices of Chang'an University. Publishing services by Elsevier B.V. on behalf of Owner. This is an open access article under the CC BY-NC-ND license (http:// creativecommons.org/licenses/by-nc-nd/4.0/).

1.

Introduction

Speeding and drink driving are the key risk factors that significantly contribute to road traffic crashes (RTCs) (WHO, 2013). Specifically, drink driving contributes to as much as 30%e40% of road deaths globally which validates the influence of alcohol on risk taking driving behaviors (WHO, 2013). Quantitatively, in 2012, approximately 10,322 deaths in United States occurred due to crashes involving drunk drivers that equate to single fatal accident every 51 min (Blincoe et al., 2002). The situation is equally frightening on global level with almost 20% of all road traffic crashes related to drink driving (Room et al., 2005). Likewise, in year 2011, 30.72% of the total (32,367) road traffic fatalities (RTFs) occurred due to speeding (FHWA, 2011). Speeding was found as a contributory factor in 44% of the total fatal road crashes in UK (Clarke et al., 2010). One conclusion from Clarke et al. (2010) study of particular relevance is that approximately 65% of the reported crashes involved speeding, drunk driving, and/or non-usage of seat belts. Another study concluded effectiveness of publicity and enforcement campaigns against drunk driving and speeding in reducing RTCs (Tay, 2005). Therefore, effective enforcement of speeding and drink driving laws is crucial in reducing the global burden of excessive RTCs. Also, it seems imperative to understand the impact of several factors that may be associated or correlated with effectiveness of enforcement levels of speed limit and drink driving laws. A rigorous analysis of country-level data files regarding the road safety will eventually reveal covariates that may be associated with high or low effectiveness of speed limit and drink driving enforcement. Thus, the objective of present study is to investigate the simultaneous association between socioeconomic factors, vehicle ownerships, fatalities, agency road safety polices, health services, and effectiveness of speed and drink driving enforcement, respectively. Utilizing country level data from World Health Organization (WHO) and International Road Federation (IRF), the bivariate ordered probit modeling framework encapsulates the effect of several explanatory variables on two response outcomes.

2.

Literature review

Several independent studies concluded drivers' non-compliance to drink driving law as a major reason for occurrence of road traffic crashes. Specifically, these studies concluded that blood-alcohol concentration (BAC) of drivers is much above

the recommended statutory limit (i.e., BAC of 0.05 g/dL) for up to half of total road traffic crashes (Baker et al., 2002; Evans, lez-Wilhelm, 2007; Hingson et al., 2002; Voas et al., 1990; Gonza 2006). Moreover, a more pronounced effect of drink driving has been observed for young aged drivers, night time driving, and weekend driving, respectively. For example, an application of conditional logistic regression technique on case control data revealed positive BACs for majority of young drivers (aged 21 or less) involved in road crashes, where alcohol intake was suspected to affect crash avoidance skills of novice and young drivers (Peck et al., 2008). Likewise, for drivers of age 40 or less driving on lower volume roads, as much as 50% of night time risk (on weekend) was encapsulated by the odd consequences of alcohol (Keall et al., 2005). Also, several achievements have been made in terms of analyzing the vulnerability of young and novice drivers to consuming alcohol and the resulting risk of crash. The occurrence of road crash as a result of driver's first drink driving offense and more preferably at young age was concluded as an important factor in perception of drivers to repeat drink driving offense (Ferrante et al., 2001). Similarly, it was found that the proportion of New Zealand young drivers who consume large amount of alcohols are 2.6 times more likely to get engage in road traffic crash than those who do not undertake drunk driving (Horwood and Fergusson, 2000). As a result, the alleviated pattern of driver's risks and crash severities that results due to consumption of alcohol are more profound for young divers than the adult ones (Walker et al., 2005; Williams, 2003). In fact, the strong association between intensive outcome of a crash (injury severity) and drink driving is validated by a broad spectrum of methodological studies including (but not limited to) discrete choice models (Jiang et al., 2015), generalized ordered logit (Abegaz et al., 2014), ordered probit (Haleem and Gan, 2011), longitudinal empirical analysis (Morrison et al., 2002), mixed logit (Li et al., 2014), ordered logistic regression (MacKenzie et al., 2015), and random coefficient heteroskedastic ordered response model (Paleti et al., 2010). Due to the strong influence of drink driving on crash risk and severity, several researchers have studied the impact of various enforcement programs in an attempt to reduce alcohol-related RTCs. Past research has also been focused on quantifying the efficacy of publicity and enforcement schemes, social marketing, effectiveness of enforcement intensity, and drink driving enforcement strategies in reducing alcohol-involved RTCs (Li et al., 2012; Nguyen et al., 2012; Tay, 2005). A detailed synthesis of literature thus revealed drink driving as an important determinant of probability of crash involvement. Having this said, stringent enforcement of drink driving law


