Transport Policy 42 (2015) 166–172

Contents lists available at ScienceDirect

Transport Policy

http://d0967-07

n Fax:E-m

journal homepage: www.elsevier.com/locate/tranpol

Factors associated with the relationship between non-fatal roadinjuries and economic growth

Teik Hua Law n

Road Safety Research Centre, Faculty of Engineering, Universiti Putra Malaysia, 43400 Serdang, Selangor, Malaysia

a r t i c l e i n f o

Article history:Received 9 October 2014Received in revised form29 May 2015Accepted 5 June 2015Available online 26 June 2015

Keywords:Kuznets relationshipNon-fatal road injuriesDemocracyPolitical stability

x.doi.org/10.1016/j.tranpol.2015.06.0040X/& 2015 Elsevier Ltd. All rights reserved.

þ60 386567129.ail address: lawteik@upm.edu.my

a b s t r a c t

This study reports the results of an empirical analysis of the Kuznets curve relationship between non-fatal road injuries and per-capita income. This relationship indicates that the number of road deathsincreases with increasing per-capita income at lower income levels, but decreases once it has exceeded athreshold level. We apply a fixed effects negative binomial regression analysis on a panel of 90 countriesover the period of 1963–2009. Results indicated evidence of an inverted U-shaped relationship betweeneconomic growth and non-fatal road injuries for both less developed and highly developed countries.Results also indicated that the turning point is higher in less developed countries than in higher de-veloped countries. The evidence presented in this study suggests that improvements in road infra-structure, the quality of regulatory institutions, and increase in the use of safer transport modes will helpreduce non-fatal road injuries.

& 2015 Elsevier Ltd. All rights reserved.

1. Introduction

The relevance of the relationship between economic growthand road safety has long been recognized. However, it has beenthe impact of economic growth on road deaths that has attractedthe most attention (Wintemute, 1985; Söderlund and Zwi, 1995;Beeck et al., 2000; Kopits and Cropper, 2005). Recently, numerousstudies have shown that there is an inverted U-shaped relation-ship between road deaths and economic growth (Bishai et al.,2006; Beeck et al., 2000; Moniruzzaman and Andersson, 2008).That is, road deaths increase at lower income levels, but decreaseonce the number has exceeded a threshold level. This relationshipis also known as the Kuznets curve relationship, which existedbetween income inequality and per-capita income (Kuznets, 1955).It is important to understand how economic growth affects roadsafety because road safety concerns could lead to a primary roadsafety policy agenda. This relationship has been explained byseveral studies to be the result of changes in the fundamentaldevelopment of a country. This includes changes in variations inrates among vulnerable non-motorized road users (Paulozzi et al.,2007), advances in medical services (Law et al., 2009, 2011; No-land, 2003), the quality of political institutions (Anbarci et al.,2006; Law et al., 2011), income equality (Anbarci et al., 2009) androad safety policies (Law et al., 2009).

The Kuznets curve hypothesis for road safety can be supportedby three theoretical explanations: the scale of economic activities,changes in vehicle composition, and a surge in demand for betterroad safety as per-capita income rises (Law et al., 2009). TheKuznets curve hypothesis contends that the number of roaddeaths increases initially as a country's economy develops, owingto the fact that the growth of a country's economy is accompaniedby a corresponding surge in demand for transportation services(Dargay and Gately, 1999). Several previous studies have foundthat one of the contributing factors to the increase in road crashesand injuries is the growing number of vehicles per-capita (Bishaiet al., 2006; Kopits and Cropper, 2005).

The composition effect depicts the change in vehicle compo-sition as a country's economy grows. This involves a change fromlow-threat and high-risk transport modes (such as pedestriansand bicyclists) to high-threat and low-risk motorized vehicles(Bhalla et al., 2007). In the initial stage of economic development,the total road death risk initially increases for the reason that anincrease in the number of motorized vehicles can pose an in-creasing threat to the predominantly more vulnerable transportmodes. Nevertheless, at higher levels of economic development,when a majority of the commuting population are less vulnerablemotor vehicle occupants, there is an increase in the number ofmotor vehicles causing a reduction in the total road death risk.This indicates an inverted U-shaped relationship between roaddeaths and income.

The abatement effect is the result of road death risk alleviation

T.H. Law / Transport Policy 42 (2015) 166–172 167

measures, which reflects both supply side and demand influences.On the demand side, at low income levels, people are less capableto invest in road safety, even though there is a demand for it, re-sulting in low abatement. Nonetheless, as the level of income rises,people pay greater attention to road safety and they can affordhigher-cost road safety enhancement measures, enabling morecommuters to switch to lower-risk modes of transport. On thesupply side, at low levels of income, society is unable to affordsufficient resources to establish the social institutions needed toregulate road safety interventions. With a higher income level,abatement efforts become more substantial because more re-sources are available to invest in road safety. Therefore, road safetyregulatory institutions become more effective. This supply side ofabatement, together with the demand side mentioned above, isexpected to yield a declining relationship between road deathsand income.

