- Email: [email protected]
Ship safety index
Transportation Research Part A 66 (2014) 75–87
Contents lists available at ScienceDirect
Transportation Research Part A journal homepage: www.elsevier.com/locate/tra
Ship safety index Kevin X. Li a, Jingbo Yin b,⇑, Lixian Fan c a b c
Dept. of International Logistics, Chung-Ang University, South Korea Dept. of International Shipping, Shanghai Jiao Tong University, China School of Management, Shanghai University, China
a r t i c l e
i n f o
Article history: Received 3 May 2012 Received in revised form 31 March 2014 Accepted 25 April 2014
Keywords: Maritime safety Binary regression Safety index Risk rank system Safety database
a b s t r a c t This paper develops a new quantitative safety index for each worldwide sea-going vessel based on their condition information and safety records. The safety index can generate a relative risk score using binary logistic regression method and a dataset with both static and dynamic information covering over 90% of the world sea-going merchant fleet. It has a widely potential usage for both industry and academic research, e.g., for port authorities to determine whether an on board inspection is needed; for insurers to determine premium rate; and for shipowners to identify functional areas for repair and maintenance. Ó 2014 Elsevier Ltd. All rights reserved.
1. Introduction There are a total of around 120,000 sea-going commercial vessels of over 100 dead weight tonnage (dwt) sailing around the world. With an increasing awareness of environmental protection and safety issues, maritime authorities around the world have been making more rigorous efforts than ever before to promote safety and security at sea (Li and Zheng, 2008; Li et al., 2014; Yin et al., 2014). Since 1993, the International Maritime Organization (IMO) has adopted the International Safety Management Code (ISM) as a minimum statutory requirement for vessel operators to establish, implement and maintain their safety management systems. Although this program is implemented as a regulatory to oversee the safety activities of the vessels at the operational level, it cannot be used to measure and predict the safety level of each vessel. To examine comparative safety levels for safety management and improvement purposes, an evaluation mechanism for measuring the overall safety level of the vessels is needed. In this paper, we establish a new quantitative safety index to serve as a proactive evaluation mechanism to measure the safety level of a vessel, which provides useful information for decision making on safety management and improvements. The safety index, using both static and dynamic data that have been collected from various sources, forms the most comprehensive data set to date on vessel information (for details see Section 2). We investigate the effects of various risk factors using multivariate logistic regression modeling, and assess how the various factors simultaneously affect a vessel’s safety level. The new safety index can be used as a safety benchmark for both industry and academic research. For example, it can be used by port authorities to determine whether the right of access to territorial waters or ports should be granted, or whether on board inspections are required for vessels calling at their ports, in order to prevent oil pollution and accidents within their territorial waters (Yap and Lam, 2013). It can be used by insurers to determine the premium rate prior to an insurance
⇑ Corresponding author. Tel.: +86 021-66137696. E-mail address: [email protected] (J. Yin). http://dx.doi.org/10.1016/j.tra.2014.04.016 0965-8564/Ó 2014 Elsevier Ltd. All rights reserved.
76
K.X. Li et al. / Transportation Research Part A 66 (2014) 75–87
contract being concluded, or by any party who is interested in the safety and quality of the vessel and who needs to check the safety index before engaging in any business with the vessel. The safety index can provide useful information for operators to understand the relative safety strengths and weaknesses in terms of manageable safety characteristics, and to identify functional areas for safety improvement and maintenance. 2. Data description For this study, a total dataset with three sub-datasets has been built. The total dataset is a combination of accounting for about 130,000 vessels, including information about 10,000 lost vessels and 120,000 existing vessels, counting more than 90% of the worldwide commercial fleet. The first sub-dataset containing the basic information (static data) of a vessel has been compiled from various sources including the World Shipping Encyclopedia (Lloyd’s Fairplay, 2008) and Lloyd’s Register of Ships. The static data describes each vessel with over 200 variables such as identity number, nationality, date of building, tonnage. The second is a casualty dataset that comprises 8023 records covering the time period from 1993 to 2008, which is a compilation of data from World Casualty Statistics by Lloyd’s Register of Shipping (Lloyd’s Fairplay, 1993–2008) and the IMO. The World Casualty Statistics consists of 2552 casualty records and the IMO website provides 6876 casualty records from 1993 to 2008. By eliminating the duplicated observations, there are 8023 records in the final casualty dataset, which includes various descriptive variables, such as, accident records of collisions, contacts, fires and explosions, foundering, hull/machinery damage, and miscellaneous wrecks/stranding/groundings. The third is an inspection dataset comprising 370,000 inspection cases in 59 countries for the time period from January 1999 to December 2008. These countries are member States of three main Memoranda of Understanding (MoU) on Port State Control (PSC) under the coordination of the IMO, including the member states of China, Japan, India, France, the UK and Canada etc. 3. Methodology – a binary regression approach Maritime safety is a complex domain, which involves technological, human and organizational factors (Kristiansen, 2005). The consequence of maritime risk ranges from severe environmental damage to large scale loss of life. Numerous studies use safety records or the statistic methods to analyze incidents (Gaarder et al., 1997; Li and Cullinane, 2003; Soares and Teixeira, 2001; Kristiansen, 2005; Li and Zheng, 2008). However, statistic information describes only the past condition of the vessels. Based on the safety records of global vessels, this study develops a maritime safety index to describe the comparative safety level of any particular vessel. We cannot observe safety directly, but safety outputs such as accidents and incidents can provide information on the underlying distributions of probability of accidents. Then, we can analyze the probability of accidents as a measure of safety performance. Traditionally, the most common way to estimate the probability of accidents is accident frequency, which is regarded as the first type of method that addresses safety levels (Soares and Teixeira, 2001). The statistics concerning the frequency of accidents provide a crucial view of safety performance as in Romer et al. (1995). Most of these studies used accident frequency as a safety measure for examining the relationship between safety performance and a particular indicator, such as the age of vessels (Faragher et al., 1979; Cashman, 1977) and the flag of vessels (Li and Wonham, 1999; Pronce, 1990). There are also some studies using frequency of accidents, which are related to a particular type of vessels (Grabowski et al., 2007; Talley, 2001; Wang et al., 2005; Forsyth, 1991; Roberts and Marlow, 2002) or a particular business section (Mostafa, 2004). All previous studies provide useful insights into the characteristics of maritime accident. However, there are limitations to all those studies. First, statistics describing the relationship between indicators and characteristics of the accidents do not reveal the degree of the influence of the determining factors. Second, there are some specific criteria and assumptions in some studies, which cannot be easily verified by other sources (Romer et al., 1995). Last, safety records reveal only the past conditions of relevant vessels but cannot predict the occurrence of a future accident directly (Gaarder et al., 1997) especially when maritime accidents are typically very rare events (Chang and Yeh, 2004; Hockaday and Chatziioanou, 1986). Logistic regression has been proved to be a powerful modeling tool to predict the probability of occurrence of an accident, by fitting data to a logit function. In recent years, it has been suggested as an appropriate analytical technique to use for the multivariate modeling of categorical dependent variables (Uncles, 1987). There are several studies in the maritime domain using logistic regression model (Bergantina and Marlow, 1998; Jin et al., 2002; Jin and Thunberg, 2005). Given this background, the main aim of this paper is to investigate which factors and to what extent these factors in affecting a vessel’s risk level by using logistic regression method. Then, the predicted probability of risk can be transformed to maritime safety index using Eq. (1). Therefore, the maritime safety index can be regarded as a combination of risk indicators.
R¼1P
ð1Þ
The probability (P) of an accident to a vessel is parameterized as an exponential function of the vessel’s operating characteristics.
K.X. Li et al. / Transportation Research Part A 66 (2014) 75–87
77
There are two possible forms for y, accident (y = 1) or no accident (y = 0). Then, this can be regarded as a binomial model since the y is one of two alternatives. In binary regression, a latent variable y⁄ gets mapped onto a binominal variable y where y⁄ e (1, +1). While we cannot observe y⁄ directly, we can observe y such that:
y ¼ 1 accident; if y > 0
ð2Þ
y ¼ 0 nonaccident; if y 6 0
ð3Þ
Consider a random m-dimensional variable X = (x1, . . ., xm), each variable describes a certain property of the vessel, then we can define the latent variable as a function of X,
y ¼ Xb þ l
ð4Þ
The matrix contains independent variables such as age, size, flag, classification society or owner, and b represents a column vector of unknown parameters (Green, 2003). This gives:
EðyjXÞ ¼ Pðy ¼ 1jXÞ ¼ Pðy > 0jXÞ
ð5Þ
¼ Pðl > XbÞ ¼ 1 FðXbÞ
This function F can take different forms, and for this study we choose the logistic cumulative distribution function. The general model can therefore be written in the form
P¼
eXb 1 þ eXb
ð6Þ
Then the safety index can be obtained as:
R¼1P ¼
1 1 þ eXb
ð7Þ
1R ¼ Xb Ln R
ð8Þ
The safety index will fall in the range [0, 1] because P 2 [0, 1]. 1R is defined as the safety odds. The log safety odds, Ln 1R , is R R the linear predictor of the binary logistic model, which means that the log safety odds is the linear predictor of the index function. The higher the value of safety index, R, the better the safety level of the vessel will be. In order to be more intuitive, we use a Green-Yellow-Red list to indicate the safety level of each vessel. When the index falls in the range [0.9, 1], the vessel will find itself on the green list, meaning that the safety level is the highest and the probability of an accident is the lowest. When the index is less than 0.8, then the vessel will find itself on the red list, meaning that the safety level is the lowest, and the probability of an accident is the highest. When the value of the safety index lies (0.8–0.9), the vessel will find itself situated in the yellow list (see Fig. 1).