3


Table 1 e Cross tabulated summary of dependent variables. Effectiveness of enforcement level of speed limit law Very low Low Medium High Total

Effectiveness of enforcement level of drink driving law Very low

Low

Medium

High

32 9 10 0 51 (30.000%)

13 17 8 2 40 (23.529%)

6 7 23 5 41 (24.117%)

0 3 18 17 38 (22.354%)

is highly likely to reduce drivers' propensity of being involved in a crash. On the other hand, speeding is the most important determinant of crash occurrence and crash severity. The effectiveness of police enforcement and/or publicity campaigns is not clear with some recent studies concluding negligible effects of police enforcement and publicity campaigns (Goldenbeld and Schagen, 2005) while others (out of many) concluding significant effects of police enforcement in reducing driver's speeding violations (Walter et al., 2011). Importantly and in special relevance to our work, a partial proportion odd model concluded both speeding and drink driving as significant contributors to crash severity with highest effects on severe and/or fatal accidents (Abegaz et al., 2014). The investigation of association between drink driving and speeding among fatally injured drivers of Norway found that approximately 71.7% of all drink driving fatal road crashes involved speeding drivers (Bogstrand et al., 2015). As a result of vast literature synthesis, it can be inferred that stringent enforcement of drink driving law and/or speed limit law is highly likely to reduce drivers' propensity of being involved in a crash. From enforcement perspective, an understanding of how effectiveness of drink driving and speed limit enforcement varies across various countries will eventually highlight important factors that may contribute to effectiveness of such enforcement programs. Also, drink driving and speeding have been found correlated in contributing to overall risk of crash involvement (Abegaz et al., 2014; Bogstrand et al., 2015). Thus, it seems important to evaluate the country-wise effectiveness of enforcement programs of speeding and drink driving simultaneously. However, recent literature possesses meager evidence of studies that categorically evaluated the factors that can simultaneously contribute to low or high effectiveness of drink driving and speeding enforcement programs. The present study attempts to bridge this gap by developing a methodological framework (bivariate ordered probit model) in order to disentangle the simultaneous complex relationships between several explanatory factors and effectiveness of drink driving and speed limit enforcement laws.

3.

Data

The data for effectiveness of drink driving and speed limit enforcement programs for 178 countries were obtained from WHO's Global Status Report on Road Safety (WHO, 2013). The overall dataset includes information on 37 different variables related to population, socioeconomic status (e.g.

Total

51 (30.00%) 36 (21.10%) 59 (34.70%) 24 (14.11%) 170

gross national income (GNI) per capita), annual fatal road traffic crashes, road safety policies, speed limits on urban and rural roads, and health care system. Also, data such as total number of registered vehicles, and hospital beds per hundred thousand population (HBPHTP) were extracted from International Road Federation (IRF). Information on presence or absence of policy for promoting walking and cycling (PWC), national child restraint law (NCRL), road safety audits of existing roads (RSAER), mandatory installation of seat belt for front and rear occupants, vital registration system (VRS), national helmet law, presence of lead agency, and drink driving law (DDL) was also extracted from WHO's GSRS e2013 (WHO, 2013). Specifically, legislative data regarding five key risk factors were collected and specific criteria were observed to be vital elements of comprehensive legislation for five key risk factors. For example, comprehensive legislation for speeding was defined as existence of national level speed limit law with an urban speed limit of either 50 km/h or less. Likewise, for data collection, comprehensive legislation for drunk driving was defined as existence of national drink driving law where blood alcohol concentration is less than or equal to 0.05 g/dL for general population. In each country, selected road safety experts were asked to respond to a structured questionnaire in an attempt to rate the effectiveness of enforcement level(s) of five key risk factors subjectively on scale 0e10 (0 being “not effective” and 10 being “highly effective”). Out of 178 countries, 170 were considered for model development due to incomplete information for remaining eight countries. The final data set contains measurements of two response variables for 170 countries, out of which, 33, 94 and 43 belong to low-income (GNI per capita < $1045), mediumincome (GNI per capita $1045e$12,746) and high-income (GNI per capita > $12,746) groups, respectively. A crosstabulated summary of the two response variables is shown in Table 1. For instance, it can be seen that compared to 48.8% of countries having at least medium level of speed limit enforcement, 46.4% of countries possessed at least medium level of drink driving enforcement too. Also, the distributions of very low and low level of speed limit and drink driving enforcement across sampled countries are approximately similar. These findings suggest that effectiveness of enforcement levels of speed limit and drink driving laws are significantly associated with each other. Finally, the indicator variables along with their description and mean values are presented in Table 2. Based on the descriptive statistics, the data seem to be of reasonable quality. Referring to Table 2, almost 43 countries (0.252 170) possessed national policy for promoting