Bishai et al. (2006) found that at higher income levels, an in-crease in per-capita income appears to reduce road deaths, butdoes not reduce road crashes and injuries. However, their analysisdoes not explain the underlying mechanism that drives these re-lationships. In addition, their analysis applied a fixed effect or-dinary regression, which is inappropriate to model count data.Thus, the objective of this study is to understand how economicgrowth affects non-fatal road injuries and what factors underliethis relationship.

The analysis presented in this study examines the relationshipbetween non-fatal road injuries and economic growth, spanning a47-year period. Several explanatory variables, which are related tothe fundamental development of a country, are included to ac-count for this relationship. In particular we include motorizationlevel, the urban-to-rural population ratio, the percentage of thepopulation below the age of 15 and over the age of 64, democracylevel, political stability, and adult alcohol consumption rate.

1 This is done by conditioning on the total number of non-fatal road injuriesthat occurred within each country during the study period.

2 Most of the developed countries use hospital data as the source, while indeveloping countries, it is mostly taken from police.

3 The Hausman test rejects the null hypothesis at 5% significance level.4 Some data are missing for some countries and years.

2. Methodology

The analysis of count data, such as non-fatal road injuries, oftendoes not follow the underlying assumption of normality, limited tonon-negative integer values and the distribution is highly skewed(Gardner et al., 1995; Cameron and Trivedi, 1998). Therefore, theconventional ordinary linear regression may not be appropriate toanalyze this sort of data (Long, 1997). Transformation of the datamay not yield normally distributed data. Furthermore, this maycause difficulty in interpreting the regression coefficients becausethe transformed data are not estimated on the original scale (Byerset al., 2003) and the accuracy of the estimated results is ques-tionable (Chang and Pocock, 2000).

In view of the above facts, the Poisson and the negative bino-mial regression methods are frequently used to model count data.However, the selection of the Poisson or the negative binomialregression is based on the difference between mean and variance.The Poisson regression model assumes the mean is equal to var-iance, also known as equidispersion. If this assumption is violated(overdispersion), the Poisson model will produce consistent coef-ficient estimates, but standard errors will be underestimated(Cameron and Trivedi, 1998; Winkelmann, 1997).

The log-likelihood ratio test is used to test the hypothesis thatthe variance and mean is equal, which indicates equidispersion.Rejection of the null hypothesis implies that the negative binomialis more appropriate than the Poisson regression. In the modelsestimated in this study, the log-likelihood ratio test showed thatthe null hypothesis can be rejected at 5% level of significance,implying that the negative binomial regression model is the pre-ferred model.

The present study used the design that combine longitudinal

and cross-sectional methods. The fixed effects negative binomialregression, derived by Hausman et al. (1984), is employed in thisstudy. The fixed effects model assumes that the country specificintercept is correlated with explanatory variables. The first ad-vantage of this model is that it can cancel out the dispersionparameters and account for heterogeneity in the data1. The secondadvantage of this model is the country specific intercepts are ableto take into account differences in accident data derivation due todifferent sources of accident data used2.

Alternatively, another method known as random effects modelassumes that the inverse of the over-dispersion parameter is dis-tributed as a beta distribution. The assumption applied in thismodel is that country specific effects as part of the error term.Another assumption is that the country specific effect is un-correlated with the explanatory variables, which is often un-realistic (Baltagi, 2001; Wooldridge, 2000).

The choice between fixed and random effects model is basedon the Hausman test (Baltagi, 2001). The null hypothesis for thetest is the country specific intercept is uncorrelated with otherexplanatory variables. Rejection of the null hypothesis indicatingfixed effects model is more appropriate. For the data analyzedhere, it was found that the more appropriate effect for all modelswas the fixed effect3.

An offset variable is included in the analysis to normalize theeffect of risk exposure on non-fatal road injuries. This is necessarybecause a country with a higher level of risk exposure shouldexperience more road crashes. The offset variable is specified asthe logarithm of a measure of risk exposure in the equation andcan be written as

x Elog log 1it i it itμ α β( ) = + + ( ) ( )

where itμ denotes the expected number of non-fatal road injuries,Eit represents an index of risk exposure, iα is the country specificintercept, xit is a vector of covariate which describes the char-acteristics of an observation unit i during a given time period t, β isthe model parameters. This equation can be rewritten as

E xlog log 2it it i itμ α β( ) − ( ) = + ( )

and then

E xlog / 3it it i itμ α β( ) = + ( )

The coefficient for the explanatory variables is interpreted aseffects on rates rather than a count. From Eq. (1), the expectednumber of non-fatal road injuries is given by E xexpit it itμ α β= ( + ).This means that the expected number of non-fatal road injuries isproportional to the level of risk exposure.