Dangerous area
Normal area
Safe area
Fig. 1. Overview of safety index.
78
K.X. Li et al. / Transportation Research Part A 66 (2014) 75–87
4. Risk indicators and hypotheses 4.1. Internal variables Indicator 1: vessel age Vessel age is one of the most important characteristics of vessels. Previous studies show that the relationship between vessel age and its safety level is negative. Cariou et al. (2008) and Li et al. (2009) noticed that the probability of belonging to the ‘‘always deficient’’ category is highest for vessels between 25 and 30 years old. Knapp and Franses, (2007b) also found that vessel age has a significant effect on casualties. Therefore, we proposed the following hypothesis: Hypothesis 1. With an increase of age, a vessel’s safety level is decreased. Indicator 2: vessel size Compared with small vessels, the maneuverability of large vessels is poor which increases the probability of accidents accordingly. Li et al. (2009) showed that the probability of a total-loss incident increases with the vessel size through analysis of total-loss incidents. Based on these, we propose that older vessels have a high level of accident. Hypothesis 2. With the increase of size, a vessel’s safety level decreases. Indicator 3: vessel type The division of vessel types for this study is based on that of Lloyd’s Fairplay (1993–2008)which includes general cargo & multipurpose, bulk, container, tanker, passenger vessel and others. The safety levels of different types of vessels are different as they carry different cargoes and navigate in different sea areas. Cariou et al. (2008) found that passenger vessels exhibit more deficiencies in comparison to other vessel categories. However, Knapp and Franses, (2007a) have shown that general cargo vessels seem to carry the highest risk. Therefore, we can propose following hypotheses for testing: Hypothesis 3. The safety levels of different types of vessels are different. Indicator 4: classification societies A classification society is a non-profit organization that sets rules and technical standards for the quality and integrity of vessels, and performs surveys to determine whether vessels are in compliance with the classification society’s rules and regulations, national laws and international conventions (Clark, 2009). Since classification is an absolute prerequisite for vessel registration and insurance cover (Clark, 1991), vessels cannot operate without being certified by a classification society. There are currently approximately 50 classification societies that provide maritime classification services, some of which are recognized within the industry to be of a better quality and standard than others. These classification societies with good reputations usually request their members to satisfy a series of strict quality requirements. In order to improve the safety level of vessels, the International Association of Classification Societies (IACS) recognized 10 member societies that enforce strict regulations. This means that vessels classified by these 10 IACS members have a high safety level. Knapp and Franses, (2007b) also showed that the quality of the classification society has a significant effect on the probability of casualties. In this study, we use 34 dummy variables to determine the relationship between the quality of classification societies and safety record of vessels. Hypothesis 4a. Different classification societies have different effects on the safety level. Hypothesis 4b. Compared with non-IACS members, the safety level of vessels classified by IACS members is higher. 4.2. External variables Indicator 5: navigation zone In accordance with the world casualty statistics (Lloyd’s Fairplay, 1993–2008), the oceans in the world are divided into 31 navigation zones (Fig. 2). Due to their unique geographical environment, different zones have different effects on the safety level of the vessels. We use 30 dummy variables to control the effect of navigation zones on safety. Hypothesis 5. Different navigation zones will have different effects on a vessel’s safety level. Indicator 6: vessel flags Li and Wonham (1999) found that open-registry vessels tend to be substandard vessels, and the safety record of developing maritime countries as a group is better than that of developed maritime countries. This study uses two dummy variables to control the impact of different registration flags on vessel safety. Hypothesis 6a. Different registry flags have different effects on a vessel’s safety level. Hypothesis 6b. Compared with open registry vessels, closed registry vessels have higher safety levels.
79
K.X. Li et al. / Transportation Research Part A 66 (2014) 75–87
Fig. 2. Geographical division of zone (Figure source: IHS Fairplay World Casualty Statistics).
Indicator 7: time and season The passing of time affects the controls for certain conditions – such as technological changes, regulations and congestion. Rose (1990) and Li et al. (2009) both showed that there has been a generally declining trend in accidents through time, mirroring the substantial decline in aggregate accident rates over the period of time. Thus, we have Hypothesis 7a. Hypothesis 7a. Time has an effect on the safety level. Similarly, different seasons of the year generate different weather conditions to navigation f for vessels. Thus, the probability of accident is different in different seasons, which can be derived from the accident records. Thus, we have Hypothesis 7b.
Table 1 List of variables. Variables Vessel characteristics X1 X2 X40–X154 Vessel type X3 X4 X5 X6 X7 Vessel classification X8 X9 X10 X11 X12 X13 X14 X15 X16 X17 X18 X19–X39 External variables X155–X184 X185 X186 X187 X188 X189 X190–X255
Measurement
Variable type
Vessel age in years Vessel size in GT Flag state at the time of accident
Continuous Continuous Dummy
1 1 1 1 1
Dummy Dummy Dummy Dummy Dummy
if if if if if
a a a a a
general cargo vessel, 0 otherwise bulker vessel, 0 otherwise container vessel, 0 otherwise tanker vessel, 0 otherwise passenger vessel, 0 otherwise
1 if vessel is classified by a member of the IACS, 0 otherwise 1 if vessel is classified by the American Bureau of Shipping, 0 otherwise 1 if vessel is classified by the Bureau Veritas, 0 otherwise 1 if vessel is classified by the China Classification Society, 0 otherwise 1 if vessel is classified by the Det Norske Veritas, 0 otherwise 1 if vessel is classified by the Germanischer Lloyd, 0 otherwise 1 if vessel is classified by the Korean Register of Shipping, 0 otherwise 1 if vessel is classified by the Lloyds Registry, 0 otherwise 1 if vessel is classified by the Nippon Kaiji Kyokai, 0 otherwise 1 if vessel is classified by the Registro Italian Navale, 0 otherwise 1 if vessel is classified by the Russian Maritime Register of Shipping, 0 otherwise Other Classification Societies
Dummy Dummy Dummy Dummy Dummy Dummy Dummy Dummy Dummy Dummy Dummy Dummy
1 if the incident occurred in the region i, 0 otherwise (i = 1, . . ., 30) 1 if a vessel operator’s country is a developed country, 0 otherwise 1 if the country where the vessel’s operator is domiciled is on open registry, 0 otherwise 1 for spring, 0 for otherwise 1 for summer, 0 for otherwise 1 for fall, 0 for otherwise Years (1993–2008)
Dummy Dummy Dummy Dummy Dummy Dummy Dummy
80
K.X. Li et al. / Transportation Research Part A 66 (2014) 75–87
Hypothesis 7b. The season of year has an effect on the safety level. Table 1 is a summary of the variables that are used in the regressions. According to the hypotheses proposed, we have Model I and II to estimate a vessel’s safety level.
Model I
y ¼ Xb ¼ b0 þ
Model II y ¼ Xb ¼ b0 þ
P7
i¼1 bj xi P8 i¼1 bi xi
þ þ
P255
i¼9 bi xi
P255
i¼185 bi xi
By using the first model, we test the effect of classification societies and flags on vessels’ safety level. By using the second model, we examine the effects of IACS, open registry and developed countries on safety level. Following a logistic modeling estimation, we calculate the linear predictor by summing each parameter estimate times its respective data element across all terms, plus the constant.
^¼b ^0 þ ^ ¼ Xb y
X8
^
bx i¼1 i i
þ
X255
^
bx i¼185 i i
ð9Þ
Then the safety level Ri can be calculated by the formula below:
R¼
1 1 þ eX b^
ð10Þ
The casualties are an accumulated figure over a time frame of 16 years where the ships with no casualties are of the same time period but are not counted 16 times. Cramer et al. (1999) investigates various sizes of a reduced dataset of zeros and did not found significant change. Therefore, adding more data on the side of the zeros does not have a serious effect on the coefficients. This means that adding more data on the side of the zeros will not add much explanatory power to the models (Knapp and Franses, 2007a). 5. Empirical study and results The models are tested with the total datasets combined by several data sources and logistic regression procedure is available within the SAS software. Maximum likelihood estimation (MLE) is used to estimate the coefficients as it is suitable for handling with categorical dependent variables (Yadlin-Weintraub, 1991). We develop two separate models. In Model I, we test which classification society associates vessels with substandard. Similar test is done with flag. In Model II, we use two dummy variables to examine the effects of IACS and open registry flags on safety level. Table 2 is the model fit summary from logistic regressions of Models I and II. The test results indicate that the models developed by this study fit the data well. For example, for Model I the likelihood ratio statistic is 14,706, well above the critical value of 135.8 for significance of 0.01 with 100 degrees of freedom. Most variables are highly significant, with p-values less than or equal to 0.01. Since the results from Model I and Model II are similar and Model I includes all the variables, so we will explain the coefficient results base on Model I except when explaining Hypotheses 4b and 6b. Table 3 presents the partial effects of the coefficients and the significance level of interest variables. The sign of an estimated logistic coefficient suggests either an increase or decrease in safety level. The coefficient itself does not measure the marginal effect of each independent variable on the dependent variable. It can be derived utilizing the estimated coefficients and the following equation for continuous variables. ^
@Rk @½1=1 þ eRX b ¼ ¼ bk Rk ð1 Rk Þ @xk @xk
ð11Þ
For the dummy variables, let xk denote the dummy explanatory variable and x+ denote the other covariates at their means. The marginal effect due to a change of xk on the predicted probabilities of y is
R½y ¼ 1jxk ¼ 1; xþ R½y ¼ 1jxk ¼ 0; xþ
ð12Þ
The results of marginal effect are listed in the last column of Table 3.
