- Email: [email protected]
Factors associated with the relationship between non-fatal road injuries and economic growth
Transport Policy 42 (2015) 166–172
Contents lists available at ScienceDirect
Transport Policy journal homepage: www.elsevier.com/locate/tranpol
Factors associated with the relationship between non-fatal road injuries and economic growth Teik Hua Law n Road Safety Research Centre, Faculty of Engineering, Universiti Putra Malaysia, 43400 Serdang, Selangor, Malaysia
art ic l e i nf o
a b s t r a c t
Article history: Received 9 October 2014 Received in revised form 29 May 2015 Accepted 5 June 2015
This study reports the results of an empirical analysis of the Kuznets curve relationship between nonfatal road injuries and per-capita income. This relationship indicates that the number of road deaths increases with increasing per-capita income at lower income levels, but decreases once it has exceeded a threshold level. We apply a fixed effects negative binomial regression analysis on a panel of 90 countries over the period of 1963–2009. Results indicated evidence of an inverted U-shaped relationship between economic 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 developed countries. The evidence presented in this study suggests that improvements in road infrastructure, the quality of regulatory institutions, and increase in the use of safer transport modes will help reduce non-fatal road injuries. & 2015 Elsevier Ltd. All rights reserved.
Keywords: Kuznets relationship Non-fatal road injuries Democracy Political stability
1. Introduction The relevance of the relationship between economic growth and road safety has long been recognized. However, it has been the impact of economic growth on road deaths that has attracted the most attention (Wintemute, 1985; Söderlund and Zwi, 1995; Beeck et al., 2000; Kopits and Cropper, 2005). Recently, numerous studies have shown that there is an inverted U-shaped relationship 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 decrease once the number has exceeded a threshold level. This relationship is also known as the Kuznets curve relationship, which existed between income inequality and per-capita income (Kuznets, 1955). It is important to understand how economic growth affects road safety because road safety concerns could lead to a primary road safety policy agenda. This relationship has been explained by several studies to be the result of changes in the fundamental development of a country. This includes changes in variations in rates among vulnerable non-motorized road users (Paulozzi et al., 2007), advances in medical services (Law et al., 2009, 2011; Noland, 2003), the quality of political institutions (Anbarci et al., 2006; Law et al., 2011), income equality (Anbarci et al., 2009) and road safety policies (Law et al., 2009). n
Fax: þ 60 386567129. E-mail address: [email protected]
http://dx.doi.org/10.1016/j.tranpol.2015.06.004 0967-070X/& 2015 Elsevier Ltd. All rights reserved.
The Kuznets curve hypothesis for road safety can be supported by three theoretical explanations: the scale of economic activities, changes in vehicle composition, and a surge in demand for better road safety as per-capita income rises (Law et al., 2009). The Kuznets curve hypothesis contends that the number of road deaths increases initially as a country's economy develops, owing to the fact that the growth of a country's economy is accompanied by a corresponding surge in demand for transportation services (Dargay and Gately, 1999). Several previous studies have found that one of the contributing factors to the increase in road crashes and injuries is the growing number of vehicles per-capita (Bishai et al., 2006; Kopits and Cropper, 2005). The composition effect depicts the change in vehicle composition as a country's economy grows. This involves a change from low-threat and high-risk transport modes (such as pedestrians and 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 an increase in the number of motorized vehicles can pose an increasing threat to the predominantly more vulnerable transport modes. Nevertheless, at higher levels of economic development, when a majority of the commuting population are less vulnerable motor vehicle occupants, there is an increase in the number of motor vehicles causing a reduction in the total road death risk. This indicates an inverted U-shaped relationship between road deaths and income. The abatement effect is the result of road death risk alleviation
T.H. Law / Transport Policy 42 (2015) 166–172
measures, which reflects both supply side and demand influences. On the demand side, at low income levels, people are less capable to invest in road safety, even though there is a demand for it, resulting in low abatement. Nonetheless, as the level of income rises, people pay greater attention to road safety and they can afford higher-cost road safety enhancement measures, enabling more commuters to switch to lower-risk modes of transport. On the supply side, at low levels of income, society is unable to afford sufficient resources to establish the social institutions needed to regulate road safety interventions. With a higher income level, abatement efforts become more substantial because more resources are available to invest in road safety. Therefore, road safety regulatory institutions become more effective. This supply side of abatement, together with the demand side mentioned above, is expected to yield a declining relationship between road deaths and income. Bishai et al. (2006) found that at higher income levels, an increase in per-capita income appears to reduce road deaths, but does not reduce road crashes and injuries. However, their analysis does not explain the underlying mechanism that drives these relationships. In addition, their analysis applied a fixed effect ordinary regression, which is inappropriate to model count data. Thus, the objective of this study is to understand how economic growth affects non-fatal road injuries and what factors underlie this relationship. The analysis presented in this study examines the relationship between non-fatal road injuries and economic growth, spanning a 47-year period. Several explanatory variables, which are related to the fundamental development of a country, are included to account for this relationship. In particular we include motorization level, the urban-to-rural population ratio, the percentage of the population below the age of 15 and over the age of 64, democracy level, political stability, and adult alcohol consumption rate.
2. Methodology The analysis of count data, such as non-fatal road injuries, often does not follow the underlying assumption of normality, limited to non-negative integer values and the distribution is highly skewed (Gardner et al., 1995; Cameron and Trivedi, 1998). Therefore, the conventional ordinary linear regression may not be appropriate to analyze this sort of data (Long, 1997). Transformation of the data may not yield normally distributed data. Furthermore, this may cause difficulty in interpreting the regression coefficients because the transformed data are not estimated on the original scale (Byers et al., 2003) and the accuracy of the estimated results is questionable (Chang and Pocock, 2000). In view of the above facts, the Poisson and the negative binomial regression methods are frequently used to model count data. However, the selection of the Poisson or the negative binomial regression is based on the difference between mean and variance. The Poisson regression model assumes the mean is equal to variance, also known as equidispersion. If this assumption is violated (overdispersion), the Poisson model will produce consistent coefficient estimates, but standard errors will be underestimated (Cameron and Trivedi, 1998; Winkelmann, 1997). The log-likelihood ratio test is used to test the hypothesis that the variance and mean is equal, which indicates equidispersion. Rejection of the null hypothesis implies that the negative binomial is more appropriate than the Poisson regression. In the models estimated in this study, the log-likelihood ratio test showed that the null hypothesis can be rejected at 5% level of significance, implying that the negative binomial regression model is the preferred model. The present study used the design that combine longitudinal
167
and cross-sectional methods. The fixed effects negative binomial regression, derived by Hausman et al. (1984), is employed in this study. The fixed effects model assumes that the country specific intercept is correlated with explanatory variables. The first advantage of this model is that it can cancel out the dispersion parameters and account for heterogeneity in the data1. The second advantage of this model is the country specific intercepts are able to take into account differences in accident data derivation due to different sources of accident data used2. Alternatively, another method known as random effects model assumes that the inverse of the over-dispersion parameter is distributed as a beta distribution. The assumption applied in this model is that country specific effects as part of the error term. Another assumption is that the country specific effect is uncorrelated with the explanatory variables, which is often unrealistic (Baltagi, 2001; Wooldridge, 2000). The choice between fixed and random effects model is based on the Hausman test (Baltagi, 2001). The null hypothesis for the test is the country specific intercept is uncorrelated with other explanatory variables. Rejection of the null hypothesis indicating fixed effects model is more appropriate. For the data analyzed here, it was found that the more appropriate effect for all models was the fixed effect3. An offset variable is included in the analysis to normalize the effect of risk exposure on non-fatal road injuries. This is necessary because a country with a higher level of risk exposure should experience more road crashes. The offset variable is specified as the logarithm of a measure of risk exposure in the equation and can be written as
log(μit ) = αi + βxit + log(Eit )
(1)
where μit denotes the expected number of non-fatal road injuries, Eit represents an index of risk exposure, αi is the country specific intercept, x it is a vector of covariate which describes the characteristics of an observation unit i during a given time period t, β is the model parameters. This equation can be rewritten as
log(μit ) − log(Eit ) = αi + βxit
(2)
and then
log(μit /Eit ) = αi + βxit
(3)
The coefficient for the explanatory variables is interpreted as effects on rates rather than a count. From Eq. (1), the expected number of non-fatal road injuries is given by μit = Eit exp(α + βx it ). This means that the expected number of non-fatal road injuries is proportional to the level of risk exposure.