4


Table 2 e Description of independent variables (co-variate). Variable Fatality indicator Hospital beds indicator Income indicator Policy indicator Child indicator Seat belt indicator Safety strategy indicator Drink indicator Audit indicator Registration indicator

Description

Mean

1 if fatalities per thousand registered vehicle (FPTRV) is less than 2.25, 0 otherwise 1 if hospital beds per hundred thousand population (HBPHTP) is greater than 71, 0 otherwise 1 if gross national income per capita is less than $1045, 0 otherwise 1 if national policy for promoting walking and cycling exists, 0 otherwise 1 if national child restraint law exists, 0 otherwise 1 if seat belt law applies to front and rear occupants, 0 otherwise 1 if national road safety strategy exists, 0 otherwise 1 if national drink driving law exists, 0 otherwise 1 if roads audits of existing roads exists, 0 otherwise 1 if vital registration system exists, 0 otherwise

0.700

walking and cycling, and 87 countries possessed national child restraint law, whereas national road safety strategy existed in 133 of the sampled countries. Other variables in Table 2 can be interpreted in similar way. For a more detailed explanation of data, interested readers are suggested to consult the WHO report (WHO, 2013).

4.

Methodology

The statistically significant association between speeding and drink driving provides the main motivation for developing analytical model that accounts simultaneously for factors influencing the effectiveness of speed limit and drink driving enforcement programs, respectively. The country level survey conducted by WHO recorded enforcement levels for all key risk factors on scale ranging from 0 to 10. For this study, we have broadly classified enforcement levels of drink driving and speed limit laws into four categories. Thus, each response variable can take four discrete outcomes demarcated as very low (0, 1, 2, 3), low (4, 5), medium (6, 7), and high (8, 9, 10) level of enforcement all on 0e10 scale. The rationale behind the categorization is three-fold: 1) as compared to countries reporting low level of enforcements, few countries across the globe reported enforcement 8 for any of enforcement laws for five key risk factors; 2) the distribution of dependent variables (effectiveness of speed limit and drink driving enforcement) is not normal and thus cannot be reasonably modeled with alternative regression approaches; 3) the cut-off point (i.e., enforcement of 8 on scale 0e10) has been worked out in accordance with the cut-off point by World Health Organization (WHO, 2013). A review of the summary statistics (Table 1) affirms that speed limit law tend to be more highly enforced than drink driving law. Only 48.8% (83) of the countries possess high or medium level of speed limit law enforcement, compared to 46.4% (79) of the countries possessing high or medium level of drink driving law enforcement, respectively. For two response variables, each having an implicit ordered pattern, bivariate ordered probit (BOP) model can be formulated as follows (Anastasopoulos et al., 2012; Greene and Hensher, 2010). Ai;1 ¼ b1 Zi;1 þ 3i;1 ; Ai;1 ¼ x if mx1 < Ai;1 < mx ; x ¼ 0; 1; /; X1

(1)

Ai;2 ¼ b2 Zi;2 þ 3i;2 ; Ai;2 ¼ x if qx1 < Ai;2 < qx ; x ¼ 0; 1; /; X2

(2)

0.094 0.194 0.252 0.511 0.641 0.782 0.952 0.470 0.882

where Ai,1 and Ai,2 indicate response variables which refer to absolute integer pattern ordering for effectiveness of enforcement levels of speed limit and drink driving laws for individual countries, b and Zi refer to estimable parameters vectors and vectors of independent variables (as indicated in Table 2) that may influence the effectiveness of enforcement levels of speed limit and drink driving laws respectively, ei is the assumed normally distributed error terms, m and q are the threshold parameters among which the response variable takes several defined value outcomes respectively, X is the integer ordered effectiveness of enforcement levels of speeding and drink driving laws. In this study, we have assumed the error terms to be normally distributed and a probit link is thus used in estimation framework. However, if the error terms are expected to be logistic distributed, ordered logit framework may also be used (Sun and Elefteriadou, 2014). Finally, through a simultaneous model estimation framework such as bivariate ordered probit (BOP) model, polychoric cross equation correlated error terms are also estimated through BOP model which are as