Table 2 Model fit summary.
Observation Number of accidents Number of non-accidents Likelihood ratio AIC Schwarz criterion
Model I
Model II
127,073 6930 120,143 19,543 39,455 41,249
127,073 6930 120,143 22,013 44,108 44,508
81
K.X. Li et al. / Transportation Research Part A 66 (2014) 75–87 Table 3 Summary of main variables and their significance: vessel Characteristics. Variable
Variable label
Variable type
Coefficient
P-value
Marginal probability
x1 x2 x3 x4 x5 x6 x7
Age ln (gt) General Cargo Bulker Container Tanker Pass./Ferry
Continuous Continuous Dummy Dummy Dummy Dummy Dummy
0.03 0.20 1.14 0.33 0.30 0.17 0.70
0.000 0.000 0.000 0.000 0.000 0.006 0.000
0.001 0.008 0.045 0.013 0.012 0.007 0.028
Result relating to Indicator 1: vessel age The results of Model I suggest that an increase in vessel age is associated with an increase in the vessel safety level, which is contrary to our expectations, so Hypothesis 1 is rejected. An increase by 1 year in vessel age leads to an increase by 0.001 in its safety level. This may be a reflection of the fact that the survival vessels are proved to be quality or well-maintained ones (Cariou et al., 2008). Result relating to Indicator 2: vessel size The second variable, which is measured by the gross tonnage, reveals that the safety level decreases as the vessel sizes increase.
b2 ¼
DR DR ¼ Dlnx2 Dx2 =x2
ð13Þ
So
DR ¼ b2 Dx2 =x2
ð14Þ
The marginal effect of X2 suggests that the safety level decreases by 0.008% with the gross tonnage increase of 1%. Result relating to Indicator 3: vessel type The coefficients for the vessel type dummy variables reveal that safety levels vary with vessel type groups. The group of general cargo has the largest positive marginal effect on the risk level, followed by the passenger vessels. This means that the general cargo vessel is the most risky one, and it is followed by passenger vessels. The tanker group has the smallest marginal effect, so the safety level of this group is higher (compared with the base group of other types). The marginal effects are plotted along with the safety level in Figs. 3–8, using results from model I. These figures can help visualize the effect and give a better interpretation. The marginal effects of age on the safety level are obtained as a function of the age, at the mean of other variables, which are calculated using Eqs. (4) and (6). As shown in Fig. 3, we can conclude that safety levels are lower for younger vessels and general cargo vessels. Result relating to Indicator 4: classification society From Table A1, we can see that the coefficients for almost all of the classification are significant. This means that the safety levels are different among different classification societies, so Hypothesis 4a is accepted. Vessels registered with the American Bureau of Shipping and Nippon Kaiji have higher safety levels compared with other classification societies.
Fig. 3. Effect of vessel age and vessel type on the safety level.
82
K.X. Li et al. / Transportation Research Part A 66 (2014) 75–87
The Registro Cubano has the largest marginal effect on the safety level, followed by the Vietnamese Register. The results of the coefficients for classification societies are reported in Appendix Table A1. The results of Model II (Table 4) suggest that compared with non-IACS members, vessels classified by IACS members are safer, so Hypothesis 4b is accepted. The negative value implies that a classification society with a good reputation ensures that the security of its vessels is sufficient for maritime adventure (Figs. 5 and 6). That is because they enforce strict regulations so as to improve the security level of their vessels. Result relating to Indicator 5: navigation zones Most of the coefficients (Appendix Table A2) of the navigation zone variables are significant, except for zone 2, zone 14, zone 19, and zone 23, so Hypothesis 5 is accepted. The most dangerous zones include zone 3, 6, 8, 9, 10, 16, 17, and 18. Zone 6 is the Suez Canal. At present, about 25,000 vessels pass through the canal every year. So this zone has an important effect on the maritime safety level. The coefficients of two special zones (X166, X167) reveal that if the vessel voyage is via the Southern China Sea (zone 12) or the Eastern Asian (zone 13), these have no significant effect on decreasing the safety level. The large number of accidents may be because the number of vessel voyages via these two zones is larger. Result relating to Indicator 6: vessel registers The coefficients (Appendix Tables A3-1 and A3-2) for the vessel flag dummy variables (X40–X153) indicate that the safety level is higher than the category ‘‘others’’ when the vessels are registered in Japan, China, Netherlands, India, Mexico, Brazil, Australia, Taiwan China, Sweden, Morocco, Belgium, South Africa and Ireland. The safety level is lower when the vessel is
Fig. 4. Effect of tonnage and vessel type on the safety level.
Fig. 5. Effect of vessel age and IACS on the safety level.
K.X. Li et al. / Transportation Research Part A 66 (2014) 75–87
Fig. 6. Effect of Tonnage and IACS on the safety level.
Fig. 7. Effect of vessel age and open registry on the safety level.
Fig. 8. Effect of Tonnage and open registry on the safety level.
83
84
K.X. Li et al. / Transportation Research Part A 66 (2014) 75–87
Table 4 Partial results of Model II. Variable
Variable label
Variable type
Coefficient
P-value
Marginal probability
x8 X185 X186
IACS member Open Registry Countries Developed Countries
Dummy Dummy Dummy
1.538921 0.373724 0.024993
0.0001 0.0001 0.0001
0.0690 0.0167 0.0011
Table 5 Safety level of different group.
registered in Panama, United States, Russia, Norway, Malta, Greece, Italy, Bahamas, Spain, Turkey, Honduras and Canada. This study accepts Hypothesis 6a. As expected, the sign of the coefficient for the open registry (X185) is positive, where Hypothesis 6b is accepted. This means that vessels registered in open registry countries have lower safety levels (Figs. 7 and 8) as the authorities are considered as no intension or resource to excise effective control over the vessels registered under their flags. Result relating to Indicator 7: time and season The incident season (X187–X189) does not seem to have a significant effect on the vessel safety levels, and the coefficients of the year (X190–X205) are not significant either. It means that some conditions, such as technological changes, regulations and congestion, vary through time but they have no significant effect on the safety level of vessels. So Hypotheses 7a and 7b are rejected. The season and year cannot affect a vessel’s safety level. In summary, the vessel safety level is primarily influenced by a vessel’s age, size, type, classification society and flag. Using the above results, when a vessel’s characteristic data is available, the probability of the vessel being involved in an accident and its safety index can be predicted using Eqs. (9) and (11). For each vessel we can obtain an individual safety index. For a group of vessels with the same characteristics, we can obtain an overall safety index. Based on the index, we can put the vessel in a different safety list (Table 5).
6. Conclusion We cannot observe risk directly, but we can measure the probability of the occurrence of an accident through historical data on safety indicators, such as vessel age, size, type, flag and classification society. In this study, we establish a comprehensive dataset and develop a safety index. The safety index can be used to generate a relative risk score for each existing vessel, and it serves as a vessel safety benchmarking and management tool for various users. For example, it can be used by port authorities to determine whether an on board inspection is needed for vessels calling at their ports, in order to prevent oil pollution and accidents within their territorial waters; it can be used by insurers to determine the premium level prior to an insurance contract. It also helps vessel operators understand their vessels’ relative safety strengths and weaknesses and identify functional areas for safety improvement. In particular, the index can help PSC authorities focus on the major safety weaknesses of an individual vessel in specific divisions when making decisions as to which should be inspected. The safety
85
K.X. Li et al. / Transportation Research Part A 66 (2014) 75–87
index system will continue to be improved, with dynamic information regarding updated vessel movements and management changes, as well as accurate inspection and survey reports.
Acknowledgement This research is partially supported by The National Natural Science Foundation of China 71302052.
Appendix See Tables A1–A3.