3. Data

The empirical analysis in this paper uses an unbalanced4 paneldataset which consists of a total of 1653 observations from 90countries (both less developed and highly developed countries) forthe period between 1963 and 2009. These countries were selectedbased on the availability of data on all explanatory variables for atleast three years. The sample is divided into two groups:- lessdeveloped and highly developed countries, according to the defi-nition of the International Monetary Fund (IMF) (2014) and the

Table 1The list of less developed countries.

No. Country Period Observations Included in Table 6

1 Albania 1992–2003 13 √2 Azerbaijan 1993–2007 12 √3 Bahrain 1987–2008 154 Bangladesh 1991–2005 10 √5 Belarus 1994–2007 76 Benin 1990–2007 87 Botswana 1968–2006 28 √8 Brazil 1975–2007 16 √9 Bulgaria 1989–2008 17 √

10 Cambodia 1992–1997 611 Cameroon 1965–2006 312 Chile 1979–2006 2113 Colombia 1985–2009 1714 Costa Rica 1986–1999 1415 Cuba 1991–1997 716 Ecuador 1990–1998 917 Egypt 1982–1994 618 El Salvador 1990–2009 919 Ethiopia 1973–2007 2320 Georgia 1993–2003 821 India 1965–2005 25 √22 Indonesia 1965–2008 32 √23 Kazakhstan 1999–2007 6 √24 Kenya 1965–2004 29 √25 Kuwait 1986–2009 11 √26 Lithuania 1993–2008 13 √27 Malawi 1964–1984 628 Malaysia 1963–2008 43 √29 Mali 1969–2008 930 Mauritius 1968–2004 2131 Mexico 1990–2004 6 √32 Moldova 1992–2007 10 √33 Mongolia 1990–1999 1034 Morocco 1969–2007 34 √35 Namibia 1991–2007 536 Nicaragua 1989–2002 837 Niger 1984–2008 738 Nigeria 1973–1993 1139 Oman 1990–2007 840 Pakistan 1972–2009 31 √41 Panama 1977–1998 1042 Peru 1969–1972 443 Philippines 1984–2000 1544 Romania 1990–1999 1045 Saudi Arab 1988–2005 946 Senegal 1967–2008 2247 South Africa 1975–2005 1848 Sri Lanka 1969–2006 24 √49 Syrian 1965–2008 13 √50 Thailand 1967–1996 3051 Togo 1973–2007 1652 Tunisia 1963–2007 3053 Turkey 1963–2007 42 √54 Uganda 1990–1993 455 Ukraine 1994–2006 10 √56 Uruguay 1990–2007 1057 Venezuela 1967–1986 1458 Vietnam 2003–2007 5 √59 Zambia 1969–1972 460 Zimbabwe 1993–1995 3

Table 2The list of highly developed countries.

No. Country Period Observations Included in Table 6

1 Australia 1980–2007 312 Austria 1963–2009 44 √3 Belgium 1963–2006 41 √4 Canada 1969–2006 32 √5 Cyprus 1981–2007 24 √6 Czech Republic 1993–2008 13 √7 Denmark 1980–2007 25 √8 Estonia 1991–2008 15 √9 Finland 1990–2008 16 √

10 France 1970–2008 30 √11 Germany 1991–2008 15 √12 Greece 1971–2008 35 √13 Hungary 1971–2008 35 √14 Ireland 1992–2006 12 √15 Israel 2003–2007 5 √16 Italy 1993–2008 11 √17 Japan 1963–2006 41 √18 Korea, Rep. 1981–2007 24 √19 Netherlands 1963–2007 37 √20 New Zealand 1963–2007 43 √21 Norway 1963–2008 43 √22 Poland 1983–2008 24 √23 Portugal 1963–2006 30 √24 Slovakia 2003–2007 5 √25 Slovenia 1991–2008 15 √26 Spain 1995–2006 9 √27 Sweden 1964–2006 31 √28 Switzerland 1963–2009 40 √29 United Kingdom 1970–2009 37 √30 United States 1978–2004 23 √

5 48% of the polity index scores in the dataset were greater than 8.6 It is defined as the number of years since the most recent regime change or

the end of transition period defined by the lack of stable political institutions.

T.H. Law / Transport Policy 42 (2015) 166–172168

Central Intelligence Agency (CIA) (2014). This will allow us to ex-amine the effect of per-capita income on non-fatal road injuriesseparately for less developed and highly developed countries.Tables 1 and 2 present the list of countries included in thisanalysis.

The data on non-fatal road injuries and motor vehicles comesfrom two international sources, the International Road Traffic andAccident Database and the International Road Federation's WorldRoad Statistics. The real per-capita Gross Domestic Product (GDP)data (USD 2005 constant price; Chain series) is taken from the

World Penn Tables version 7.0. These data sets were used as aproxy for per-capita income. We also include the square of per-capita GDP to capture the potential quadratic effect on non-fatalroad injuries (Kuznets's effect).