3i;1 3i;2

N

0 1 r ; 0 r 1

(3)

where r is the coefficient of cross-equation error terms ranging between 0 and 1. Please note that bivariate ordered probit model accounts for the correlation between error terms of two univariate ordered probit models. Ignoring the potential correlation may under- or over-estimate the model residuals and standard errors associated with estimable parameters, which may further affect the statistical significance of estimable parameters. Afterwards, BOP model can be estimated while accounting for the joint probability of ordered selection (i.e., yi,1 ¼ j and yi,2 ¼ k), as follow P Ai;1 ¼ x; Ai;2 ¼ y Zi;1 ; Zi;2 ∅2 mx b1 Zi;1 ; qy b2 Zi;2 ; r ¼ ∅2 mx1 b1 Zi;1 ; qy b2 Zi;2 ; r ∅2 mx b1 Zi;1 ; qy1 b2 Zi;2 ; r ∅2 mx1 b1 Zi;1 ; qy b2 Zi;2 ; r

(4)

where ∅ refers to the standard normal cumulative distribution function (Anastasopoulos et al., 2012; Greene and Hensher, 2010). In Eq. (4), if r z 0 the estimation framework reduces to two univariate ordered probit models in which case bivariate ordered probit model will effectively be equal to two


5


univariate ordered probit models. However, in the case under discussion, we hypothesize under null hypothesis that error terms of Eqs. (1) and (2) are significantly correlated and thus should be modeled simultaneously. We discuss in detail the afore-mentioned hypothesis and its implications in results and analysis section. Parameter estimates along with conceivable signs are of premier interest while interpreting the effect of explanatory variable on extreme outcomes of response variable (Washington et al., 2010). For instance, a positive sign of parameter estimate indicates an increase in Z is highly likely to increase the probability of getting highest ordered outcome (i.e., A ¼ high effectiveness of enforcement level) while simultaneously decreasing the probability of getting lowest ordered outcome (i.e., A ¼ very low enforcement level). In order to understand the effects of explanatory variables on intermediate categories of response variables, we compute marginal effects as follow (Anastasopoulos et al., 2012; Greene, 2008; Washington et al., 2010). PðA ¼ xÞ ¼ ½4ðux1 bBÞ 4ðux bBÞb vZ

(5)

where P(A ¼ x) refers to the probability of a certain country experiencing x level of effectiveness of enforcement level of either speed limit or drink driving law. The threshold parameters denoted by u are the absolute integer ordered effectiveness of enforcement levels of speeding and drink driving, respectively. While 4 refers to probability mass function (pmf) of assumed standard normal distribution.

5.

Results and analysis

We estimated bivariate ordered probit model that showed statistically significant (at 95% confidence level) correlation between effectiveness of speed limit and drink driving enforcement programs within each country (Table 3). The parameter r (polychoric correlation parameter) elucidates the association between unobserved factors (not captured by the data) that are likely to affect the outcomes within individual countries. The correlation parameter (i.e., 0.218) suggests a positive correlation between effectiveness of enforcement levels of speed limit and drink driving laws. Hence forward, it is likely that such unobserved factors tend to simultaneously increase or decrease the effectiveness of enforcement levels of speed limit and drink driving laws for different countries. The statistically significant polychoric correlation coefficient (i.e., 0.218 with t-stat of 3.815) establishes the fact that error terms of enforcement levels of speed limit and drink driving laws are significantly correlated and should be modeled jointly. As a result, we fail to reject the null hypothesis that errors terms are uncorrelated. Moreover, the log-likelihood at convergence are given for the final bivariate ordered probit model and two univariate ordered probit counterparts (Table 3). In Table 3, as compared to the combined log-likelihood of two univariate models (i.e., 394.49), modeling enforcement levels of speed limit and drink driving laws simultaneously has resulted in significant improvement in log-likelihood (i.e., log-likelihood of 314.23) for bivariate ordered probit model, respectively.