Table A1 Vessel characteristic results. Variable
Variable label
Coefficient
P-value
Marginal probability
x9 x10 x11 x12 x13 x14 x15 x16 x17 x18 x19 x20 x21 x22 x23 x24 x25 x26 x27 x28 x29 x30 x31 x32 x33 x34 x35 x36 x37 x38
American Bureau of Shipping Bureau Veritas China Classification Society Det Norske Veritas Germanischer Lloyd South Korean Register Lloyds Register Nippon Kaiji Registro Italiano Russian Register Biro Klass Indonesia Bulgarski Koraben Registar China Corp Register Croation Register Hellenic Register Indian Register Yugoslavia Register Korea Classification Society Polish Register Turk Loydu Vietnamese Register Registro Cubano Rinave Portugesa Russian River Register East German Register Bulgarian Register of Shipping Isthmus Maritime Classification Society Isthmus Bureau of Shipping Maritime Lloyds-Georgia Romanian Register
2.21 1.58 0.88 1.85 1.64 1.42 1.61 2.00 1.02 1.22 1.36 1.06 0.01 1.66 0.27 1.95 0.31 0.63 0.24 0.23 2.48 3.81 0.86 3.33 0.27 6.00 1.88 3.72 4.69 4.04
0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.035 0.958 0.003 0.222 0.000 0.958 0.027 0.320 0.208 0.000 0.000 0.116 0.340 0.967 0.206 0.590 0.255 0.141 0.006
0.088 0.063 0.035 0.074 0.065 0.057 0.064 0.080 0.041 0.049 0.054 0.042 0.001 0.066 0.011 0.078 0.012 0.025 0.010 0.009 0.099 0.152 0.034 0.133 0.011 0.239 0.075 0.148 0.187 0.161
Table A2 Summary of main variables and their significance: zone. Variable
Coefficient
P-value
Marginal probability
Variable
Coefficient
P-value
Marginal probability
Zone1 Zone2 Zone3 Zone4 Zone5 Zone6 Zone7 Zone8 Zone9 Zone10 Zone11 Zone12 Zone13 Zone14 Zone15
0.81 0.02 11.88 1.11 1.57 2.72 1.30 1.53 1.14 2.63 0.41 1.35 0.49 11.27 0.69
0.000 0.862 0.203 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.009 0.000 0.000 0.263 0.001
0.032 0.001 0.473 0.044 0.063 0.108 0.052 0.061 0.045 0.105 0.016 0.054 0.020 0.449 0.027
Zone16 Zone17 Zone18 Zone19 Zone20 Zone21 Zone22 Zone23 Zone24 Zone25 Zone26 Zone27 Zone28 Zone29 Zone30
2.81 10.52 2.07 0.08 2.15 2.63 2.06 0.07 0.70 2.55 11.54 0.55 2.61 1.16 3.30
0.000 0.305 0.000 0.424 0.000 0.000 0.000 0.574 0.000 0.000 0.181 0.008 0.000 0.000 0.000
0.112 0.419 0.083 0.003 0.085 0.105 0.082 0.003 0.028 0.102 0.460 0.022 0.104 0.046 0.132
86
K.X. Li et al. / Transportation Research Part A 66 (2014) 75–87
Table A3-1 Vessel flag results. Variable
Coefficient
P-value
Marginal probability
Variable
Coefficient
P-value
Marginal probability
x40 x41 x42 x43 x44 x45 x46 x47 x48 x49 x50 x51 x52 x53 x54 x55 x56 x57 x58 x59 x60 x61 x62 x63 x64 x65 x66 x67 x68 x69 x70 x71 x72 x73 x74 x75 x76 x77 x78 x79
1.01 1.37 0.85 1.55 0.57 0.30 2.37 0.38 0.62 0.63 0.77 0.97 0.73 0.39 0.99 0.78 0.22 0.88 0.08 0.71 1.24 0.99 0.45 0.67 0.79 0.81 0.39 0.55 1.33 0.09 1.12 0.87 0.60 3.75 0.10 0.68 2.89 2.89 1.16 0.21
0.000 0.000 0.000 0.000 0.001 0.034 0.000 0.010 0.000 0.000 0.000 0.000 0.000 0.014 0.000 0.000 0.195 0.000 0.625 0.000 0.000 0.000 0.036 0.000 0.000 0.000 0.048 0.016 0.000 0.616 0.000 0.000 0.002 0.000 0.564 0.061 0.000 0.000 0.002 0.397
0.040 0.055 0.034 0.062 0.023 0.012 0.094 0.015 0.025 0.025 0.031 0.039 0.029 0.015 0.039 0.031 0.009 0.035 0.003 0.028 0.049 0.040 0.018 0.027 0.032 0.032 0.016 0.022 0.053 0.004 0.045 0.035 0.024 0.149 0.004 0.027 0.115 0.115 0.046 0.008
x80 x81 x82 x83 x84 x85 x86 x87 x88 x89 x90 x91 x92 x93 x94 x95 x96 x97 x98 x99 x100 x101 x102 x103 x104 x105 x106 x107 x108 x109 x110 x111 x112 x113 x114 x115 x116 x117 x118 x119
1.18 0.04 0.12 1.73 0.18 3.26 0.89 0.10 1.88 0.30 0.21 0.36 0.49 0.20 1.09 0.81 0.28 0.62 0.00 0.28 0.47 0.98 1.56 1.80 0.23 0.21 1.25 1.35 0.78 1.33 0.55 1.61 0.72 0.27 1.97 1.36 0.11 0.63 0.70 0.22
0.000 0.893 0.735 0.000 0.464 0.000 0.000 0.746 0.000 0.377 0.391 0.140 0.119 0.464 0.000 0.004 0.276 0.102 0.993 0.266 0.065 0.006 0.000 0.001 0.398 0.423 0.000 0.001 0.001 0.001 0.037 0.000 0.020 0.498 0.000 0.040 0.715 0.250 0.043 0.548
0.047 0.002 0.005 0.069 0.007 0.130 0.036 0.004 0.075 0.012 0.008 0.014 0.020 0.008 0.043 0.032 0.011 0.025 0.000 0.011 0.019 0.039 0.062 0.072 0.009 0.009 0.050 0.054 0.031 0.053 0.022 0.064 0.029 0.011 0.079 0.054 0.004 0.025 0.028 0.009
Table A3-2 Vessel flag results (Continue). Variable
Coefficient
P-value
Marginal probability
Variable
Coefficient
P-value
Marginal probability
x120 x121 x122 x123 x124 x125 x126 x127 x128 x129 x130 x131 x132 x133 x134 x135 x136
0.19 4.87 1.60 1.11 2.33 0.74 0.08 0.80 0.60 1.02 1.17 12.69 1.05 0.95 0.87 0.15 1.14
0.720 0.329 0.002 0.069 0.000 0.022 0.907 0.018 0.093 0.001 0.018 0.182 0.003 0.163 0.012 0.730 0.003
0.007 0.194 0.064 0.044 0.093 0.030 0.003 0.032 0.024 0.041 0.047 0.506 0.042 0.038 0.035 0.006 0.045
x137 x138 x139 x140 x141 x142 x143 x144 x145 x146 x147 x148 x149 x150 x151 x152 x153
1.18 0.26 0.64 0.07 0.09 1.73 1.11 0.12 0.95 5.93 0.19 0.43 2.03 1.54 1.33 2.40 0.89
0.008 0.668 0.145 0.908 0.887 0.000 0.000 0.804 0.010 0.315 0.591 0.258 0.000 0.012 0.165 0.000 0.018
0.047 0.010 0.026 0.003 0.003 0.069 0.044 0.005 0.038 0.236 0.008 0.017 0.081 0.061 0.053 0.096 0.035
K.X. Li et al. / Transportation Research Part A 66 (2014) 75–87
87
References Bergantina, A., Marlow, P., 1998. Factors influencing the choice of flag: empirical evidence. Marit. Policy Manag. 25 (2), 157–174. Cariou, P., Mejia Jr., M.Q., Wolff, F.C., 2008. On the effectiveness of port state control inspections. Transp. Res. E 44 (3), 491–503. Cashman, J.P., 1977. Analysis of world merchant ship losses 1967–1975, paper no. 2 in Safety At Sea, Proceedings, London, 23–27 May. Chang, Y.H., Yeh, C.H., 2004. A new airline safety index. Transp. Res. B 38 (4), 369–383. Clark, P.D., 1991. Vessel classification isn’t insurance policy, appeals court rules. Sea Law, 3, http://navlaw.com/articles/index.html. Clark, P.D., 2009. An American admiralty law viewpoint on the changing role of classification societies. Sea Law, 5, http://navlaw.com/articles/index.html.. Cramer, M., Franses, P.H., Slagter, E., 1999. Censored Regression Analysis in Large Samples with Many Zero Observations. Econometric Institute Research Report 9939/A. Faragher, W. E., Pizzo, J. T., Reinburg, L. R., Reiter, A. D., Graham, W. C., 1979. Deep water ports approach/exit hazard and risk assessment. Planning Research Corporation. McLean, VA. U.S. Coast Guard, Report No. CG-D-6-79. Forsyth, C.J., 1991. Factors affecting tanker safety. Marit. Policy Manag. 18 (3), 313–319. Gaarder, S., Rognstad, K., Olofsson, M., 1997. The impact of human element in maritime risk management. In: Guedes Soares, C. (Ed.), Advances in Safety and Reliability. Pergamon, Oxford, pp. 857–868. Grabowski, M., Ayyalasomayajula, P., Merrick, J., Mccafferty, D., 2007. Accident precursors and safety nets: leading indicators of tanker operations safety. Marit. Policy Manag. 34 (5), 405–425. Green, W.H., 2003. Econometric Analysis, 5th ed. Prentice Hall, Engle, Cliffs, NJ. Hockaday, S.L.M., Chatziioanou, A., 1986. An analytical method for aircraft collision risk estimation. Transp. Res. B 20 (5), 415–428. Jin, D., Thunberg, E., 2005. An analysis of fishing vessel accidents in fishing areas off the northeastern United States. Saf. Sci. 43 (8), 523–540. Jin, D., Kite-Powell, H.L., Thunberg, E., Solow, A.R., Talley, W.K., 2002. A model of fishing vessel accident probability. J. Saf. Res. 33 (4), 497–510. Knapp, S., Franses, P.H., 2007a. Econometric analysis on the effect of port state control inspections on the probability of casualty. Can targeting of substandard vessels for inspections be improved? Marine Policy 31 (4), 550–563. Knapp, S., Franses, P.H., 2007b. A global view on port state control – econometric analysis of the differences across port state control regimes. Marit. Policy Manag. 34 (5), 453–483. Kristiansen, S., 2005. Maritime Transportation: Safety Management and Risk Analysis. Elsevier, Oxford. Li, K.X., Cullinane, K., 2003. An economic approach to maritime risk management and safety regulation. Marit. Econ. Logist. 5 (3), 268–284. Li, K.X., Wonham, J., 1999. Who is safe and who is at risk: a study of 20-year-record on accident total loss in different flags. Marit. Policy Manag. 26 (2), 137– 144. Li, K.X., Zheng, H., 2008. Enforcement of law by the Port State Control (PSC). Marit. Policy Manag. 35 (1), 61–71. Li, K.X., Wang, Y., Min, J., 2009. Quantitative analysis of materiality in maritime insurance. Marit. Policy Manag. 36 (5), 437–455. Li, K.X., Yin, J., Bang, H.S., Yang, Z., Wang, J., 2014. Bayesian network with quantitative input for maritime risk analysis. Transportmetrica A 10 (2), 89–118. Lloyd’s Fairplay, 1993–2008. World casualty statistics. Annual Report. Lloyd’s Fairplay, 2008. World Shipping Encyclopedia. Version 2008. Mostafa, M.M., 2004. Forecasting the Suez Canal traffic: a neural network analysis. Marit. Policy Manag. 31 (2), 139–156. Pronce, P.V., 1990. An analysis of maritime total losses worldwide and for selected flags. Marit. Technol. 27 (2), 114–116. Roberts, S.E., Marlow, P.B., 2002. Casualties in dry bulk shipping (1963–1996). Marine Policy 26 (6), 437–450. Romer, H., Petersen, H.J.S., Haastrup, P., 1995. Maritime accident frequencies: review and recent empirical results. J. Navigation 48 (3), 410–424. Rose, N.L., 1990. Profitability and product quality: economic determinants of airline safety performance. J. Polit. Econ. 98 (5), 944–964. Soares, C.G., Teixeira, A.P., 2001. Risk assessment in maritime transportation. Reliab. Eng. Syst. Saf. 74 (3), 299–309. Talley, W.K., 2001. Determinants of the property damage costs of bulk barge accidents. Marit. Policy Manag. 28 (2), 175–186. Uncles, M.D., 1987. A beta-logistic model of mode choice: goodness of fit and intertemporal dependence. Transp. Res. B 21 (3), 195–205. Wang, J., Pillay, A., Kwon, Y.S., Wall, A.D., Loughran, C.G., 2005. An analysis of fishing vessel accidents. Accid. Anal. Prev. 37 (6), 1019–1024. Yadlin-Weintraub, M., 1991. Some procedures for the statistical and empirical analyses of probabilistic discrete decisions models. Transp. Res. B 25 (4), 215– 236. Yap, W.Y., Lam, J.S.L., 2013. 80 Million-TEU container port? Sustainability issues in port and coastal development. Ocean Coast. Manag. 71, 13–25. Yin, J., Fan, L., Yang, Z., Li, K.X., 2014. Slow steaming of liner trade: its economic and environmental impacts. Marit. Policy Manag. 41 (2), 149–158.