3. Data The empirical analysis in this paper uses an unbalanced4 panel dataset which consists of a total of 1653 observations from 90 countries (both less developed and highly developed countries) for the period between 1963 and 2009. These countries were selected based on the availability of data on all explanatory variables for at least three years. The sample is divided into two groups:- less developed and highly developed countries, according to the definition of the International Monetary Fund (IMF) (2014) and the 1 This is done by conditioning on the total number of non-fatal road injuries that occurred within each country during the study period. 2 Most of the developed countries use hospital data as the source, while in developing 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.
168
T.H. Law / Transport Policy 42 (2015) 166–172
Table 1 The list of less developed countries.
Table 2 The list of highly developed countries.
No.
Country
Period
Observations
Included in Table 6
No.
Country
Period
Observations
Included in Table 6
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60
Albania Azerbaijan Bahrain Bangladesh Belarus Benin Botswana Brazil Bulgaria Cambodia Cameroon Chile Colombia Costa Rica Cuba Ecuador Egypt El Salvador Ethiopia Georgia India Indonesia Kazakhstan Kenya Kuwait Lithuania Malawi Malaysia Mali Mauritius Mexico Moldova Mongolia Morocco Namibia Nicaragua Niger Nigeria Oman Pakistan Panama Peru Philippines Romania Saudi Arab Senegal South Africa Sri Lanka Syrian Thailand Togo Tunisia Turkey Uganda Ukraine Uruguay Venezuela Vietnam Zambia Zimbabwe
1992–2003 1993–2007 1987–2008 1991–2005 1994–2007 1990–2007 1968–2006 1975–2007 1989–2008 1992–1997 1965–2006 1979–2006 1985–2009 1986–1999 1991–1997 1990–1998 1982–1994 1990–2009 1973–2007 1993–2003 1965–2005 1965–2008 1999–2007 1965–2004 1986–2009 1993–2008 1964–1984 1963–2008 1969–2008 1968–2004 1990–2004 1992–2007 1990–1999 1969–2007 1991–2007 1989–2002 1984–2008 1973–1993 1990–2007 1972–2009 1977–1998 1969–1972 1984–2000 1990–1999 1988–2005 1967–2008 1975–2005 1969–2006 1965–2008 1967–1996 1973–2007 1963–2007 1963–2007 1990–1993 1994–2006 1990–2007 1967–1986 2003–2007 1969–1972 1993–1995
13 12 15 10 7 8 28 16 17 6 3 21 17 14 7 9 6 9 23 8 25 32 6 29 11 13 6 43 9 21 6 10 10 34 5 8 7 11 8 31 10 4 15 10 9 22 18 24 13 30 16 30 42 4 10 10 14 5 4 3
√ √
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30
Australia Austria Belgium Canada Cyprus Czech Republic Denmark Estonia Finland France Germany Greece Hungary Ireland Israel Italy Japan Korea, Rep. Netherlands New Zealand Norway Poland Portugal Slovakia Slovenia Spain Sweden Switzerland United Kingdom United States
1980–2007 1963–2009 1963–2006 1969–2006 1981–2007 1993–2008 1980–2007 1991–2008 1990–2008 1970–2008 1991–2008 1971–2008 1971–2008 1992–2006 2003–2007 1993–2008 1963–2006 1981–2007 1963–2007 1963–2007 1963–2008 1983–2008 1963–2006 2003–2007 1991–2008 1995–2006 1964–2006 1963–2009 1970–2009 1978–2004
31 44 41 32 24 13 25 15 16 30 15 35 35 12 5 11 41 24 37 43 43 24 30 5 15 9 31 40 37 23
√ √ √ √ √ √ √ √ √ √ √ √ √ √ √ √ √ √ √ √ √ √ √ √ √ √ √ √ √
√
√ √ √
√ √ √ √ √ √ √
√ √ √
√
√ √
√ √
√
Central Intelligence Agency (CIA) (2014). This will allow us to examine the effect of per-capita income on non-fatal road injuries separately for less developed and highly developed countries. Tables 1 and 2 present the list of countries included in this analysis. The data on non-fatal road injuries and motor vehicles comes from two international sources, the International Road Traffic and Accident Database and the International Road Federation's World Road 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 a proxy for per-capita income. We also include the square of percapita GDP to capture the potential quadratic effect on non-fatal road 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 either democratic or autocratic. It is a composite of five elements, which includes competitiveness of political participation, regulation of political participation, competitiveness of executive recruitment, openness of executive recruitment, and institutionalized constraints on executive 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 polity score from 2 to 3 is not necessarily the same as when the score increases from 6 to 7. Therefore, the polity index is not truly a continuous variable (Keefer et al., 2011). As the polity index scores in this study were heavily clustered toward the upper end of the scale5, they were categorized into two levels:- 10 to 8 (low democracy, coded as 0) and 9–10 (high democracy, coded as 1). Another political variable that might impact the level of road injuries would be the political stability of a country. As a proxy for political stability, we used regime durability drawn from the Polity IV database, which is defined as the length of time since the most recent regime change6. Higher regime durability indicates higher political stability. The urban-to-rural population ratio, which was obtained from the World Development Indicator Database, is used to measure the urbanization level. Urbanization refers to urban areas have higher population density (Sato and Yamamoto, 2005). This variable was used to account for the effect of population density on non-fatal road injuries in urban areas. Data on the percentage of the 5
48% of the polity index scores in the dataset were greater than 8. 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. 6
T.H. Law / Transport Policy 42 (2015) 166–172 12
12 11 10 9 8 7 6 5 4
5
6
7
8 ln(GDP)
9
10
11
ln(non-fatal road injuries per 10,000 population)
ln(non-fatal road injuries per 10,000 population)
13
169
4. Results and discussions Figs. 1 and 2 show the actual and the estimated non-fatal road injuries for both less developed and highly developed countries. As shown in Fig. 1, for less developed countries, the non-fatal road injuries to per-capita income relationship is positive over most of the sample range. This implies that in less developed countries, increases in per-capita income lead to an increase in non-fatal road injuries. On the other hand, for more developed countries, non-fatal road injuries initially increase and then decrease as income rises. This can be visualized in Fig. 2, which shows that there is an inverted U-shaped relationship between non-fatal road injuries and per-capita income for highly developed countries. Tables 3 and 4 report descriptive statistics for all variables, for less and highly developed countries, respectively. As shown, the number of injuries rose significantly in less developed countries, but declined in more developed countries as income and population increased. Although highly developed countries had higher total vehicles per thousand population than less developed countries, less developed countries had higher percentages of motorcycles. Compared to less developed countries, highly developed countries were also characterized by higher urbanization, higher alcohol consumption rate, and greater political stability and democracy. Table 5 lists results from the fixed effects negative binomial model on an unbalanced panel of less developed and highly developed countries for which data are available between 1963 and 2009. The dependent variable used is non-fatal road injuries, which includes slight and serious injuries as the result of road crashes. The log population was used as an offset variable. The percentage of the population below the age of 15 and over the age of 64, the urban-to-rural population ratio, vehicles per thousand population and per-capita GDP variables are interacted with the country group dummy7 (countries are classified as less developed and highly developed countries). These interaction terms measure the effect of the country's fundamental development on non-fatal road injuries in different country groups. The likelihood ratio test is used for two purposes. First it is used to test overall model fit and significance. Second, it examines the null hypothesis of equidispersion. The likelihood ratio tests show that, for all models, the null hypothesis that all coefficients (except the intercept) are jointly equal to zero is rejected. The value of the likelihood ratio statistic increases from Models A to C, which 7 The country group dummy variable is coded as 1 for highly developed countries and 0 for less developed countries.