The first political variable used in this study is the polity index,which was published in the polity IV database (Marshall and Jaggers,2009). This index measures the degree to which a country is eitherdemocratic or autocratic. It is a composite of five elements, whichincludes competitiveness of political participation, regulation of poli-tical participation, competitiveness of executive recruitment, opennessof executive recruitment, and institutionalized constraints on execu-tive power. This index is based on a scale that ranges from �10 to 10,where 10 represents the most democratic. The polity index is ordinal,not cardinal. For example, the magnitude of a change in the polityscore from 2 to 3 is not necessarily the same as when the score in-creases from 6 to 7. Therefore, the polity index is not truly a con-tinuous variable (Keefer et al., 2011). As the polity index scores in thisstudy were heavily clustered toward the upper end of the scale5, theywere categorized into two levels:- �10 to 8 (low democracy, coded as0) and 9–10 (high democracy, coded as 1). Another political variablethat might impact the level of road injuries would be the politicalstability of a country. As a proxy for political stability, we used regimedurability drawn from the Polity IV database, which is defined as thelength of time since the most recent regime change6. Higher regimedurability indicates higher political stability.

The urban-to-rural population ratio, which was obtained fromthe World Development Indicator Database, is used to measure theurbanization level. Urbanization refers to urban areas have higherpopulation density (Sato and Yamamoto, 2005). This variable wasused to account for the effect of population density on non-fatalroad injuries in urban areas. Data on the percentage of the

4

5

6

7

8

9

10

11

12

13

5 6 7 8 9 10 11

ln(n

on-f

atal

roa

d in

juri

es p

er 1

0,00

0 po

pula

tion)

ln(GDP)

Fig. 1. The actual and the estimated non-fatal road injuries for less developedcountries.

8

9

10

11

12

7 8 9 10 11 12

ln(n

on-f

atal

roa

d in

juri

es p

er 1

0,00

0 po

pula

tion)

ln(GDP)

Fig. 2. The actual and the estimated non-fatal road injuries for highly developedcountries.

T.H. Law / Transport Policy 42 (2015) 166–172 169

population below the age of 15 and over the age of 64 were takenfrom the same database. Another variable used in this study is therecorded adult (above 15 years of age) per-capita alcohol con-sumption, taken from the World Health Organization's Global Al-cohol Database.

4. Results and discussions

Figs. 1 and 2 show the actual and the estimated non-fatal roadinjuries for both less developed and highly developed countries. Asshown in Fig. 1, for less developed countries, the non-fatal roadinjuries to per-capita income relationship is positive over most ofthe sample range. This implies that in less developed countries,increases in per-capita income lead to an increase in non-fatal roadinjuries. On the other hand, for more developed countries, non-fatalroad injuries initially increase and then decrease as income rises.This can be visualized in Fig. 2, which shows that there is an in-verted U-shaped relationship between non-fatal road injuries andper-capita income for highly developed countries. Tables 3 and 4report descriptive statistics for all variables, for less and highly de-veloped countries, respectively. As shown, the number of injuriesrose significantly in less developed countries, but declined in moredeveloped countries as income and population increased. Althoughhighly developed countries had higher total vehicles per thousandpopulation than less developed countries, less developed countrieshad higher percentages of motorcycles. Compared to less developedcountries, highly developed countries were also characterized byhigher urbanization, higher alcohol consumption rate, and greaterpolitical stability and democracy.

Table 5 lists results from the fixed effects negative binomialmodel on an unbalanced panel of less developed and highly de-veloped countries for which data are available between 1963 and2009. The dependent variable used is non-fatal road injuries,which includes slight and serious injuries as the result of roadcrashes. The log population was used as an offset variable. Thepercentage of the population below the age of 15 and over the ageof 64, the urban-to-rural population ratio, vehicles per thousandpopulation and per-capita GDP variables are interacted with thecountry group dummy7 (countries are classified as less developedand highly developed countries). These interaction terms measurethe effect of the country's fundamental development on non-fatalroad injuries in different country groups.

The likelihood ratio test is used for two purposes. First it is usedto test overall model fit and significance. Second, it examines thenull hypothesis of equidispersion. The likelihood ratio tests showthat, for all models, the null hypothesis that all coefficients (exceptthe intercept) are jointly equal to zero is rejected. The value of thelikelihood ratio statistic increases from Models A to C, which

7 The country group dummy variable is coded as 1 for highly developedcountries and 0 for less developed countries.

indicates that model fit improves when more variables are in-cluded into the model. The likelihood ratio tests also reject the nullhypothesis of equidispersion8, which suggests that the negativebinomial model is more appropriate than the Poisson model.