Contents lists available at ScienceDirect
Transportation Research Part A journal homepage: www.elsevier.com/locate/tra
Ship safety index Kevin X. Li a, Jingbo Yin b,⇑, Lixian Fan c a b c
Dept. of International Logistics, Chung-Ang University, South Korea Dept. of International Shipping, Shanghai Jiao Tong University, China School of Management, Shanghai University, China
a r t i c l e
i n f o
Article history: Received 3 May 2012 Received in revised form 31 March 2014 Accepted 25 April 2014
Keywords: Maritime safety Binary regression Safety index Risk rank system Safety database
a b s t r a c t This paper develops a new quantitative safety index for each worldwide sea-going vessel based on their condition information and safety records. The safety index can generate a relative risk score using binary logistic regression method and a dataset with both static and dynamic information covering over 90% of the world sea-going merchant fleet. It has a widely potential usage for both industry and academic research, e.g., for port authorities to determine whether an on board inspection is needed; for insurers to determine premium rate; and for shipowners to identify functional areas for repair and maintenance. Ó 2014 Elsevier Ltd. All rights reserved.
1. Introduction There are a total of around 120,000 sea-going commercial vessels of over 100 dead weight tonnage (dwt) sailing around the world. With an increasing awareness of environmental protection and safety issues, maritime authorities around the world have been making more rigorous efforts than ever before to promote safety and security at sea (Li and Zheng, 2008; Li et al., 2014; Yin et al., 2014). Since 1993, the International Maritime Organization (IMO) has adopted the International Safety Management Code (ISM) as a minimum statutory requirement for vessel operators to establish, implement and maintain their safety management systems. Although this program is implemented as a regulatory to oversee the safety activities of the vessels at the operational level, it cannot be used to measure and predict the safety level of each vessel. To examine comparative safety levels for safety management and improvement purposes, an evaluation mechanism for measuring the overall safety level of the vessels is needed. In this paper, we establish a new quantitative safety index to serve as a proactive evaluation mechanism to measure the safety level of a vessel, which provides useful information for decision making on safety management and improvements. The safety index, using both static and dynamic data that have been collected from various sources, forms the most comprehensive data set to date on vessel information (for details see Section 2). We investigate the effects of various risk factors using multivariate logistic regression modeling, and assess how the various factors simultaneously affect a vessel’s safety level. The new safety index can be used as a safety benchmark for both industry and academic research. For example, it can be used by port authorities to determine whether the right of access to territorial waters or ports should be granted, or whether on board inspections are required for vessels calling at their ports, in order to prevent oil pollution and accidents within their territorial waters (Yap and Lam, 2013). It can be used by insurers to determine the premium rate prior to an insurance
⇑ Corresponding author. Tel.: +86 021-66137696. E-mail address: [email protected] (J. Yin). http://dx.doi.org/10.1016/j.tra.2014.04.016 0965-8564/Ó 2014 Elsevier Ltd. All rights reserved.
76
K.X. Li et al. / Transportation Research Part A 66 (2014) 75–87
contract being concluded, or by any party who is interested in the safety and quality of the vessel and who needs to check the safety index before engaging in any business with the vessel. The safety index can provide useful information for operators to understand the relative safety strengths and weaknesses in terms of manageable safety characteristics, and to identify functional areas for safety improvement and maintenance. 2. Data description For this study, a total dataset with three sub-datasets has been built. The total dataset is a combination of accounting for about 130,000 vessels, including information about 10,000 lost vessels and 120,000 existing vessels, counting more than 90% of the worldwide commercial fleet. The first sub-dataset containing the basic information (static data) of a vessel has been compiled from various sources including the World Shipping Encyclopedia (Lloyd’s Fairplay, 2008) and Lloyd’s Register of Ships. The static data describes each vessel with over 200 variables such as identity number, nationality, date of building, tonnage. The second is a casualty dataset that comprises 8023 records covering the time period from 1993 to 2008, which is a compilation of data from World Casualty Statistics by Lloyd’s Register of Shipping (Lloyd’s Fairplay, 1993–2008) and the IMO. The World Casualty Statistics consists of 2552 casualty records and the IMO website provides 6876 casualty records from 1993 to 2008. By eliminating the duplicated observations, there are 8023 records in the final casualty dataset, which includes various descriptive variables, such as, accident records of collisions, contacts, fires and explosions, foundering, hull/machinery damage, and miscellaneous wrecks/stranding/groundings. The third is an inspection dataset comprising 370,000 inspection cases in 59 countries for the time period from January 1999 to December 2008. These countries are member States of three main Memoranda of Understanding (MoU) on Port State Control (PSC) under the coordination of the IMO, including the member states of China, Japan, India, France, the UK and Canada etc. 3. Methodology – a binary regression approach Maritime safety is a complex domain, which involves technological, human and organizational factors (Kristiansen, 2005). The consequence of maritime risk ranges from severe environmental damage to large scale loss of life. Numerous studies use safety records or the statistic methods to analyze incidents (Gaarder et al., 1997; Li and Cullinane, 2003; Soares and Teixeira, 2001; Kristiansen, 2005; Li and Zheng, 2008). However, statistic information describes only the past condition of the vessels. Based on the safety records of global vessels, this study develops a maritime safety index to describe the comparative safety level of any particular vessel. We cannot observe safety directly, but safety outputs such as accidents and incidents can provide information on the underlying distributions of probability of accidents. Then, we can analyze the probability of accidents as a measure of safety performance. Traditionally, the most common way to estimate the probability of accidents is accident frequency, which is regarded as the first type of method that addresses safety levels (Soares and Teixeira, 2001). The statistics concerning the frequency of accidents provide a crucial view of safety performance as in Romer et al. (1995). Most of these studies used accident frequency as a safety measure for examining the relationship between safety performance and a particular indicator, such as the age of vessels (Faragher et al., 1979; Cashman, 1977) and the flag of vessels (Li and Wonham, 1999; Pronce, 1990). There are also some studies using frequency of accidents, which are related to a particular type of vessels (Grabowski et al., 2007; Talley, 2001; Wang et al., 2005; Forsyth, 1991; Roberts and Marlow, 2002) or a particular business section (Mostafa, 2004). All previous studies provide useful insights into the characteristics of maritime accident. However, there are limitations to all those studies. First, statistics describing the relationship between indicators and characteristics of the accidents do not reveal the degree of the influence of the determining factors. Second, there are some specific criteria and assumptions in some studies, which cannot be easily verified by other sources (Romer et al., 1995). Last, safety records reveal only the past conditions of relevant vessels but cannot predict the occurrence of a future accident directly (Gaarder et al., 1997) especially when maritime accidents are typically very rare events (Chang and Yeh, 2004; Hockaday and Chatziioanou, 1986). Logistic regression has been proved to be a powerful modeling tool to predict the probability of occurrence of an accident, by fitting data to a logit function. In recent years, it has been suggested as an appropriate analytical technique to use for the multivariate modeling of categorical dependent variables (Uncles, 1987). There are several studies in the maritime domain using logistic regression model (Bergantina and Marlow, 1998; Jin et al., 2002; Jin and Thunberg, 2005). Given this background, the main aim of this paper is to investigate which factors and to what extent these factors in affecting a vessel’s risk level by using logistic regression method. Then, the predicted probability of risk can be transformed to maritime safety index using Eq. (1). Therefore, the maritime safety index can be regarded as a combination of risk indicators.