10
9
8
Fig. 1. The actual and the estimated non-fatal road injuries for less developed countries.
population below the age of 15 and over the age of 64 were taken from the same database. Another variable used in this study is the recorded adult (above 15 years of age) per-capita alcohol consumption, taken from the World Health Organization's Global Alcohol Database.
11
7
8
9
10
11
12
ln(GDP)
Fig. 2. The actual and the estimated non-fatal road injuries for highly developed countries.
indicates that model fit improves when more variables are included into the model. The likelihood ratio tests also reject the null hypothesis of equidispersion8, which suggests that the negative binomial model is more appropriate than the Poisson model. The results in Models A to C in Table 5 confirm a statistically significant inverted U-shaped or Kuznets relationship between per-capita GDP and non-fatal road injuries for less developed and highly developed countries. This is shown in Figs. 3 and 4, respectively. The inverted U-shaped relationship remains statistically significant after controlling the other explanatory variables. This suggests that some underlying factors that explain the changes of non-fatal road injuries and are correlated with percapita GDP are not considered in the analysis. The results also indicated that for a highly developed country, the turning point was in the range of US$11,301 to US$21,284, while for a less developed country, it was higher (US$25,474 to USD$2,039,037). The high turning point for a less developed country is consistent with Fig. 1, which shows a positive relationship between non-fatal road injuries and per-capita GDP over most of the sample range. We performed additional analysis using data between 2003 and 2004. Tables 1 and 2 present the list of countries included in this analysis. The result in Table 6 is consistent with that in Model A in Table 5, indicating that the result is not a mere consequence of the improvements in road accident data quality as a country develops. Two political variables, polity and durability, were used to examine the effect of the quality of regulatory institutions on nonfatal road injuries. The negative and statistically significant coefficient for polity and durability in Model C shows that greater levels of democracy and political stability are associated with lower nonfatal road injuries. Damania et al. (2004) indicated that an increase in political stability would lead to a more judicial efficiency in law enforcement. In more politically stable countries, judicial institutions are more effective in enforcing road safety regulations and policies. Moreover, Payne (1995) revealed that citizens of democracies had more freedom to express their concerns and demands about environmental issues and put pressure on policymakers to respond positively to such demands. These results are consistent with Law et al.'s (2013) findings that countries with a higher level of democracy and political stability were more likely to implement road safety legislations with greater judicial efficiency. Consequently, this would reduce the number of non-fatal road injuries. The estimated results for the percentage of the population below the age of 15 and over the age of 64 indicated that more young and elderly people are associated with increased non-fatal road injuries in less developed countries. Several studies reported that the population at these age groups are more vulnerable physically 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.
170
T.H. Law / Transport Policy 42 (2015) 166–172
Table 3 Descriptive statistics for less developed countries. Variable
The number of Injuries Population (thousands of people) Real Gross Domestic Product per-capita (US$2005 constant prices) Percentage of the population below 15 and above 64 years of age Total vehicles per thousand population Percentage of motorcycles Urban-to-rural population ratio Adult (above 15 years of age) alcohol consumption(liters per population) Political stability Democracy, 10 (strongly democratic) and 10 (strongly autocratic) a
Between 1963 and 1989 Obs. Mean
Std. dev.
Between 1990 and 2009 Obs. Mean
Std. dev.
397 397 397 397 397 397 397 397 397 397
29,090.7 123,418.5 3846.4 4.15 50.15 0.65 4.83 2.83 13.75 14a
470 470 470 470 470 470 470 470 470 470
58,504.7 146,914.1 6846.1 5.82 102.95 0.62 8.11 4.09 18.71 12a
19,980.8 52,636.1 3247.5 45.7 34.9 0.49 1.33 2.58 12.55 1.07
29,885.1 55,929.1 6135.7 38.8 95.5 0.56 2.99 3.95 18.78 3.15
The interquartile range was used instead of standard deviation because the polity index is ordinal, not cardinal.
Li et al., 2003). However, this effect is smaller for highly developed countries. This could be attributed to the fact that young and elderly people in more developed countries are more likely to transport in a safer mode. The urbanization variable showed a statistically significant difference between the two country groups. The estimated results suggest that an increase in the urban-to-rural population ratio can lead to an increase in non-fatal road injuries for less developed countries but a decrease in non-fatal road injuries for more developed countries. A plausible explanation is that proper urban infrastructure and public transport services in highly developed countries may reduce conflicts among vulnerable road users (such as pedestrians, motorcyclists and bicyclists) and other motor vehicle users, thus decreasing the number of non-fatal road injuries. A similar result was found for vehicles per thousand population. The results indicated that more vehicles per thousand population would lead to an increase in non-fatal road injuries in less developed countries. The opposite is true in highly developed countries. We interpret this result for increases in road crash risk exposure levels as the result of increases in the degree of mobility. Nevertheless, in highly developed countries, the road crash exposure level reduces due to the use of safer vehicles and better road infrastructure. A variable for the percentage of motorcycles was used to represent the relative degree of motorcycle use among different countries and over time. This variable is positively and significantly associated with non-fatal road injuries, confirming previous findings9. This suggests that a higher percentage of motorcycles increases the number of non-fatal road injuries. The empirical model estimates that for a 1% increase in motorcycles,
Table 5 Kuznets relationship for non-fatal road injuries for the period of 1963–2009. Variable
ln(RGDPCH) LD ln(RGDPCH) 2LD ln(RGDPCH) HD ln(RGDPCH)2HD VEHP LD VEHP HD 0.0004nn MCV ln(AGE) LD ln(AGE) HD URBAN LD URBAN HD ALC POL (1 if POLITY 4 8; 0 otherwise) DURABLE Constant 32.904nn 32.747nn 90 N Log likelihood Likelihood ratio test Turning point (USD) LD Turning point (USD) HD
Model A
Model B
Model C
Coef.
Coef.
Coef.
1.646nn 0.056nn 2.295nn 0.123nn
3.606nn 0.179nn 4.902nn 0.254nn 0.0024nn
3.596nn 0.177nn 4.918nn 0.252nn 0.0025nn 0.0004nn
0.024nn 1.828nn 0.394nn 0.013n 0.007n 0.073nn
0.024nn 1.783nn 0.301nn 0.013n 0.004n 0.071nn 0.068nn 0.002nn 18.053nn
Groups
90
90
1653 16,666 1113nn 2,039,037 11,301
1653 16,495 1453nn 23,984 15,285
1653 16,490 1464nn 26,076 17,080
LD – Less developed countries, HD – Highly developed countries. n
Significant at 5%. Significant at 1%.
nn
Table 4 Descriptive statistics for highly developed countries. Variable
The number of Injuries Population (thousands of people) Real Gross Domestic Product per-capita (US$2005 constant prices) Percentage of the population below 15 and above 64 years of age Total vehicles per thousand population Percentage of motorcycles Urban-to-rural population ratio Adult (above 15 years of age) alcohol consumption(liters per population) Political stability Democracy, 10 (strongly democratic) and 10 (strongly autocratic) a
Between 1963 and 1989 Obs. Mean
Std. dev.
Between 1990 and 2009 Obs. Mean
Std. dev.