The results in Models A to C in Table 5 confirm a statisticallysignificant inverted U-shaped or Kuznets relationship betweenper-capita GDP and non-fatal road injuries for less developed andhighly developed countries. This is shown in Figs. 3 and 4, re-spectively. The inverted U-shaped relationship remains statisti-cally significant after controlling the other explanatory variables.This suggests that some underlying factors that explain thechanges of non-fatal road injuries and are correlated with per-capita GDP are not considered in the analysis. The results alsoindicated that for a highly developed country, the turning pointwas in the range of US$11,301 to US$21,284, while for a less de-veloped country, it was higher (US$25,474 to USD$2,039,037). Thehigh turning point for a less developed country is consistent withFig. 1, which shows a positive relationship between non-fatal roadinjuries and per-capita GDP over most of the sample range. Weperformed additional analysis using data between 2003 and 2004.Tables 1 and 2 present the list of countries included in this ana-lysis. The result in Table 6 is consistent with that in Model A inTable 5, indicating that the result is not a mere consequence of theimprovements in road accident data quality as a country develops.

Two political variables, polity and durability, were used to ex-amine the effect of the quality of regulatory institutions on non-fatal road injuries. The negative and statistically significant coeffi-cient for polity and durability in Model C shows that greater levelsof democracy and political stability are associated with lower non-fatal road injuries. Damania et al. (2004) indicated that an increasein political stability would lead to a more judicial efficiency in lawenforcement. In more politically stable countries, judicial institu-tions are more effective in enforcing road safety regulations andpolicies. Moreover, Payne (1995) revealed that citizens of democ-racies had more freedom to express their concerns and demandsabout environmental issues and put pressure on policymakers torespond positively to such demands. These results are consistentwith Law et al.'s (2013) findings that countries with a higher level ofdemocracy and political stability were more likely to implementroad safety legislations with greater judicial efficiency. Conse-quently, this would reduce the number of non-fatal road injuries.

The estimated results for the percentage of the populationbelow the age of 15 and over the age of 64 indicated that moreyoung and elderly people are associated with increased non-fatalroad injuries in less developed countries. Several studies reportedthat the population at these age groups are more vulnerablephysically and mentally and are more likely to suffer road injuries(Peden et al., 2004; Evans, 1998, 2000; Williams and Carsten, 1989;

8 Equidispersion indicates that the dispersion parameter is zero.

Table 3Descriptive statistics for less developed countries.

Variable Between 1963 and 1989 Between 1990 and 2009Obs. Mean Std. dev. Obs. Mean Std. dev.

The number of Injuries 397 19,980.8 29,090.7 470 29,885.1 58,504.7Population (thousands of people) 397 52,636.1 123,418.5 470 55,929.1 146,914.1Real Gross Domestic Product per-capita (US$2005 constant prices) 397 3247.5 3846.4 470 6135.7 6846.1Percentage of the population below 15 and above 64 years of age 397 45.7 4.15 470 38.8 5.82Total vehicles per thousand population 397 34.9 50.15 470 95.5 102.95Percentage of motorcycles 397 0.49 0.65 470 0.56 0.62Urban-to-rural population ratio 397 1.33 4.83 470 2.99 8.11Adult (above 15 years of age) alcohol consumption(liters per population) 397 2.58 2.83 470 3.95 4.09Political stability 397 12.55 13.75 470 18.78 18.71Democracy, 10 (strongly democratic) and �10 (strongly autocratic) 397 �1.07 14a 470 3.15 12a

a The interquartile range was used instead of standard deviation because the polity index is ordinal, not cardinal.

Table 5Kuznets relationship for non-fatal road injuries for the period of 1963–2009.

Variable Model A Model B Model C

Coef. Coef. Coef.

ln(RGDPCH) LD 1.646nn 3.606nn 3.596nn

ln(RGDPCH) 2LD �0.056nn �0.179nn �0.177nn

ln(RGDPCH) HD 2.295nn 4.902nn 4.918nn

ln(RGDPCH)2HD �0.123nn �0.254nn �0.252nn

VEHP LD 0.0024nn 0.0025nn

VEHP HD �0.0004nn

�0.0004nn

MCV 0.024nn 0.024nn

ln(AGE) LD 1.828nn 1.783nn

ln(AGE) HD 0.394nn 0.301nn

URBAN LD 0.013n 0.013n

URBAN HD �0.007n �0.004n

ALC 0.073nn 0.071nn

POL (1 if POLITY 48; 0 otherwise) �0.068nn

DURABLE �0.002nn

Constant �18.053nn

�32.904nn

�32.747nn Groups 90 9090N 1653 1653 1653Log likelihood �16,666 �16,495 �16,490Likelihood ratio test 1113nn 1453nn 1464nn

Turning point (USD) LD 2,039,037 23,984 26,076Turning point (USD) HD 11,301 15,285 17,080

LD – Less developed countries, HD – Highly developed countries.n Significant at 5%.nn Significant at 1%.

T.H. Law / Transport Policy 42 (2015) 166–172170

Li et al., 2003). However, this effect is smaller for highly developedcountries. This could be attributed to the fact that young and el-derly people in more developed countries are more likely totransport in a safer mode.