Table 3 e Parameter estimates for bivariate ordered probit model. Variable

Effectiveness of enforcement level of speed limit law

Effectiveness of enforcement level of drink driving law

Coefficient t-stat Coefficient t-stat Constant Fatality indicator Hospital beds indicator Income indicator Policy indicator Child indicator Seat belt indicator Safety strategy indicator Drink indicator Audit indicator Registration indicator m1 m2 Log-likelihood (univariate) Log-likelihood (bivariate) r

0.745 0.897 0.972

2.822 3.657 2.225

0.521 0.503 0.325 0.351

1.879 2.023 1.286 1.409

0.762 2.770 195.94

Coefficient (0.218)

6.094 10.480

1.949 0.539 0.786 0.901 0.281

2.747 2.003 2.378 2.554 1.261

1.371 0.317 0.621 0.833 1.909 198.55

2.412 1.707 1.425 6.047 10.075

314.23 t-stat (3.815)

Altogether, these results reveal the statistical superiority of bivariate ordered probit model as compared to two univariate ordered probit models, and the need to model and investigate enforcement levels of speed limit and drink driving laws jointly. The detailed results of the bivariate ordered probit model are presented (Table 3) for understanding the nature and impact of various explanatory variables. Analyzing the parameter estimates for effectiveness of enforcement level of speed limit law, the fatality indicator variable (fatalities per thousand registered vehicles of less than 2.25) has positive coefficient suggesting higher likelihood of medium or high effectiveness of enforcement level of speed limit law. Regarding coding the fatality indicatory variable, value of 2.25 is selected as it corresponds to only 21.8% of the countries. In other words, approximately 78.2% of the countries are observed to possess fatalities per thousand registered vehicles of greater than 2.25. Likewise, positive coefficient of the hospital bed indicator variable shows that countries with more than 71 hospital beds per hundred thousand population (HBPHTP) have higher likelihood of possessing medium or high effectiveness of enforcement level of speed limit law and vice versa. The cut-off value for this indicator variable represents 10% (16 countries) of the total countries with HBPHTP of greater than 71. With a focus, countries with fewer fatalities and more hospital beds are very much likely to possess integrated health care system simultaneously with an efficient and safe driving culture. These findings are consistent with the results of past studies that concluded associations between pre-hospital care and health care proportional spending and RTCs (Anwaar et al., 2012; Soole et al., 2013). The policy (presence of national policy for promoting walking and cycling) and child indicator (presence of national child restraint law) variable intuitively exhibit positive


6


coefficients (0.521 and 0.503) with all countries that possess such laws being more likely to exhibit higher effectiveness of enforcement levels of speed limits. This is understandable as optimal usage of child restraint is found effective in reducing road traffic fatalities, both for adults and children which may passively portray safe speeding choice of considerable driver's population (Arbogast et al., 2005; Brown et al., 2006). On the other hand, policy for promoting walking and cycling (PWC) characterize road agency's concern regarding legislating road user's safety, both for vulnerable and non-vulnerable users. For example, such policy is very much likely to be mandated in urban areas which may in turn reduce vehicle kilometers traveled due to adequate pedestrian and bicycle pathways, thus likely resulting in enhanced safety particularly for vulnerable road users. Also, such a policy is likely to remove dangerous proportion of speeding drivers by providing them alternative forms of travel commutes (i.e., biking). The seat belt indicator (mandatory application of seat belt for both front and rear occupants) marginally significantly increases the probability of very low or low effectiveness of speed limit enforcement, respectively. This result is perceivable as habitual risk taking drivers may feel safe under the added security which seat belts may provide and may not consider over-speeding as a threat to their personal safety. Nevertheless, further investigation is needed for studying the effects of compliance of risk taking drivers to seat belt usage and their speeding choice, respectively. The existence of national road safety strategy (NRSS) is also found correlated with medium/high effectiveness of speed limit enforcement. The importance of presence of road safety strategies towards effectiveness of enforcement programs have long been validated by several researchers too (Amador and Willis, 2014). The influences of fatality, hospital beds, and policy indicator variables all have consistent effects on effectiveness of drink driving enforcement as they do on effectiveness of speed limit enforcement. For example, countries with less than 2.25 FPTRV are more likely to possess medium or high enforcement level of drink driving law. Similar effects were observed for hospital beds and policy indicator variables too. While the coefficient estimates presented in Table 3 administer the effects of explanatory variables on extreme outcomes of response variable, Table 4 presents marginal effects related to the significant explanatory variables (Duncan et al., 1998; Washington et al., 2010). For instance, with each one-unit increase in country possessing policy for PWC, the probabilities of exhibiting high level of effectiveness of speed limit and drink driving enforcement were observed to increase by 7.05% and 13.73%, respectively. Similarly, for each country, the income indicator variable (countries with GNI per capita less than $1045) pronounce the strong negative association between low income and corresponding effectiveness of enforcement level of drink driving law. For example, a one-unit increase in number of countries having GNI per capita of less than $1045 results in an average 0.281 unit increase in the probability of such countries to exhibit very low level of effectiveness of drink driving enforcement. The effects of low income turned out to be unreasonable in speed limit enforcement equation due to statistically insignificant parameter estimates. Countries under low income group may have deficit resources to