388 388 388 388 388 388 388 388 388 388
570,993.9 46,829.9 5911.8 2.14 167.54 0.65 4.47 3.71 43.03 0a
398 398 398 398 398 398 398 398 398 398
540,195.9 49,839.04 8856.05 1.89 159.29 0.12 5.62 2.64 46.82 0a
225,880.4 29,373.2 19,113.4 35.21 298.63 0.27 3.94 10.85 51.68 8.25
207,760.9 30,606.5 25,379.9 32.76 432.17 0.11 4.11 10.35 53.15 9.69
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 drivers on 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
T.H. Law / Transport Policy 42 (2015) 166–172
ln(non-fatal road injuries per 10,000 population)
12 10 8 6 4 2 0
0
5
10
15
20
25
30
ln(GDP)
ln(non-fatal road injuries per 10,000 population)
Fig. 3. The effect of per-capita income on non-fatal road injuries for less developed countries based on Model A (Table 5).
10 8 6 4 2 0
0
2
4
6
8
10 ln(GDP)
12
14
16
18
20
Fig. 4. The effect of per-capita income on road injuries for highly developed countries based on Model A (Table 5). Table 6 Kuznets relationship for non-fatal road injuries for the period of 2003–2004. Variable
Model A Coef.
ln(RGDPCH) LD ln(RGDPCH) 2LD ln(RGDPCH) HD ln(RGDPCH)2HD Constant Groups N Log likelihood Likelihood ratio test Turning point (USD) Turning point (USD)
6.919n 0.327n 8.548nn 0.468nn 41.543nn 51 102 448.12 82.27nn 38,294 9174
LD HD
171
Several limitations of the current study should be noted. First, the dependent variable, non-fatal road injuries, which consists of slight and serious road injuries. Thus, an estimation of the effect of explanatory variables on slight and serious road injuries separately is not applicable10. Second, the political variables, democracy and political stability indexes, are used as proxy variables to determine the likelihood of enacting and enforcing road safety legislation, without taking the effect of enforcement resources into consideration. Consequently, all else being equal, two countries with an equal score of democracy and political stability will have an equal reduction in non-fatal road injuries. However, in reality, it is expected that in a country with a higher quality of political institutions, increases in enforcement resources should lead to a greater reduction in non-fatal road injuries. Lastly, previous studies (Law et al., 2009, 2011; Noland, 2003) found that improvement in the level of medical care and technology is associated with reductions in the number of road deaths. Consequently, one would expect that many road deaths could be prevented and result in non-fatal road injuries as medical care and technology advances. However, the effect of such advances could not be observed in the analysis for two reasons. First, increases in non-fatal road injuries may be due to greater accessibility and use of medical care among those who are moderately injured, as medical care and technology continue to improve. Second, even if there is a shift from fatal to non-fatal road injuries, it would be a minor percentage of the total injuries and might not be detectable statistically. This study examines the empirical Kuznets curve relationship between non-fatal road injuries and per-capita income. Although the evidence implies the Kuznets curve effect on non-fatal road injuries, we cannot assume that today's less developed countries will inevitably follow a similar path. The results suggest that improvements in road infrastructure, the quality of regulatory institutions (as represented by political stability and democratic institutions), and increase in the use of safer transport modes will help reduce non-fatal road injuries. However, less developed countries generally suffer from lower judicial efficiency in law enforcement (WHO, 2013)11 and higher levels of motorcycle ownership, which tend to persist over time. In view of this, it will be beneficial to improve the quality of regulatory institutions and promote the use of safer transport modes, especially in less developed countries, to achieve a sustainable improvement in road safety. Economic growth may be a prerequisite, but it alone does not guarantee that positive change will happen.
LD – Less developed countries, HD – Highly developed countries. n
Significant at 5%. Significant at 1%.
nn
lead to a significant increase in non-fatal road injuries. The results indicated that a one-liter per-capita increase in alcohol consumption led to a 7.5% increase in non-fatal injuries per thousand population. Previous studies found that driver skills related to perception, information processing and judgement are significantly impaired even at lower blood alcohol concentration levels (Moskowitz et al., 2000; Moskowitz and Robinson, 1988; Moskowitz and Burns, 1990).
5. Conclusions This paper investigates the empirical relationship between non-fatal road injuries and per-capita GDP using an international panel data set over the period of 1963–2009. The empirical results suggested that non-fatal road injuries follow an inverted U-shaped, or Kuznets curve, relationship with per-capita GDP for less developed and highly developed countries.
Acknowledgment This research was funded by Universiti Putra Malaysia's Research University Grant Scheme (Grant no. 05-05-10-1108RU). The funding source played absolutely no role in the study design nor any collection, analysis, or interpretation of the data.
References Anbarci, N., Escaleras, M., Register, C., 2006. Traffic fatalities and public sector corruption. KYKLOS 59, 327–344. Anbarci, N., Escaleras, M., Register, C.A., 2009. Traffic fatalities: does income inequality create an externality? Can. J. Econ. 42, 244–266. Baltagi, B., 2001. Econometric Analysis of Panel Data. John Wiley and Sons, West Sussex.
10 Separate data on slight and serious road injuries are not available for most of the less developed countries. 11 According to the World Health Organization data, less than 35% of low- and middle-income countries had road safety policies in place to protect vulnerable road users, such as pedestrians and cyclists.
172
T.H. Law / Transport Policy 42 (2015) 166–172
Beeck, E.F.V., GJJM, Borsboom, Mackenbach, J.P., 2000. Economic development and traffic accident mortality in the industrialized world, 1962–1990. Int. J. Epidemiol. 29, 503–509. Bhalla, K., Ezzati, M., Mahal, A., Salomon, J., Reich, M., 2007. A risk-based method for modelling traffic fatalities. Risk Anal. 27, 125–136. Bishai, D., Quresh, A., James, P., Ghaffar, A., 2006. National road casualties and economic development. Health Econ. 15, 65–81. Byers, A.L., Allore, H., Gill, T.M., Peduzzi, P.N., 2003. Application of negative binomial modelling for discrete outcomes: a case study in ageing research. J. Clin. Epidemiol. 56, 559–564. Cameron, A.C., Trivedi, P.K., 1998. Regression Analysis of Count Data. Cambridge University Press, Cambridge. Chang, B.H., Pocock, S., 2000. Analysing data with clumping at zero: an example demonstration. 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 and regulatory 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 other road 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 with an 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 of mortality 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 between 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 Administration Report No. DOT HS 807 280. Department of Transportation, Washington, DC. Moskowitz, H., Burns, M., 1990. Effects of alcohol on driving performance. Alcohol Health Res. World 14, 12–14. Moskowitz, H., Burns, M., Fiorentino, D., Smiley, A., Zador, P., 2000. Driver Characteristics and Impairment at Various BACs. Southern California Research Institute, National Highway Traffic Safety Administration, Washington, DC. Noland, R.B., 2003. Medical treatment and traffic fatality reductions in industrialized countries. Accid. Anal. Prev. 35, 877–883. Paulozzi, L.J., Ryan, G.W., Espitia-Hardeman, V.E., Xi, Y., 2007. Economic development's effect on road transport-related mortality among different types of road 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 Organization, 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 motorcycle accidents: short-term impacts of the running headlights campaign and regulation. J. Traffic Med. 23, 17–28. Sato, Y., Yamamoto, K., 2005. Population concentration, urbanization, and demographic transition. J. Urban Econ. 58, 45–61. Söderlund, N., Zwi, A.B., 1995. Traffic-related mortality in industrialized and less developed countries. Bull. World Health Organ. 73, 175–182. Williams, A.F., Carsten, O., 1989. Driver age and crash involvement. Am. J. Public Health 79, 326–327. Winkelmann, R., 1997. Econometric Analysis of Count Data. Springer Verlag, New York. 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: Supporting A Decade of Action, Geneva. Wooldridge, J.M., 2000. Introductory Econometrics: A Modern Approach. SouthWestern College Publishing, New York.