The urbanization variable showed a statistically significantdifference between the two country groups. The estimated resultssuggest that an increase in the urban-to-rural population ratio canlead to an increase in non-fatal road injuries for less developedcountries but a decrease in non-fatal road injuries for more de-veloped countries. A plausible explanation is that proper urbaninfrastructure and public transport services in highly developedcountries may reduce conflicts among vulnerable road users (suchas pedestrians, motorcyclists and bicyclists) and other motor ve-hicle users, thus decreasing the number of non-fatal road injuries.

A similar result was found for vehicles per thousand popula-tion. The results indicated that more vehicles per thousand po-pulation would lead to an increase in non-fatal road injuries in lessdeveloped countries. The opposite is true in highly developedcountries. We interpret this result for increases in road crash riskexposure levels as the result of increases in the degree of mobility.Nevertheless, in highly developed countries, the road crash ex-posure level reduces due to the use of safer vehicles and betterroad infrastructure. A variable for the percentage of motorcycleswas used to represent the relative degree of motorcycle use amongdifferent countries and over time. This variable is positively andsignificantly associated with non-fatal road injuries, confirmingprevious findings9. This suggests that a higher percentage of mo-torcycles increases the number of non-fatal road injuries. Theempirical model estimates that for a 1% increase in motorcycles,

Table 4Descriptive statistics for highly developed countries.

Variable Between 1963 and 1989 Between 1990 and 2009Obs. Mean Std. dev. Obs. Mean Std. dev.

The number of Injuries 388 225,880.4 570,993.9 398 207,760.9 540,195.9Population (thousands of people) 388 29,373.2 46,829.9 398 30,606.5 49,839.04Real Gross Domestic Product per-capita (US$2005 constant prices) 388 19,113.4 5911.8 398 25,379.9 8856.05Percentage of the population below 15 and above 64 years of age 388 35.21 2.14 398 32.76 1.89Total vehicles per thousand population 388 298.63 167.54 398 432.17 159.29Percentage of motorcycles 388 0.27 0.65 398 0.11 0.12Urban-to-rural population ratio 388 3.94 4.47 398 4.11 5.62Adult (above 15 years of age) alcohol consumption(liters per population) 388 10.85 3.71 398 10.35 2.64Political stability 388 51.68 43.03 398 53.15 46.82Democracy, 10 (strongly democratic) and �10 (strongly autocratic) 388 8.25 0a 398 9.69 0a

a The interquartile range was used instead of standard deviation because the polity index is ordinal, not cardinal.

9 Motorcyclists are more vulnerable to injury than other motor vehicle driverson the road (Radin Umar et al., 1995; Preusser et al., 1995).

non-fatal road injuries per thousand population increase by 2.4%.The result presented in Model C reveals a positive association

between alcohol consumption rate and non-fatal road injuries.This implies that an increase in the alcohol consumption rate can

0

2

4

6

8

10

12

0 5 10 15 20 25 30

ln(n

on-f

atal

roa

d in

juri

es p

er 1

0,00

0 po

pula

tion)

ln(GDP)

Fig. 3. The effect of per-capita income on non-fatal road injuries for less developedcountries based on Model A (Table 5).

0

2

4

6

8

10

0 2 4 6 8 10 12 14 16 18 20

ln(n

on-f

atal

roa

d in

juri

es p

er 1

0,00

0 po

pula

tion)

ln(GDP)

Fig. 4. The effect of per-capita income on road injuries for highly developedcountries based on Model A (Table 5).

Table 6Kuznets relationship for non-fatal road injuries for the period of 2003–2004.

Variable Model ACoef.

ln(RGDPCH) LD 6.919n

ln(RGDPCH) 2LD �0.327n

ln(RGDPCH) HD 8.548nn

ln(RGDPCH)2HD �0.468nn

Constant �41.543nn

Groups 51N 102Log likelihood �448.12Likelihood ratio test 82.27nn

Turning point (USD) LD 38,294Turning point (USD) HD 9174

LD – Less developed countries, HD – Highly developed countries.n Significant at 5%.nn Significant at 1%.

T.H. Law / Transport Policy 42 (2015) 166–172 171

lead to a significant increase in non-fatal road injuries. The resultsindicated that a one-liter per-capita increase in alcohol con-sumption led to a 7.5% increase in non-fatal injuries per thousandpopulation. Previous studies found that driver skills related toperception, information processing and judgement are sig-nificantly impaired even at lower blood alcohol concentration le-vels (Moskowitz et al., 2000; Moskowitz and Robinson, 1988;Moskowitz and Burns, 1990).

10 Separate data on slight and serious road injuries are not available for most ofthe less developed countries.

11 According to the World Health Organization data, less than 35% of low- andmiddle-income countries had road safety policies in place to protect vulnerableroad users, such as pedestrians and cyclists.