R¼1P
ð1Þ
The probability (P) of an accident to a vessel is parameterized as an exponential function of the vessel’s operating characteristics.
K.X. Li et al. / Transportation Research Part A 66 (2014) 75–87
77
There are two possible forms for y, accident (y = 1) or no accident (y = 0). Then, this can be regarded as a binomial model since the y is one of two alternatives. In binary regression, a latent variable y⁄ gets mapped onto a binominal variable y where y⁄ e (1, +1). While we cannot observe y⁄ directly, we can observe y such that:
y ¼ 1 accident; if y > 0
ð2Þ
y ¼ 0 nonaccident; if y 6 0
ð3Þ
Consider a random m-dimensional variable X = (x1, . . ., xm), each variable describes a certain property of the vessel, then we can define the latent variable as a function of X,
y ¼ Xb þ l
ð4Þ
The matrix contains independent variables such as age, size, flag, classification society or owner, and b represents a column vector of unknown parameters (Green, 2003). This gives:
EðyjXÞ ¼ Pðy ¼ 1jXÞ ¼ Pðy > 0jXÞ
ð5Þ
¼ Pðl > XbÞ ¼ 1 FðXbÞ
This function F can take different forms, and for this study we choose the logistic cumulative distribution function. The general model can therefore be written in the form
P¼
eXb 1 þ eXb
ð6Þ
Then the safety index can be obtained as:
R¼1P ¼
1 1 þ eXb
ð7Þ
1R ¼ Xb Ln R
ð8Þ
The safety index will fall in the range [0, 1] because P 2 [0, 1]. 1R is defined as the safety odds. The log safety odds, Ln 1R , is R R the linear predictor of the binary logistic model, which means that the log safety odds is the linear predictor of the index function. The higher the value of safety index, R, the better the safety level of the vessel will be. In order to be more intuitive, we use a Green-Yellow-Red list to indicate the safety level of each vessel. When the index falls in the range [0.9, 1], the vessel will find itself on the green list, meaning that the safety level is the highest and the probability of an accident is the lowest. When the index is less than 0.8, then the vessel will find itself on the red list, meaning that the safety level is the lowest, and the probability of an accident is the highest. When the value of the safety index lies (0.8–0.9), the vessel will find itself situated in the yellow list (see Fig. 1).
Dangerous area
Normal area
Safe area
Fig. 1. Overview of safety index.
78
K.X. Li et al. / Transportation Research Part A 66 (2014) 75–87
4. Risk indicators and hypotheses 4.1. Internal variables Indicator 1: vessel age Vessel age is one of the most important characteristics of vessels. Previous studies show that the relationship between vessel age and its safety level is negative. Cariou et al. (2008) and Li et al. (2009) noticed that the probability of belonging to the ‘‘always deficient’’ category is highest for vessels between 25 and 30 years old. Knapp and Franses, (2007b) also found that vessel age has a significant effect on casualties. Therefore, we proposed the following hypothesis: Hypothesis 1. With an increase of age, a vessel’s safety level is decreased. Indicator 2: vessel size Compared with small vessels, the maneuverability of large vessels is poor which increases the probability of accidents accordingly. Li et al. (2009) showed that the probability of a total-loss incident increases with the vessel size through analysis of total-loss incidents. Based on these, we propose that older vessels have a high level of accident. Hypothesis 2. With the increase of size, a vessel’s safety level decreases. Indicator 3: vessel type The division of vessel types for this study is based on that of Lloyd’s Fairplay (1993–2008)which includes general cargo & multipurpose, bulk, container, tanker, passenger vessel and others. The safety levels of different types of vessels are different as they carry different cargoes and navigate in different sea areas. Cariou et al. (2008) found that passenger vessels exhibit more deficiencies in comparison to other vessel categories. However, Knapp and Franses, (2007a) have shown that general cargo vessels seem to carry the highest risk. Therefore, we can propose following hypotheses for testing: Hypothesis 3. The safety levels of different types of vessels are different. Indicator 4: classification societies A classification society is a non-profit organization that sets rules and technical standards for the quality and integrity of vessels, and performs surveys to determine whether vessels are in compliance with the classification society’s rules and regulations, national laws and international conventions (Clark, 2009). Since classification is an absolute prerequisite for vessel registration and insurance cover (Clark, 1991), vessels cannot operate without being certified by a classification society. There are currently approximately 50 classification societies that provide maritime classification services, some of which are recognized within the industry to be of a better quality and standard than others. These classification societies with good reputations usually request their members to satisfy a series of strict quality requirements. In order to improve the safety level of vessels, the International Association of Classification Societies (IACS) recognized 10 member societies that enforce strict regulations. This means that vessels classified by these 10 IACS members have a high safety level. Knapp and Franses, (2007b) also showed that the quality of the classification society has a significant effect on the probability of casualties. In this study, we use 34 dummy variables to determine the relationship between the quality of classification societies and safety record of vessels. Hypothesis 4a. Different classification societies have different effects on the safety level. Hypothesis 4b. Compared with non-IACS members, the safety level of vessels classified by IACS members is higher. 4.2. External variables Indicator 5: navigation zone In accordance with the world casualty statistics (Lloyd’s Fairplay, 1993–2008), the oceans in the world are divided into 31 navigation zones (Fig. 2). Due to their unique geographical environment, different zones have different effects on the safety level of the vessels. We use 30 dummy variables to control the effect of navigation zones on safety. Hypothesis 5. Different navigation zones will have different effects on a vessel’s safety level. Indicator 6: vessel flags Li and Wonham (1999) found that open-registry vessels tend to be substandard vessels, and the safety record of developing maritime countries as a group is better than that of developed maritime countries. This study uses two dummy variables to control the impact of different registration flags on vessel safety. Hypothesis 6a. Different registry flags have different effects on a vessel’s safety level. Hypothesis 6b. Compared with open registry vessels, closed registry vessels have higher safety levels.
79
K.X. Li et al. / Transportation Research Part A 66 (2014) 75–87
Fig. 2. Geographical division of zone (Figure source: IHS Fairplay World Casualty Statistics).
Indicator 7: time and season The passing of time affects the controls for certain conditions – such as technological changes, regulations and congestion. Rose (1990) and Li et al. (2009) both showed that there has been a generally declining trend in accidents through time, mirroring the substantial decline in aggregate accident rates over the period of time. Thus, we have Hypothesis 7a. Hypothesis 7a. Time has an effect on the safety level. Similarly, different seasons of the year generate different weather conditions to navigation f for vessels. Thus, the probability of accident is different in different seasons, which can be derived from the accident records. Thus, we have Hypothesis 7b.