Contents lists available at ScienceDirect
Transport Policy journal homepage: www.elsevier.com/locate/tranpol
Factors associated with the relationship between non-fatal road injuries and economic growth Teik Hua Law n Road Safety Research Centre, Faculty of Engineering, Universiti Putra Malaysia, 43400 Serdang, Selangor, Malaysia
art ic l e i nf o
a b s t r a c t
Article history: Received 9 October 2014 Received in revised form 29 May 2015 Accepted 5 June 2015
This study reports the results of an empirical analysis of the Kuznets curve relationship between nonfatal road injuries and per-capita income. This relationship indicates that the number of road deaths increases with increasing per-capita income at lower income levels, but decreases once it has exceeded a threshold level. We apply a fixed effects negative binomial regression analysis on a panel of 90 countries over the period of 1963–2009. Results indicated evidence of an inverted U-shaped relationship between economic 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 developed countries. The evidence presented in this study suggests that improvements in road infrastructure, the quality of regulatory institutions, and increase in the use of safer transport modes will help reduce non-fatal road injuries. & 2015 Elsevier Ltd. All rights reserved.
Keywords: Kuznets relationship Non-fatal road injuries Democracy Political stability
1. Introduction The relevance of the relationship between economic growth and road safety has long been recognized. However, it has been the impact of economic growth on road deaths that has attracted the most attention (Wintemute, 1985; Söderlund and Zwi, 1995; Beeck et al., 2000; Kopits and Cropper, 2005). Recently, numerous studies have shown that there is an inverted U-shaped relationship 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 decrease once the number has exceeded a threshold level. This relationship is also known as the Kuznets curve relationship, which existed between income inequality and per-capita income (Kuznets, 1955). It is important to understand how economic growth affects road safety because road safety concerns could lead to a primary road safety policy agenda. This relationship has been explained by several studies to be the result of changes in the fundamental development of a country. This includes changes in variations in rates among vulnerable non-motorized road users (Paulozzi et al., 2007), advances in medical services (Law et al., 2009, 2011; Noland, 2003), the quality of political institutions (Anbarci et al., 2006; Law et al., 2011), income equality (Anbarci et al., 2009) and road safety policies (Law et al., 2009). n
Fax: þ 60 386567129. E-mail address: [email protected]
http://dx.doi.org/10.1016/j.tranpol.2015.06.004 0967-070X/& 2015 Elsevier Ltd. All rights reserved.
The Kuznets curve hypothesis for road safety can be supported by three theoretical explanations: the scale of economic activities, changes in vehicle composition, and a surge in demand for better road safety as per-capita income rises (Law et al., 2009). The Kuznets curve hypothesis contends that the number of road deaths increases initially as a country's economy develops, owing to the fact that the growth of a country's economy is accompanied by a corresponding surge in demand for transportation services (Dargay and Gately, 1999). Several previous studies have found that one of the contributing factors to the increase in road crashes and injuries is the growing number of vehicles per-capita (Bishai et al., 2006; Kopits and Cropper, 2005). The composition effect depicts the change in vehicle composition as a country's economy grows. This involves a change from low-threat and high-risk transport modes (such as pedestrians and 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 an increase in the number of motorized vehicles can pose an increasing threat to the predominantly more vulnerable transport modes. Nevertheless, at higher levels of economic development, when a majority of the commuting population are less vulnerable motor vehicle occupants, there is an increase in the number of motor vehicles causing a reduction in the total road death risk. This indicates an inverted U-shaped relationship between road deaths and income. The abatement effect is the result of road death risk alleviation
T.H. Law / Transport Policy 42 (2015) 166–172
measures, which reflects both supply side and demand influences. On the demand side, at low income levels, people are less capable to invest in road safety, even though there is a demand for it, resulting in low abatement. Nonetheless, as the level of income rises, people pay greater attention to road safety and they can afford higher-cost road safety enhancement measures, enabling more commuters to switch to lower-risk modes of transport. On the supply side, at low levels of income, society is unable to afford sufficient resources to establish the social institutions needed to regulate road safety interventions. With a higher income level, abatement efforts become more substantial because more resources are available to invest in road safety. Therefore, road safety regulatory institutions become more effective. This supply side of abatement, together with the demand side mentioned above, is expected to yield a declining relationship between road deaths and income. Bishai et al. (2006) found that at higher income levels, an increase in per-capita income appears to reduce road deaths, but does not reduce road crashes and injuries. However, their analysis does not explain the underlying mechanism that drives these relationships. In addition, their analysis applied a fixed effect ordinary regression, which is inappropriate to model count data. Thus, the objective of this study is to understand how economic growth affects non-fatal road injuries and what factors underlie this relationship. The analysis presented in this study examines the relationship between non-fatal road injuries and economic growth, spanning a 47-year period. Several explanatory variables, which are related to the fundamental development of a country, are included to account for this relationship. In particular we include motorization level, the urban-to-rural population ratio, the percentage of the population below the age of 15 and over the age of 64, democracy level, political stability, and adult alcohol consumption rate.
2. Methodology The analysis of count data, such as non-fatal road injuries, often does not follow the underlying assumption of normality, limited to non-negative integer values and the distribution is highly skewed (Gardner et al., 1995; Cameron and Trivedi, 1998). Therefore, the conventional ordinary linear regression may not be appropriate to analyze this sort of data (Long, 1997). Transformation of the data may not yield normally distributed data. Furthermore, this may cause difficulty in interpreting the regression coefficients because the transformed data are not estimated on the original scale (Byers et al., 2003) and the accuracy of the estimated results is questionable (Chang and Pocock, 2000). In view of the above facts, the Poisson and the negative binomial regression methods are frequently used to model count data. However, the selection of the Poisson or the negative binomial regression is based on the difference between mean and variance. The Poisson regression model assumes the mean is equal to variance, also known as equidispersion. If this assumption is violated (overdispersion), the Poisson model will produce consistent coefficient estimates, but standard errors will be underestimated (Cameron and Trivedi, 1998; Winkelmann, 1997). The log-likelihood ratio test is used to test the hypothesis that the variance and mean is equal, which indicates equidispersion. Rejection of the null hypothesis implies that the negative binomial is more appropriate than the Poisson regression. In the models estimated in this study, the log-likelihood ratio test showed that the null hypothesis can be rejected at 5% level of significance, implying that the negative binomial regression model is the preferred model. The present study used the design that combine longitudinal
167
and cross-sectional methods. The fixed effects negative binomial regression, derived by Hausman et al. (1984), is employed in this study. The fixed effects model assumes that the country specific intercept is correlated with explanatory variables. The first advantage of this model is that it can cancel out the dispersion parameters and account for heterogeneity in the data1. The second advantage of this model is the country specific intercepts are able to take into account differences in accident data derivation due to different sources of accident data used2. Alternatively, another method known as random effects model assumes that the inverse of the over-dispersion parameter is distributed as a beta distribution. The assumption applied in this model is that country specific effects as part of the error term. Another assumption is that the country specific effect is uncorrelated with the explanatory variables, which is often unrealistic (Baltagi, 2001; Wooldridge, 2000). The choice between fixed and random effects model is based on the Hausman test (Baltagi, 2001). The null hypothesis for the test is the country specific intercept is uncorrelated with other explanatory variables. Rejection of the null hypothesis indicating fixed effects model is more appropriate. For the data analyzed here, it was found that the more appropriate effect for all models was the fixed effect3. An offset variable is included in the analysis to normalize the effect of risk exposure on non-fatal road injuries. This is necessary because a country with a higher level of risk exposure should experience more road crashes. The offset variable is specified as the logarithm of a measure of risk exposure in the equation and can be written as
log(μit ) = αi + βxit + log(Eit )
(1)
where μit denotes the expected number of non-fatal road injuries, Eit represents an index of risk exposure, αi is the country specific intercept, x it is a vector of covariate which describes the characteristics of an observation unit i during a given time period t, β is the model parameters. This equation can be rewritten as
log(μit ) − log(Eit ) = αi + βxit
(2)
and then
log(μit /Eit ) = αi + βxit
(3)
The coefficient for the explanatory variables is interpreted as effects on rates rather than a count. From Eq. (1), the expected number of non-fatal road injuries is given by μit = Eit exp(α + βx it ). This means that the expected number of non-fatal road injuries is proportional to the level of risk exposure.