5. Conclusions

This paper investigates the empirical relationship betweennon-fatal road injuries and per-capita GDP using an internationalpanel data set over the period of 1963–2009. The empirical resultssuggested that non-fatal road injuries follow an invertedU-shaped, or Kuznets curve, relationship with per-capita GDP forless developed and highly developed countries.

Several limitations of the current study should be noted. First,the dependent variable, non-fatal road injuries, which consists ofslight and serious road injuries. Thus, an estimation of the effect ofexplanatory variables on slight and serious road injuries separatelyis not applicable10. Second, the political variables, democracy andpolitical stability indexes, are used as proxy variables to determinethe likelihood of enacting and enforcing road safety legislation,without taking the effect of enforcement resources into con-sideration. Consequently, all else being equal, two countries withan equal score of democracy and political stability will have anequal reduction in non-fatal road injuries. However, in reality, it isexpected that in a country with a higher quality of political in-stitutions, increases in enforcement resources should lead to agreater reduction in non-fatal road injuries. Lastly, previous stu-dies (Law et al., 2009, 2011; Noland, 2003) found that improve-ment in the level of medical care and technology is associated withreductions in the number of road deaths. Consequently, one wouldexpect that many road deaths could be prevented and result innon-fatal road injuries as medical care and technology advances.However, the effect of such advances could not be observed in theanalysis for two reasons. First, increases in non-fatal road injuriesmay be due to greater accessibility and use of medical care amongthose who are moderately injured, as medical care and technologycontinue to improve. Second, even if there is a shift from fatal tonon-fatal road injuries, it would be a minor percentage of the totalinjuries and might not be detectable statistically.

This study examines the empirical Kuznets curve relationshipbetween non-fatal road injuries and per-capita income. Althoughthe evidence implies the Kuznets curve effect on non-fatal roadinjuries, we cannot assume that today's less developed countrieswill inevitably follow a similar path. The results suggest that im-provements in road infrastructure, the quality of regulatory in-stitutions (as represented by political stability and democraticinstitutions), and increase in the use of safer transport modes willhelp reduce non-fatal road injuries. However, less developedcountries generally suffer from lower judicial efficiency in lawenforcement (WHO, 2013)11 and higher levels of motorcycleownership, which tend to persist over time. In view of this, it willbe beneficial to improve the quality of regulatory institutions andpromote the use of safer transport modes, especially in less de-veloped countries, to achieve a sustainable improvement in roadsafety. Economic growth may be a prerequisite, but it alone doesnot guarantee that positive change will happen.

Acknowledgment

This research was funded by Universiti Putra Malaysia's Re-search University Grant Scheme (Grant no. 05-05-10-1108RU). Thefunding source played absolutely no role in the study design norany collection, analysis, or interpretation of the data.

References

Anbarci, N., Escaleras, M., Register, C., 2006. Traffic fatalities and public sectorcorruption. KYKLOS 59, 327–344.

Anbarci, N., Escaleras, M., Register, C.A., 2009. Traffic fatalities: does income in-equality create an externality? Can. J. Econ. 42, 244–266.

Baltagi, B., 2001. Econometric Analysis of Panel Data. John Wiley and Sons, WestSussex.

T.H. Law / Transport Policy 42 (2015) 166–172172

Beeck, E.F.V., GJJM, Borsboom, Mackenbach, J.P., 2000. Economic development andtraffic accident mortality in the industrialized world, 1962–1990. Int. J. Epide-miol. 29, 503–509.

Bhalla, K., Ezzati, M., Mahal, A., Salomon, J., Reich, M., 2007. A risk-based method formodelling traffic fatalities. Risk Anal. 27, 125–136.

Bishai, D., Quresh, A., James, P., Ghaffar, A., 2006. National road casualties andeconomic development. Health Econ. 15, 65–81.

Byers, A.L., Allore, H., Gill, T.M., Peduzzi, P.N., 2003. Application of negative binomialmodelling for discrete outcomes: a case study in ageing research. J. Clin. Epi-demiol. 56, 559–564.

Cameron, A.C., Trivedi, P.K., 1998. Regression Analysis of Count Data. CambridgeUniversity Press, Cambridge.

Chang, B.H., Pocock, S., 2000. Analysing data with clumping at zero: an exampledemonstration. Biometrics 53, 1036–1043.

CIA (Central Intelligence Agency), 2014. The World Factbook. URL: ⟨https://www.cia.gov/library/publications/the-world-factbook/appendix/appendix-b.html⟩.

Damania, R., Fredriksson, P.G., Mani, M., 2004. The persistence of corruption andregulatory compliance failures: theory and evidence. Public Choice 121,363–390.

Dargay, J., Gately, D., 1999. Income's effect on car and vehicle ownership. Transp.Res. Part A 33, 101–138.

Evans, L., 1998. Older driver involvement in fatal and severe traffic crashes. J.Gerontol.: Soc. Sci. 43, S186–S193.

Evans, L., 2000. Risk older drivers face themselves and threats they pose to otherroad users. Int. J. Epidemiol. 29, 315–322.