Table 4 e Marginal effects. Effectiveness of enforcement level Speed limit law Fatality indicator Hospital beds indicator Policy indicator Child indicator Seat belt indicator Safety strategy indicator Drink driving law Fatality indicator Hospital beds indicator Income indicator Policy indicator Drink indicator Audit indicator Registration indicator

Very low

Low Medium High

0.307 0.134 0.114 0.134 0.010 0.197

0.024 0.057 0.037 0.031 0.002 0.019

0.215 0.090 0.081 0.097 0.007 0.142

0.116 0.101 0.070 0.068 0.005 0.074

0.196 0.119 0.281 0.150 0.315 0.148 0.184

0.018 0.046 0.004 0.050 0.038 0.032 0.003

0.093 0.048 0.132 0.063 0.146 0.068 0.088

0.121 0.116 0.144 0.137 0.130 0.112 0.099

initiate, monitor, enforce and penalize drink driving violations (Ferguson et al., 1999). Also, out of all low income countries that possess very low level of drink driving law enforcement (24 countries), 70.8% (17 countries) belong to African continent whose RTFs were attributed to poor and inadequate drink driving enforcement in past studies (Aguwa and Anosike, 1982; Odero, 1998). To provide some insight into the impact of highway agency road safety policies on effectiveness of enforcement level of drink driving law, the drink indicator variable suggests that existence of national drink driving law (DDL) is likely to result, for majority of countries, in high level of effectiveness of drink driving enforcement. Intuitively, almost 73% of the countries (possessing medium or high effectiveness of drink driving enforcement) belong to middle income (>$4125) groups or above. With this said, middle and/or high income group countries may exhibit determined political will, good and ethical police practice (something rare in low income countries), efficient enforcement campaigns, educational schemes for public awareness, and rigorous penalties for law offenders. The audit and registration indicator variable intuitively features the influence of existence of road safety audits of existing roads and presence of vital registration system (VRS) towards the effectiveness of enforcement level of drink driving law. As such, countries that possess such laws are more likely to assess their highway networks for necessary safety standards, thus leading to properly engineered and enforced highway system with due circumspection for unsafe and intoxicated drivers. Additionally, a well versed database (existence of vital registration system) is likely to assist road agencies and policy makers for identifying potential areas for resource allocation and funding.

6.

Conclusions

Given the potential relationship between speeding and drink driving, no efforts (to the best of our knowledge) have been concentrated on investigating the simultaneous impact of socioeconomic factors, fatalities, and agency related policies on effectiveness of enforcement levels of speed limit and drink driving laws. Thus, a bivariate ordered probit model has



been developed to investigate the impact of factors (covariates) on the effectiveness of speed limit and drink driving enforcement, while taking into account the correlation between two response outcomes. The statistically significant correlation and efficient parameter estimates validate the efficiency of the proposed model as compared to univariate ordered probit models. The impact of fatality, hospital beds, and policy indicator variables all have consistent effects (direct relationship) on drink driving enforcement as they do on speed limit enforcement. Presence of drink driving law, road audits of new roads, and vital registration system were also found associated with likely medium or high effectiveness of drink driving enforcement, respectively. Furthermore, the computation of marginal effects provides insight regarding the impact of explanatory variables on extreme and intermediate categories of the response variables. The child, seat belt, and road safety indicator variables pronounced the intuitive impact of agency policies on effectiveness of speed limit law enforcement. All other being equal, countries with mandatory application of seat belt law for both front and rear occupants are likely to exhibit very low or low level of speed limit law enforcement. However, further investigation is needed for studying the effects of compliance of drivers to seat belt usage and their speeding choice while taking into account the driver's characteristics, trip type and purpose, and number of occupants, respectively.