3. Data The empirical analysis in this paper uses an unbalanced4 panel dataset which consists of a total of 1653 observations from 90 countries (both less developed and highly developed countries) for the period between 1963 and 2009. These countries were selected based on the availability of data on all explanatory variables for at least three years. The sample is divided into two groups:- less developed and highly developed countries, according to the definition of the International Monetary Fund (IMF) (2014) and the 1 This is done by conditioning on the total number of non-fatal road injuries that occurred within each country during the study period. 2 Most of the developed countries use hospital data as the source, while in developing 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.
168
T.H. Law / Transport Policy 42 (2015) 166–172
Table 1 The list of less developed countries.
Table 2 The list of highly developed countries.
No.
Country
Period
Observations
Included in Table 6
No.
Country
Period
Observations
Included in Table 6
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60
Albania Azerbaijan Bahrain Bangladesh Belarus Benin Botswana Brazil Bulgaria Cambodia Cameroon Chile Colombia Costa Rica Cuba Ecuador Egypt El Salvador Ethiopia Georgia India Indonesia Kazakhstan Kenya Kuwait Lithuania Malawi Malaysia Mali Mauritius Mexico Moldova Mongolia Morocco Namibia Nicaragua Niger Nigeria Oman Pakistan Panama Peru Philippines Romania Saudi Arab Senegal South Africa Sri Lanka Syrian Thailand Togo Tunisia Turkey Uganda Ukraine Uruguay Venezuela Vietnam Zambia Zimbabwe
1992–2003 1993–2007 1987–2008 1991–2005 1994–2007 1990–2007 1968–2006 1975–2007 1989–2008 1992–1997 1965–2006 1979–2006 1985–2009 1986–1999 1991–1997 1990–1998 1982–1994 1990–2009 1973–2007 1993–2003 1965–2005 1965–2008 1999–2007 1965–2004 1986–2009 1993–2008 1964–1984 1963–2008 1969–2008 1968–2004 1990–2004 1992–2007 1990–1999 1969–2007 1991–2007 1989–2002 1984–2008 1973–1993 1990–2007 1972–2009 1977–1998 1969–1972 1984–2000 1990–1999 1988–2005 1967–2008 1975–2005 1969–2006 1965–2008 1967–1996 1973–2007 1963–2007 1963–2007 1990–1993 1994–2006 1990–2007 1967–1986 2003–2007 1969–1972 1993–1995
13 12 15 10 7 8 28 16 17 6 3 21 17 14 7 9 6 9 23 8 25 32 6 29 11 13 6 43 9 21 6 10 10 34 5 8 7 11 8 31 10 4 15 10 9 22 18 24 13 30 16 30 42 4 10 10 14 5 4 3
√ √
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30
Australia Austria Belgium Canada Cyprus Czech Republic Denmark Estonia Finland France Germany Greece Hungary Ireland Israel Italy Japan Korea, Rep. Netherlands New Zealand Norway Poland Portugal Slovakia Slovenia Spain Sweden Switzerland United Kingdom United States
1980–2007 1963–2009 1963–2006 1969–2006 1981–2007 1993–2008 1980–2007 1991–2008 1990–2008 1970–2008 1991–2008 1971–2008 1971–2008 1992–2006 2003–2007 1993–2008 1963–2006 1981–2007 1963–2007 1963–2007 1963–2008 1983–2008 1963–2006 2003–2007 1991–2008 1995–2006 1964–2006 1963–2009 1970–2009 1978–2004
31 44 41 32 24 13 25 15 16 30 15 35 35 12 5 11 41 24 37 43 43 24 30 5 15 9 31 40 37 23
√ √ √ √ √ √ √ √ √ √ √ √ √ √ √ √ √ √ √ √ √ √ √ √ √ √ √ √ √
√
√ √ √
√ √ √ √ √ √ √
√ √ √
√
√ √
√ √
√
Central Intelligence Agency (CIA) (2014). This will allow us to examine the effect of per-capita income on non-fatal road injuries separately for less developed and highly developed countries. Tables 1 and 2 present the list of countries included in this analysis. The data on non-fatal road injuries and motor vehicles comes from two international sources, the International Road Traffic and Accident Database and the International Road Federation's World Road 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 a proxy for per-capita income. We also include the square of percapita GDP to capture the potential quadratic effect on non-fatal road 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 either democratic or autocratic. It is a composite of five elements, which includes competitiveness of political participation, regulation of political participation, competitiveness of executive recruitment, openness of executive recruitment, and institutionalized constraints on executive 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 polity score from 2 to 3 is not necessarily the same as when the score increases from 6 to 7. Therefore, the polity index is not truly a continuous variable (Keefer et al., 2011). As the polity index scores in this study were heavily clustered toward the upper end of the scale5, they were categorized into two levels:- 10 to 8 (low democracy, coded as 0) and 9–10 (high democracy, coded as 1). Another political variable that might impact the level of road injuries would be the political stability of a country. As a proxy for political stability, we used regime durability drawn from the Polity IV database, which is defined as the length of time since the most recent regime change6. Higher regime durability indicates higher political stability. The urban-to-rural population ratio, which was obtained from the World Development Indicator Database, is used to measure the urbanization level. Urbanization refers to urban areas have higher population density (Sato and Yamamoto, 2005). This variable was used to account for the effect of population density on non-fatal road injuries in urban areas. Data on the percentage of the 5
48% of the polity index scores in the dataset were greater than 8. 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. 6
T.H. Law / Transport Policy 42 (2015) 166–172 12
12 11 10 9 8 7 6 5 4
5
6
7
8 ln(GDP)
9
10
11
ln(non-fatal road injuries per 10,000 population)
ln(non-fatal road injuries per 10,000 population)
13
169
4. Results and discussions Figs. 1 and 2 show the actual and the estimated non-fatal road injuries for both less developed and highly developed countries. As shown in Fig. 1, for less developed countries, the non-fatal road injuries to per-capita income relationship is positive over most of the sample range. This implies that in less developed countries, increases in per-capita income lead to an increase in non-fatal road injuries. On the other hand, for more developed countries, non-fatal road injuries initially increase and then decrease as income rises. This can be visualized in Fig. 2, which shows that there is an inverted U-shaped relationship between non-fatal road injuries and per-capita income for highly developed countries. Tables 3 and 4 report descriptive statistics for all variables, for less and highly developed countries, respectively. As shown, the number of injuries rose significantly in less developed countries, but declined in more developed countries as income and population increased. Although highly developed countries had higher total vehicles per thousand population than less developed countries, less developed countries had higher percentages of motorcycles. Compared to less developed countries, highly developed countries were also characterized by higher urbanization, higher alcohol consumption rate, and greater political stability and democracy. Table 5 lists results from the fixed effects negative binomial model on an unbalanced panel of less developed and highly developed countries for which data are available between 1963 and 2009. The dependent variable used is non-fatal road injuries, which includes slight and serious injuries as the result of road crashes. The log population was used as an offset variable. The percentage of the population below the age of 15 and over the age of 64, the urban-to-rural population ratio, vehicles per thousand population and per-capita GDP variables are interacted with the country group dummy7 (countries are classified as less developed and highly developed countries). These interaction terms measure the effect of the country's fundamental development on non-fatal road injuries in different country groups. The likelihood ratio test is used for two purposes. First it is used to test overall model fit and significance. Second, it examines the null hypothesis of equidispersion. The likelihood ratio tests show that, for all models, the null hypothesis that all coefficients (except the intercept) are jointly equal to zero is rejected. The value of the likelihood ratio statistic increases from Models A to C, which 7 The country group dummy variable is coded as 1 for highly developed countries and 0 for less developed countries.