Gardner, W., Mulvey, E.P., Shaw, E.C., 1995. Regression analyses of counts and rates:Poisson, overdispersed Poisson, and negative binomial models. Psychol. Bull.118, 392–404.

Hausman, J., Hall, B.H., Griliches, Z., 1984. Econometric models for count data withan application to the patents—R & D relationship. Econometrica 52, 909–938.

IMF (International Monetary Fund), 2014. World Economic Outlook – Legacies,Clouds Uncertainties. URL: http://www.imf.org/external/pubs/ft/weo/2014/02/pdf/text.pdf.

Keefer, P., Neumayer, E., Plümper, T., 2011. Earthquake propensity and the politics ofmortality prevention. World Dev. 39, 1530–1541.

Kopits, E., Cropper, M., 2005. Traffic fatalities and economic growth. Accid. Anal.Prev. 37, 169–178.

Kuznets, S., 1955. Economic growth and incomes inequality. Am. Econ. Rev. 1, 1–28.Law, T.H., Noland, R.B., Evans, A.W., 2009. Factors associated with relationship be-

tween motorcycle deaths and economic growth. Accid. Anal. Prev. 41, 234–240.Law, T.H., Noland, R.B., Evans, A.W., 2011. The sources of the Kuznets relationship

between road fatalities and economic growth. J. Transp. Geogr. 19, 355–365.Law, T.H., Noland, R.B., Evans, A.W., 2013. Factors associated with the enactment of

safety belt and motorcycle helmet laws. Risk Anal. 33, 1367–1378.Li, G., Braver, E.R., Chen, L.H., 2003. Fragility versus excessive crash involvement as

determinants of high death rates per car-mile of travel among older drivers.

Accid. Anal. Prev. 35, 227–235.Long, J.S., 1997. Regression Models for Categorical and Limited Dependent Variable.

Sage, Thousand Oaks, CA.Marshall, M.G., Jaggers, K., 2009. Polity IV Project: Dataset Users' Manual.Moniruzzaman, S., Andersson, R., 2008. Economic development as a determinant of

injury mortality – a longitudinal approach. Soc. Sci. Med. 66, 1699–1708.Moskowitz, H., Robinson, C.D., 1988. Effects of Low Doses of Alcohol on Driving

Related Skills: A Review of the Evidence. National Highway Traffic Safety Ad-ministration Report No. DOT HS 807 280. Department of Transportation, Wa-shington, DC.

Moskowitz, H., Burns, M., 1990. Effects of alcohol on driving performance. AlcoholHealth Res. World 14, 12–14.

Moskowitz, H., Burns, M., Fiorentino, D., Smiley, A., Zador, P., 2000. Driver Char-acteristics and Impairment at Various BACs. Southern California Research In-stitute, National Highway Traffic Safety Administration, Washington, DC.

Noland, R.B., 2003. Medical treatment and traffic fatality reductions in in-dustrialized countries. Accid. Anal. Prev. 35, 877–883.

Paulozzi, L.J., Ryan, G.W., Espitia-Hardeman, V.E., Xi, Y., 2007. Economic devel-opment's effect on road transport-related mortality among different types ofroad users: a cross-sectional international study. Accid. Anal. Prev. 39, 606–617.

Payne, R.A., 1995. Freedom Environ. J. Democr. 6, 41–55.Peden, M., Scurfield, R., Sleet, D., Mohan, D., Hyder, A.A., Jarawan, E., Mathers, C.,

2004. The World Report on Road Traffic Injury Prevention. World Health Or-ganization, Geneva, Switzerland.

Preusser, D.F., Williams, A.F., Ulmer, R.G., 1995. Analysis of fatal motorcycle crashes:crash typing. Accid. Anal. Prev. 27, 845–851.

Radin Umar, R.S., Mackay, G.M., Hills, B.L., 1995. Preliminary analysis of motorcycleaccidents: short-term impacts of the running headlights campaign and reg-ulation. J. Traffic Med. 23, 17–28.

Sato, Y., Yamamoto, K., 2005. Population concentration, urbanization, and demo-graphic transition. J. Urban Econ. 58, 45–61.

Söderlund, N., Zwi, A.B., 1995. Traffic-related mortality in industrialized and lessdeveloped countries. Bull. World Health Organ. 73, 175–182.

Williams, A.F., Carsten, O., 1989. Driver age and crash involvement. Am. J. PublicHealth 79, 326–327.

Winkelmann, R., 1997. Econometric Analysis of Count Data. Springer Verlag, NewYork.

Wintemute, G.J., 1985. Is motor vehicle-related mortality a disease of development?Accid. Anal. Prev. 17, 223–237.

World Health Organization, 2013. Global Status Report on Road Safety: SupportingA Decade of Action, Geneva.

Wooldridge, J.M., 2000. Introductory Econometrics: A Modern Approach. South-Western College Publishing, New York.