Table 1 List of variables. Variables Vessel characteristics X1 X2 X40–X154 Vessel type X3 X4 X5 X6 X7 Vessel classification X8 X9 X10 X11 X12 X13 X14 X15 X16 X17 X18 X19–X39 External variables X155–X184 X185 X186 X187 X188 X189 X190–X255
Measurement
Variable type
Vessel age in years Vessel size in GT Flag state at the time of accident
Continuous Continuous Dummy
1 1 1 1 1
Dummy Dummy Dummy Dummy Dummy
if if if if if
a a a a a
general cargo vessel, 0 otherwise bulker vessel, 0 otherwise container vessel, 0 otherwise tanker vessel, 0 otherwise passenger vessel, 0 otherwise
1 if vessel is classified by a member of the IACS, 0 otherwise 1 if vessel is classified by the American Bureau of Shipping, 0 otherwise 1 if vessel is classified by the Bureau Veritas, 0 otherwise 1 if vessel is classified by the China Classification Society, 0 otherwise 1 if vessel is classified by the Det Norske Veritas, 0 otherwise 1 if vessel is classified by the Germanischer Lloyd, 0 otherwise 1 if vessel is classified by the Korean Register of Shipping, 0 otherwise 1 if vessel is classified by the Lloyds Registry, 0 otherwise 1 if vessel is classified by the Nippon Kaiji Kyokai, 0 otherwise 1 if vessel is classified by the Registro Italian Navale, 0 otherwise 1 if vessel is classified by the Russian Maritime Register of Shipping, 0 otherwise Other Classification Societies
Dummy Dummy Dummy Dummy Dummy Dummy Dummy Dummy Dummy Dummy Dummy Dummy
1 if the incident occurred in the region i, 0 otherwise (i = 1, . . ., 30) 1 if a vessel operator’s country is a developed country, 0 otherwise 1 if the country where the vessel’s operator is domiciled is on open registry, 0 otherwise 1 for spring, 0 for otherwise 1 for summer, 0 for otherwise 1 for fall, 0 for otherwise Years (1993–2008)
Dummy Dummy Dummy Dummy Dummy Dummy Dummy
80
K.X. Li et al. / Transportation Research Part A 66 (2014) 75–87
Hypothesis 7b. The season of year has an effect on the safety level. Table 1 is a summary of the variables that are used in the regressions. According to the hypotheses proposed, we have Model I and II to estimate a vessel’s safety level.
Model I
y ¼ Xb ¼ b0 þ
Model II y ¼ Xb ¼ b0 þ
P7
i¼1 bj xi P8 i¼1 bi xi
þ þ
P255
i¼9 bi xi
P255
i¼185 bi xi
By using the first model, we test the effect of classification societies and flags on vessels’ safety level. By using the second model, we examine the effects of IACS, open registry and developed countries on safety level. Following a logistic modeling estimation, we calculate the linear predictor by summing each parameter estimate times its respective data element across all terms, plus the constant.
^¼b ^0 þ ^ ¼ Xb y
X8
^
bx i¼1 i i
þ
X255
^
bx i¼185 i i
ð9Þ
Then the safety level Ri can be calculated by the formula below:
R¼
1 1 þ eX b^
ð10Þ
The casualties are an accumulated figure over a time frame of 16 years where the ships with no casualties are of the same time period but are not counted 16 times. Cramer et al. (1999) investigates various sizes of a reduced dataset of zeros and did not found significant change. Therefore, adding more data on the side of the zeros does not have a serious effect on the coefficients. This means that adding more data on the side of the zeros will not add much explanatory power to the models (Knapp and Franses, 2007a). 5. Empirical study and results The models are tested with the total datasets combined by several data sources and logistic regression procedure is available within the SAS software. Maximum likelihood estimation (MLE) is used to estimate the coefficients as it is suitable for handling with categorical dependent variables (Yadlin-Weintraub, 1991). We develop two separate models. In Model I, we test which classification society associates vessels with substandard. Similar test is done with flag. In Model II, we use two dummy variables to examine the effects of IACS and open registry flags on safety level. Table 2 is the model fit summary from logistic regressions of Models I and II. The test results indicate that the models developed by this study fit the data well. For example, for Model I the likelihood ratio statistic is 14,706, well above the critical value of 135.8 for significance of 0.01 with 100 degrees of freedom. Most variables are highly significant, with p-values less than or equal to 0.01. Since the results from Model I and Model II are similar and Model I includes all the variables, so we will explain the coefficient results base on Model I except when explaining Hypotheses 4b and 6b. Table 3 presents the partial effects of the coefficients and the significance level of interest variables. The sign of an estimated logistic coefficient suggests either an increase or decrease in safety level. The coefficient itself does not measure the marginal effect of each independent variable on the dependent variable. It can be derived utilizing the estimated coefficients and the following equation for continuous variables. ^
@Rk @½1=1 þ eRX b ¼ ¼ bk Rk ð1 Rk Þ @xk @xk
ð11Þ
For the dummy variables, let xk denote the dummy explanatory variable and x+ denote the other covariates at their means. The marginal effect due to a change of xk on the predicted probabilities of y is
R½y ¼ 1jxk ¼ 1; xþ R½y ¼ 1jxk ¼ 0; xþ
ð12Þ
The results of marginal effect are listed in the last column of Table 3.
Table 2 Model fit summary.
Observation Number of accidents Number of non-accidents Likelihood ratio AIC Schwarz criterion
Model I
Model II
127,073 6930 120,143 19,543 39,455 41,249
127,073 6930 120,143 22,013 44,108 44,508
81
K.X. Li et al. / Transportation Research Part A 66 (2014) 75–87 Table 3 Summary of main variables and their significance: vessel Characteristics. Variable
Variable label
Variable type
Coefficient
P-value
Marginal probability
x1 x2 x3 x4 x5 x6 x7
Age ln (gt) General Cargo Bulker Container Tanker Pass./Ferry
Continuous Continuous Dummy Dummy Dummy Dummy Dummy
0.03 0.20 1.14 0.33 0.30 0.17 0.70
0.000 0.000 0.000 0.000 0.000 0.006 0.000
0.001 0.008 0.045 0.013 0.012 0.007 0.028
Result relating to Indicator 1: vessel age The results of Model I suggest that an increase in vessel age is associated with an increase in the vessel safety level, which is contrary to our expectations, so Hypothesis 1 is rejected. An increase by 1 year in vessel age leads to an increase by 0.001 in its safety level. This may be a reflection of the fact that the survival vessels are proved to be quality or well-maintained ones (Cariou et al., 2008). Result relating to Indicator 2: vessel size The second variable, which is measured by the gross tonnage, reveals that the safety level decreases as the vessel sizes increase.
b2 ¼
DR DR ¼ Dlnx2 Dx2 =x2
ð13Þ
So
DR ¼ b2 Dx2 =x2
ð14Þ
The marginal effect of X2 suggests that the safety level decreases by 0.008% with the gross tonnage increase of 1%. Result relating to Indicator 3: vessel type The coefficients for the vessel type dummy variables reveal that safety levels vary with vessel type groups. The group of general cargo has the largest positive marginal effect on the risk level, followed by the passenger vessels. This means that the general cargo vessel is the most risky one, and it is followed by passenger vessels. The tanker group has the smallest marginal effect, so the safety level of this group is higher (compared with the base group of other types). The marginal effects are plotted along with the safety level in Figs. 3–8, using results from model I. These figures can help visualize the effect and give a better interpretation. The marginal effects of age on the safety level are obtained as a function of the age, at the mean of other variables, which are calculated using Eqs. (4) and (6). As shown in Fig. 3, we can conclude that safety levels are lower for younger vessels and general cargo vessels. Result relating to Indicator 4: classification society From Table A1, we can see that the coefficients for almost all of the classification are significant. This means that the safety levels are different among different classification societies, so Hypothesis 4a is accepted. Vessels registered with the American Bureau of Shipping and Nippon Kaiji have higher safety levels compared with other classification societies.
Fig. 3. Effect of vessel age and vessel type on the safety level.
82
K.X. Li et al. / Transportation Research Part A 66 (2014) 75–87
The Registro Cubano has the largest marginal effect on the safety level, followed by the Vietnamese Register. The results of the coefficients for classification societies are reported in Appendix Table A1. The results of Model II (Table 4) suggest that compared with non-IACS members, vessels classified by IACS members are safer, so Hypothesis 4b is accepted. The negative value implies that a classification society with a good reputation ensures that the security of its vessels is sufficient for maritime adventure (Figs. 5 and 6). That is because they enforce strict regulations so as to improve the security level of their vessels. Result relating to Indicator 5: navigation zones Most of the coefficients (Appendix Table A2) of the navigation zone variables are significant, except for zone 2, zone 14, zone 19, and zone 23, so Hypothesis 5 is accepted. The most dangerous zones include zone 3, 6, 8, 9, 10, 16, 17, and 18. Zone 6 is the Suez Canal. At present, about 25,000 vessels pass through the canal every year. So this zone has an important effect on the maritime safety level. The coefficients of two special zones (X166, X167) reveal that if the vessel voyage is via the Southern China Sea (zone 12) or the Eastern Asian (zone 13), these have no significant effect on decreasing the safety level. The large number of accidents may be because the number of vessel voyages via these two zones is larger. Result relating to Indicator 6: vessel registers The coefficients (Appendix Tables A3-1 and A3-2) for the vessel flag dummy variables (X40–X153) indicate that the safety level is higher than the category ‘‘others’’ when the vessels are registered in Japan, China, Netherlands, India, Mexico, Brazil, Australia, Taiwan China, Sweden, Morocco, Belgium, South Africa and Ireland. The safety level is lower when the vessel is
Fig. 4. Effect of tonnage and vessel type on the safety level.
Fig. 5. Effect of vessel age and IACS on the safety level.
K.X. Li et al. / Transportation Research Part A 66 (2014) 75–87
Fig. 6. Effect of Tonnage and IACS on the safety level.
Fig. 7. Effect of vessel age and open registry on the safety level.