10
9
8
Fig. 1. The actual and the estimated non-fatal road injuries for less developed countries.
population below the age of 15 and over the age of 64 were taken from the same database. Another variable used in this study is the recorded adult (above 15 years of age) per-capita alcohol consumption, taken from the World Health Organization's Global Alcohol Database.
11
7
8
9
10
11
12
ln(GDP)
Fig. 2. The actual and the estimated non-fatal road injuries for highly developed countries.
indicates that model fit improves when more variables are included into the model. The likelihood ratio tests also reject the null hypothesis of equidispersion8, which suggests that the negative binomial model is more appropriate than the Poisson model. The results in Models A to C in Table 5 confirm a statistically significant inverted U-shaped or Kuznets relationship between per-capita GDP and non-fatal road injuries for less developed and highly developed countries. This is shown in Figs. 3 and 4, respectively. The inverted U-shaped relationship remains statistically significant after controlling the other explanatory variables. This suggests that some underlying factors that explain the changes of non-fatal road injuries and are correlated with percapita GDP are not considered in the analysis. The results also indicated that for a highly developed country, the turning point was in the range of US$11,301 to US$21,284, while for a less developed country, it was higher (US$25,474 to USD$2,039,037). The high turning point for a less developed country is consistent with Fig. 1, which shows a positive relationship between non-fatal road injuries and per-capita GDP over most of the sample range. We performed additional analysis using data between 2003 and 2004. Tables 1 and 2 present the list of countries included in this analysis. The result in Table 6 is consistent with that in Model A in Table 5, indicating that the result is not a mere consequence of the improvements in road accident data quality as a country develops. Two political variables, polity and durability, were used to examine the effect of the quality of regulatory institutions on nonfatal road injuries. The negative and statistically significant coefficient for polity and durability in Model C shows that greater levels of democracy and political stability are associated with lower nonfatal road injuries. Damania et al. (2004) indicated that an increase in political stability would lead to a more judicial efficiency in law enforcement. In more politically stable countries, judicial institutions are more effective in enforcing road safety regulations and policies. Moreover, Payne (1995) revealed that citizens of democracies had more freedom to express their concerns and demands about environmental issues and put pressure on policymakers to respond positively to such demands. These results are consistent with Law et al.'s (2013) findings that countries with a higher level of democracy and political stability were more likely to implement road safety legislations with greater judicial efficiency. Consequently, this would reduce the number of non-fatal road injuries. The estimated results for the percentage of the population below the age of 15 and over the age of 64 indicated that more young and elderly people are associated with increased non-fatal road injuries in less developed countries. Several studies reported that the population at these age groups are more vulnerable physically 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.
170
T.H. Law / Transport Policy 42 (2015) 166–172
Table 3 Descriptive statistics for less developed countries. Variable
The number of Injuries Population (thousands of people) Real Gross Domestic Product per-capita (US$2005 constant prices) Percentage of the population below 15 and above 64 years of age Total vehicles per thousand population Percentage of motorcycles Urban-to-rural population ratio Adult (above 15 years of age) alcohol consumption(liters per population) Political stability Democracy, 10 (strongly democratic) and 10 (strongly autocratic) a
Between 1963 and 1989 Obs. Mean
Std. dev.
Between 1990 and 2009 Obs. Mean
Std. dev.
397 397 397 397 397 397 397 397 397 397
29,090.7 123,418.5 3846.4 4.15 50.15 0.65 4.83 2.83 13.75 14a
470 470 470 470 470 470 470 470 470 470
58,504.7 146,914.1 6846.1 5.82 102.95 0.62 8.11 4.09 18.71 12a
19,980.8 52,636.1 3247.5 45.7 34.9 0.49 1.33 2.58 12.55 1.07
29,885.1 55,929.1 6135.7 38.8 95.5 0.56 2.99 3.95 18.78 3.15
The interquartile range was used instead of standard deviation because the polity index is ordinal, not cardinal.
Li et al., 2003). However, this effect is smaller for highly developed countries. This could be attributed to the fact that young and elderly people in more developed countries are more likely to transport in a safer mode. The urbanization variable showed a statistically significant difference between the two country groups. The estimated results suggest that an increase in the urban-to-rural population ratio can lead to an increase in non-fatal road injuries for less developed countries but a decrease in non-fatal road injuries for more developed countries. A plausible explanation is that proper urban infrastructure and public transport services in highly developed countries may reduce conflicts among vulnerable road users (such as pedestrians, motorcyclists and bicyclists) and other motor vehicle users, thus decreasing the number of non-fatal road injuries. A similar result was found for vehicles per thousand population. The results indicated that more vehicles per thousand population would lead to an increase in non-fatal road injuries in less developed countries. The opposite is true in highly developed countries. We interpret this result for increases in road crash risk exposure levels as the result of increases in the degree of mobility. Nevertheless, in highly developed countries, the road crash exposure level reduces due to the use of safer vehicles and better road infrastructure. A variable for the percentage of motorcycles was used to represent the relative degree of motorcycle use among different countries and over time. This variable is positively and significantly associated with non-fatal road injuries, confirming previous findings9. This suggests that a higher percentage of motorcycles increases the number of non-fatal road injuries. The empirical model estimates that for a 1% increase in motorcycles,
Table 5 Kuznets relationship for non-fatal road injuries for the period of 1963–2009. Variable
ln(RGDPCH) LD ln(RGDPCH) 2LD ln(RGDPCH) HD ln(RGDPCH)2HD VEHP LD VEHP HD 0.0004nn MCV ln(AGE) LD ln(AGE) HD URBAN LD URBAN HD ALC POL (1 if POLITY 4 8; 0 otherwise) DURABLE Constant 32.904nn 32.747nn 90 N Log likelihood Likelihood ratio test Turning point (USD) LD Turning point (USD) HD
Model A
Model B
Model C
Coef.
Coef.
Coef.
1.646nn 0.056nn 2.295nn 0.123nn
3.606nn 0.179nn 4.902nn 0.254nn 0.0024nn
3.596nn 0.177nn 4.918nn 0.252nn 0.0025nn 0.0004nn
0.024nn 1.828nn 0.394nn 0.013n 0.007n 0.073nn
0.024nn 1.783nn 0.301nn 0.013n 0.004n 0.071nn 0.068nn 0.002nn 18.053nn
Groups
90
90
1653 16,666 1113nn 2,039,037 11,301
1653 16,495 1453nn 23,984 15,285
1653 16,490 1464nn 26,076 17,080
LD – Less developed countries, HD – Highly developed countries. n
Significant at 5%. Significant at 1%.
nn
Table 4 Descriptive statistics for highly developed countries. Variable
The number of Injuries Population (thousands of people) Real Gross Domestic Product per-capita (US$2005 constant prices) Percentage of the population below 15 and above 64 years of age Total vehicles per thousand population Percentage of motorcycles Urban-to-rural population ratio Adult (above 15 years of age) alcohol consumption(liters per population) Political stability Democracy, 10 (strongly democratic) and 10 (strongly autocratic) a
Between 1963 and 1989 Obs. Mean
Std. dev.
Between 1990 and 2009 Obs. Mean
Std. dev.