references

Abegaz, T., Berhane, Y., Worku, A., et al., 2014. Effects of excessive speeding and falling asleep while driving on crash injury severity in Ethiopia: a generalized ordered logit model analysis. Accident Analysis & Prevention 71, 15e21. Aguwa, C., Anosike, E., 1982. Road accidents in Nigeria: level of alcohol in the blood of automobile drivers. Central African Journal of Medicine 28 (7), 171e174. Amador, L., Willis, C.J., 2014. Demonstrating a correlation between the maturity of road safety practices and road safety incidents. Traffic Injury Prevention 15 (6), 591e597. Anastasopoulos, P.C., Karlaftis, M., Haddock, J., et al., 2012. Household automobile and motorcycle ownership analyzed with random parameters bivariate ordered probit model. Transportation Research Record 2279, 12e20. Anwaar, A., Anastasopoulos, P., Ong, G.P., et al., 2012. Factors affecting highway safety, health care services, and motorizationdan exploratory empirical analysis using aggregate data. Journal of Transportation Safety & Security 4 (2), 94e115. Arbogast, K.B., Ghati, Y., Menon, R.A., et al., 2005. Field investigation of child restraints in side impact crashes. Traffic Injury Prevention 6 (4), 351e360. Baker, S.P., Braver, E., Chen, L., et al., 2002. Drinking histories of fatally injured drivers. Injury Prevention 8 (3), 221e226. Blincoe, L.J., Seay, A., Zaloshnja, E., et al., 2002. The Economic Impact of Motor Vehicle Crashes 2000. DOTHS809446. U.S. Department of Transportation, Washington DC. Bogstrand, S.T., Larsson, M., Holtan, A., et al., 2015. Associations between driving under the influence of alcohol or drugs, speeding and seatbelt use among fatally injured car drivers in Norway. Accident Analysis & Prevention 78, 14e19. Brown, J., Mccaskill, M.E., Henderson, M., et al., 2006. Serious injury is associated with suboptimal restraint use in child

7

motor vehicle occupants. Journal of Paediatrics and Child Health 42 (6), 345e349. Clarke, D.D., Ward, P., Bartle, C., et al., 2010. Killer crashes: fatal road traffic accidents in the UK. Accident Analysis & Prevention 42 (2), 764e770. Duncan, C.S., Khattak, A.J., Council, F.M., et al., 1998. Applying the ordered probit model to injury severity in truck-passenger car rear-end collisions. Transportation Research Record 1635, 63e71. Evans, L., 1990. The fraction of traffic fatalities attributable to alcohol. Accident Analysis & Prevention 22 (6), 587e602. Ferguson, M., Sheehan, M., Davey, J., et al., 1999. Drink Driving Rehabilitation: The Present Context. CR 184. Australian Transport Safety Bureau, Canberra. Ferrante, A.M., Rosman, D.L., Marom, Y., 2001. Novice drink drivers, recidivism and crash involvement. Accident Analysis & Prevention 33 (2), 221e227. FHWA, 2011. Speed Management Safety. Available at: http:// safety.fhwa.dot.gov/speedmgt/ (Accessed 31 May 2015). Goldenbeld, C., Schagen, I.V., 2005. The effects of speed enforcement with mobile radar on speed and accidents: an evaluation study on rural roads in the Dutch province Friesland. Accident Analysis & Prevention 37 (6), 1135e1144. lez-Wilhelm, L., 2007. Prevalence of alcohol and illicit drugs Gonza in blood specimens from drivers involved in traffic law offenses. Systematic review of cross-sectional studies. Traffic Injury Prevention 8 (2), 189e198. Greene, W.H., Hensher, D.A., 2010. Modeling Ordered Choices: A Primer. Cambridge University Press, Cambridge. Greene, W.H., 2008. LIMDEP, Version 9.0.4. Econometric Software, Inc., Plainview. Haleem, K., Gan, A., 2011. Identifying traditional and nontraditional predictors of crash injury severity on major urban roadways. Traffic Injury Prevention 12 (3), 223e234. Hingson, R.W., Heeren, T., Zakocs, R.C., et al., 2002. Magnitude of alcohol-related mortality and morbidity among US college students ages 18-24. Journal of Studies on Alcohol 63 (2), 136e144. Horwood, L.J., Fergusson, D.M., 2000. Drink driving and traffic accidents in young people. Accident Analysis & Prevention 32 (6), 805e814. Jiang, X., Zheng, H., Qiu, Y., et al., 2015. Differences in injury severities between 2-vehicle and 3-vehicle crashes. Traffic Injury Prevention 16 (3), 289e297. Keall, M.D., William, J.F., Patterson, T.L., et al., 2005. The contribution of alcohol to night time crash risk and other risks of night driving. Accident Analysis & Prevention 37 (5), 816e824. Li, Y., Sze, N.N., Wong, S.C., 2014. Effective measures for combating drink-driving offenses: an attitudinal model for Hong Kong. Transportmetrica A: Transport Science 10 (8), 722e739. Li, Y., Xie, D., Nie, G., et al., 2012. The drink driving situation in China. Traffic Injury Prevention 13 (2), 101e108. MacKenzie, J.E., Watling, C.N., Leal, N.L., et al., 2015. What aspects of demographic, personality, attitudes and perceptions of law enforcement influence self-reported likelihood of drink driving? Journal of Risk Research 18 (9), 1203e1219. Morrison, L., Begg, D.J., Langley, J.D., 2002. Personal and situational influences on drink driving and sober driving among a cohort of young adults. Injury Prevention 8 (2), 111e115. Nguyen, N., Passmore, J., Pham, C., et al., 2012. Drink driving prevention in Ha Nam and Ninh Binh provinces of Viet Nam. Injury prevention 18 (S1), A188. Odero, W., 1998. Alcohol-related road traffic injuries in Eldoret, Kenya. East African Medical Journal 75 (12), 708e711. Paleti, R., Eluru, N., Bhat, C.R., 2010. Examining the influence of aggressive driving behavior on driver injury severity in traffic crashes. Accident Analysis & Prevention 42 (6), 1839e1854. Peck, R.C., Gebers, M.A., Voas, R.B., et al., 2008. The relationship between blood alcohol concentration (BAC), age, and crash risk. Journal of Safety Research 39 (3), 311e319.