388 388 388 388 388 388 388 388 388 388
570,993.9 46,829.9 5911.8 2.14 167.54 0.65 4.47 3.71 43.03 0a
398 398 398 398 398 398 398 398 398 398
540,195.9 49,839.04 8856.05 1.89 159.29 0.12 5.62 2.64 46.82 0a
225,880.4 29,373.2 19,113.4 35.21 298.63 0.27 3.94 10.85 51.68 8.25
207,760.9 30,606.5 25,379.9 32.76 432.17 0.11 4.11 10.35 53.15 9.69
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 drivers on 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
T.H. Law / Transport Policy 42 (2015) 166–172
ln(non-fatal road injuries per 10,000 population)
12 10 8 6 4 2 0
0
5
10
15
20
25
30
ln(GDP)
ln(non-fatal road injuries per 10,000 population)
Fig. 3. The effect of per-capita income on non-fatal road injuries for less developed countries based on Model A (Table 5).
10 8 6 4 2 0
0
2
4
6
8
10 ln(GDP)
12
14
16
18
20
Fig. 4. The effect of per-capita income on road injuries for highly developed countries based on Model A (Table 5). Table 6 Kuznets relationship for non-fatal road injuries for the period of 2003–2004. Variable
Model A Coef.
ln(RGDPCH) LD ln(RGDPCH) 2LD ln(RGDPCH) HD ln(RGDPCH)2HD Constant Groups N Log likelihood Likelihood ratio test Turning point (USD) Turning point (USD)
6.919n 0.327n 8.548nn 0.468nn 41.543nn 51 102 448.12 82.27nn 38,294 9174
LD HD
171
Several limitations of the current study should be noted. First, the dependent variable, non-fatal road injuries, which consists of slight and serious road injuries. Thus, an estimation of the effect of explanatory variables on slight and serious road injuries separately is not applicable10. Second, the political variables, democracy and political stability indexes, are used as proxy variables to determine the likelihood of enacting and enforcing road safety legislation, without taking the effect of enforcement resources into consideration. Consequently, all else being equal, two countries with an equal score of democracy and political stability will have an equal reduction in non-fatal road injuries. However, in reality, it is expected that in a country with a higher quality of political institutions, increases in enforcement resources should lead to a greater reduction in non-fatal road injuries. Lastly, previous studies (Law et al., 2009, 2011; Noland, 2003) found that improvement in the level of medical care and technology is associated with reductions in the number of road deaths. Consequently, one would expect that many road deaths could be prevented and result in non-fatal road injuries as medical care and technology advances. However, the effect of such advances could not be observed in the analysis for two reasons. First, increases in non-fatal road injuries may be due to greater accessibility and use of medical care among those who are moderately injured, as medical care and technology continue to improve. Second, even if there is a shift from fatal to non-fatal road injuries, it would be a minor percentage of the total injuries and might not be detectable statistically. This study examines the empirical Kuznets curve relationship between non-fatal road injuries and per-capita income. Although the evidence implies the Kuznets curve effect on non-fatal road injuries, we cannot assume that today's less developed countries will inevitably follow a similar path. The results suggest that improvements in road infrastructure, the quality of regulatory institutions (as represented by political stability and democratic institutions), and increase in the use of safer transport modes will help reduce non-fatal road injuries. However, less developed countries generally suffer from lower judicial efficiency in law enforcement (WHO, 2013)11 and higher levels of motorcycle ownership, which tend to persist over time. In view of this, it will be beneficial to improve the quality of regulatory institutions and promote the use of safer transport modes, especially in less developed countries, to achieve a sustainable improvement in road safety. Economic growth may be a prerequisite, but it alone does not guarantee that positive change will happen.
LD – Less developed countries, HD – Highly developed countries. n
Significant at 5%. Significant at 1%.
nn
lead to a significant increase in non-fatal road injuries. The results indicated that a one-liter per-capita increase in alcohol consumption led to a 7.5% increase in non-fatal injuries per thousand population. Previous studies found that driver skills related to perception, information processing and judgement are significantly impaired even at lower blood alcohol concentration levels (Moskowitz et al., 2000; Moskowitz and Robinson, 1988; Moskowitz and Burns, 1990).
5. Conclusions This paper investigates the empirical relationship between non-fatal road injuries and per-capita GDP using an international panel data set over the period of 1963–2009. The empirical results suggested that non-fatal road injuries follow an inverted U-shaped, or Kuznets curve, relationship with per-capita GDP for less developed and highly developed countries.
Acknowledgment This research was funded by Universiti Putra Malaysia's Research University Grant Scheme (Grant no. 05-05-10-1108RU). The funding source played absolutely no role in the study design nor any collection, analysis, or interpretation of the data.
References Anbarci, N., Escaleras, M., Register, C., 2006. Traffic fatalities and public sector corruption. KYKLOS 59, 327–344. Anbarci, N., Escaleras, M., Register, C.A., 2009. Traffic fatalities: does income inequality create an externality? Can. J. Econ. 42, 244–266. Baltagi, B., 2001. Econometric Analysis of Panel Data. John Wiley and Sons, West Sussex.
10 Separate data on slight and serious road injuries are not available for most of the less developed countries. 11 According to the World Health Organization data, less than 35% of low- and middle-income countries had road safety policies in place to protect vulnerable road users, such as pedestrians and cyclists.
172
T.H. Law / Transport Policy 42 (2015) 166–172
Beeck, E.F.V., GJJM, Borsboom, Mackenbach, J.P., 2000. Economic development and traffic accident mortality in the industrialized world, 1962–1990. Int. J. Epidemiol. 29, 503–509. Bhalla, K., Ezzati, M., Mahal, A., Salomon, J., Reich, M., 2007. A risk-based method for modelling traffic fatalities. Risk Anal. 27, 125–136. Bishai, D., Quresh, A., James, P., Ghaffar, A., 2006. National road casualties and economic development. Health Econ. 15, 65–81. Byers, A.L., Allore, H., Gill, T.M., Peduzzi, P.N., 2003. Application of negative binomial modelling for discrete outcomes: a case study in ageing research. J. Clin. Epidemiol. 56, 559–564. Cameron, A.C., Trivedi, P.K., 1998. Regression Analysis of Count Data. Cambridge University Press, Cambridge. Chang, B.H., Pocock, S., 2000. Analysing data with clumping at zero: an example demonstration. 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 and regulatory 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 other road 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 with an 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 of mortality 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 between 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 Administration Report No. DOT HS 807 280. Department of Transportation, Washington, DC. Moskowitz, H., Burns, M., 1990. Effects of alcohol on driving performance. Alcohol Health Res. World 14, 12–14. Moskowitz, H., Burns, M., Fiorentino, D., Smiley, A., Zador, P., 2000. Driver Characteristics and Impairment at Various BACs. Southern California Research Institute, National Highway Traffic Safety Administration, Washington, DC. Noland, R.B., 2003. Medical treatment and traffic fatality reductions in industrialized countries. Accid. Anal. Prev. 35, 877–883. Paulozzi, L.J., Ryan, G.W., Espitia-Hardeman, V.E., Xi, Y., 2007. Economic development's effect on road transport-related mortality among different types of road 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 Organization, 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 motorcycle accidents: short-term impacts of the running headlights campaign and regulation. J. Traffic Med. 23, 17–28. Sato, Y., Yamamoto, K., 2005. Population concentration, urbanization, and demographic transition. J. Urban Econ. 58, 45–61. Söderlund, N., Zwi, A.B., 1995. Traffic-related mortality in industrialized and less developed countries. Bull. World Health Organ. 73, 175–182. Williams, A.F., Carsten, O., 1989. Driver age and crash involvement. Am. J. Public Health 79, 326–327. Winkelmann, R., 1997. Econometric Analysis of Count Data. Springer Verlag, New York. 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: Supporting A Decade of Action, Geneva. Wooldridge, J.M., 2000. Introductory Econometrics: A Modern Approach. SouthWestern College Publishing, New York.