8


Room, R., Babor, T., Rehm, J., 2005. Alcohol and public health. The Lancet 365 (9458), 519e530. Soole, D.W., Watson, B.C., Fleiter, J.J., 2013. Effects of average speed enforcement on speed compliance and crashes: a review of the literature. Accident Analysis & Prevention 54, 46e56. Sun, D.J., Elefteriadou, L., 2014. A driver behavior based lanechanging model for urban arterial streets. Transportation Science 48 (2), 184e205. Tay, R., 2005. The effectiveness of enforcement and publicity campaigns on serious crashes involving young male drivers: are drink driving and speeding similar? Accident Analysis & Prevention 37 (5), 922e929. Voas, R.B., Romano, E., Tippetts, A.S., et al., 2006. Drinking status and fatal crashes: which drinkers contribute most to the problem? Journal of Studies on Alcohol and Drugs 67 (5), 722e729. Walker, S., Waiters, E., Grube, J.W., et al., 2005. Young people driving after drinking and riding with drinking drivers: drinking locationsdwhat do they tell us? Traffic Injury Prevention 6 (3), 212e218. Walter, L., Broughton, J., Knowles, J., 2011. The effects of increased police enforcement along a route in London. Accident Analysis & Prevention 43 (3), 1219e1227. Washington, S.P., Karlaftis, M.G., Mannering, F.L., 2010. Statistical and Econometric Methods for Transportation Data Analysis, second ed. CRC Press, Boca Raton. WHO, 2013. Global Status Report on Road Safety, Supporting a Decade of Action. World Health Organization, Geneva. Williams, A.F., 2003. Teenage drivers: patterns of risk. Journal of Safety Research 34 (1), 5e15.

Behram Wali is a PhD student in Department of Civil and Environmental Engineering at The University of Tennessee. He is majorly involved in projects related to connected and automated vehicles, alternative fuel vehicles infrastructure deployment, and proactive transportation safety. He obtained his MS from National University of Sciences and Technology, Pakistan in 2015.

Anwaar Ahmed, PhD, is an associate professor of Department of Transportation and Geotechnical Engineering at Military College of Engineering, National University of Sciences and Technology, Risalpur, Pakistan. He received his MS and PhD from Purdue University, USA. His research interests include highway safety, highway asset management, highway user charging systems and econometric modeling.

Shahid Iqbal, PhD, is presently the dean of Military College of Engineering, National University of Sciences and Technology, Risalpur, Pakistan. He received his MS and PhD from Michigan State University, USA. His research interests include structural reliability, structural fire safety, design of structures against extreme loading and learning focused on outcome based education.

Arshad Hussain received his PhD degree in Department of Transportation Engineering from the Southwest Jiaotong University, China. He is an assistant professor at National University of Sciences and Technology, Pakistan. He is mainly involved in the study of road materials including properties and evaluation, analysis and design, maintenance and recycling, planning and construction.