Fig. 8. Effect of Tonnage and open registry on the safety level.
83
84
K.X. Li et al. / Transportation Research Part A 66 (2014) 75–87
Table 4 Partial results of Model II. Variable
Variable label
Variable type
Coefficient
P-value
Marginal probability
x8 X185 X186
IACS member Open Registry Countries Developed Countries
Dummy Dummy Dummy
1.538921 0.373724 0.024993
0.0001 0.0001 0.0001
0.0690 0.0167 0.0011
Table 5 Safety level of different group.
registered in Panama, United States, Russia, Norway, Malta, Greece, Italy, Bahamas, Spain, Turkey, Honduras and Canada. This study accepts Hypothesis 6a. As expected, the sign of the coefficient for the open registry (X185) is positive, where Hypothesis 6b is accepted. This means that vessels registered in open registry countries have lower safety levels (Figs. 7 and 8) as the authorities are considered as no intension or resource to excise effective control over the vessels registered under their flags. Result relating to Indicator 7: time and season The incident season (X187–X189) does not seem to have a significant effect on the vessel safety levels, and the coefficients of the year (X190–X205) are not significant either. It means that some conditions, such as technological changes, regulations and congestion, vary through time but they have no significant effect on the safety level of vessels. So Hypotheses 7a and 7b are rejected. The season and year cannot affect a vessel’s safety level. In summary, the vessel safety level is primarily influenced by a vessel’s age, size, type, classification society and flag. Using the above results, when a vessel’s characteristic data is available, the probability of the vessel being involved in an accident and its safety index can be predicted using Eqs. (9) and (11). For each vessel we can obtain an individual safety index. For a group of vessels with the same characteristics, we can obtain an overall safety index. Based on the index, we can put the vessel in a different safety list (Table 5).
6. Conclusion We cannot observe risk directly, but we can measure the probability of the occurrence of an accident through historical data on safety indicators, such as vessel age, size, type, flag and classification society. In this study, we establish a comprehensive dataset and develop a safety index. The safety index can be used to generate a relative risk score for each existing vessel, and it serves as a vessel safety benchmarking and management tool for various users. For example, it can be used by port authorities to determine whether an on board inspection is needed for vessels calling at their ports, in order to prevent oil pollution and accidents within their territorial waters; it can be used by insurers to determine the premium level prior to an insurance contract. It also helps vessel operators understand their vessels’ relative safety strengths and weaknesses and identify functional areas for safety improvement. In particular, the index can help PSC authorities focus on the major safety weaknesses of an individual vessel in specific divisions when making decisions as to which should be inspected. The safety
85
K.X. Li et al. / Transportation Research Part A 66 (2014) 75–87
index system will continue to be improved, with dynamic information regarding updated vessel movements and management changes, as well as accurate inspection and survey reports.
Acknowledgement This research is partially supported by The National Natural Science Foundation of China 71302052.
Appendix See Tables A1–A3.
Table A1 Vessel characteristic results. Variable
Variable label
Coefficient
P-value
Marginal probability
x9 x10 x11 x12 x13 x14 x15 x16 x17 x18 x19 x20 x21 x22 x23 x24 x25 x26 x27 x28 x29 x30 x31 x32 x33 x34 x35 x36 x37 x38
American Bureau of Shipping Bureau Veritas China Classification Society Det Norske Veritas Germanischer Lloyd South Korean Register Lloyds Register Nippon Kaiji Registro Italiano Russian Register Biro Klass Indonesia Bulgarski Koraben Registar China Corp Register Croation Register Hellenic Register Indian Register Yugoslavia Register Korea Classification Society Polish Register Turk Loydu Vietnamese Register Registro Cubano Rinave Portugesa Russian River Register East German Register Bulgarian Register of Shipping Isthmus Maritime Classification Society Isthmus Bureau of Shipping Maritime Lloyds-Georgia Romanian Register
2.21 1.58 0.88 1.85 1.64 1.42 1.61 2.00 1.02 1.22 1.36 1.06 0.01 1.66 0.27 1.95 0.31 0.63 0.24 0.23 2.48 3.81 0.86 3.33 0.27 6.00 1.88 3.72 4.69 4.04
0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.035 0.958 0.003 0.222 0.000 0.958 0.027 0.320 0.208 0.000 0.000 0.116 0.340 0.967 0.206 0.590 0.255 0.141 0.006
0.088 0.063 0.035 0.074 0.065 0.057 0.064 0.080 0.041 0.049 0.054 0.042 0.001 0.066 0.011 0.078 0.012 0.025 0.010 0.009 0.099 0.152 0.034 0.133 0.011 0.239 0.075 0.148 0.187 0.161
Table A2 Summary of main variables and their significance: zone. Variable
Coefficient
P-value
Marginal probability
Variable
Coefficient
P-value
Marginal probability
Zone1 Zone2 Zone3 Zone4 Zone5 Zone6 Zone7 Zone8 Zone9 Zone10 Zone11 Zone12 Zone13 Zone14 Zone15
0.81 0.02 11.88 1.11 1.57 2.72 1.30 1.53 1.14 2.63 0.41 1.35 0.49 11.27 0.69
0.000 0.862 0.203 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.009 0.000 0.000 0.263 0.001
0.032 0.001 0.473 0.044 0.063 0.108 0.052 0.061 0.045 0.105 0.016 0.054 0.020 0.449 0.027
Zone16 Zone17 Zone18 Zone19 Zone20 Zone21 Zone22 Zone23 Zone24 Zone25 Zone26 Zone27 Zone28 Zone29 Zone30
2.81 10.52 2.07 0.08 2.15 2.63 2.06 0.07 0.70 2.55 11.54 0.55 2.61 1.16 3.30
0.000 0.305 0.000 0.424 0.000 0.000 0.000 0.574 0.000 0.000 0.181 0.008 0.000 0.000 0.000
0.112 0.419 0.083 0.003 0.085 0.105 0.082 0.003 0.028 0.102 0.460 0.022 0.104 0.046 0.132
86
K.X. Li et al. / Transportation Research Part A 66 (2014) 75–87
Table A3-1 Vessel flag results. Variable
Coefficient
P-value
Marginal probability
Variable
Coefficient
P-value
Marginal probability
x40 x41 x42 x43 x44 x45 x46 x47 x48 x49 x50 x51 x52 x53 x54 x55 x56 x57 x58 x59 x60 x61 x62 x63 x64 x65 x66 x67 x68 x69 x70 x71 x72 x73 x74 x75 x76 x77 x78 x79
1.01 1.37 0.85 1.55 0.57 0.30 2.37 0.38 0.62 0.63 0.77 0.97 0.73 0.39 0.99 0.78 0.22 0.88 0.08 0.71 1.24 0.99 0.45 0.67 0.79 0.81 0.39 0.55 1.33 0.09 1.12 0.87 0.60 3.75 0.10 0.68 2.89 2.89 1.16 0.21
0.000 0.000 0.000 0.000 0.001 0.034 0.000 0.010 0.000 0.000 0.000 0.000 0.000 0.014 0.000 0.000 0.195 0.000 0.625 0.000 0.000 0.000 0.036 0.000 0.000 0.000 0.048 0.016 0.000 0.616 0.000 0.000 0.002 0.000 0.564 0.061 0.000 0.000 0.002 0.397
0.040 0.055 0.034 0.062 0.023 0.012 0.094 0.015 0.025 0.025 0.031 0.039 0.029 0.015 0.039 0.031 0.009 0.035 0.003 0.028 0.049 0.040 0.018 0.027 0.032 0.032 0.016 0.022 0.053 0.004 0.045 0.035 0.024 0.149 0.004 0.027 0.115 0.115 0.046 0.008
x80 x81 x82 x83 x84 x85 x86 x87 x88 x89 x90 x91 x92 x93 x94 x95 x96 x97 x98 x99 x100 x101 x102 x103 x104 x105 x106 x107 x108 x109 x110 x111 x112 x113 x114 x115 x116 x117 x118 x119
1.18 0.04 0.12 1.73 0.18 3.26 0.89 0.10 1.88 0.30 0.21 0.36 0.49 0.20 1.09 0.81 0.28 0.62 0.00 0.28 0.47 0.98 1.56 1.80 0.23 0.21 1.25 1.35 0.78 1.33 0.55 1.61 0.72 0.27 1.97 1.36 0.11 0.63 0.70 0.22
0.000 0.893 0.735 0.000 0.464 0.000 0.000 0.746 0.000 0.377 0.391 0.140 0.119 0.464 0.000 0.004 0.276 0.102 0.993 0.266 0.065 0.006 0.000 0.001 0.398 0.423 0.000 0.001 0.001 0.001 0.037 0.000 0.020 0.498 0.000 0.040 0.715 0.250 0.043 0.548
0.047 0.002 0.005 0.069 0.007 0.130 0.036 0.004 0.075 0.012 0.008 0.014 0.020 0.008 0.043 0.032 0.011 0.025 0.000 0.011 0.019 0.039 0.062 0.072 0.009 0.009 0.050 0.054 0.031 0.053 0.022 0.064 0.029 0.011 0.079 0.054 0.004 0.025 0.028 0.009
Table A3-2 Vessel flag results (Continue). Variable
Coefficient
P-value
Marginal probability
Variable
Coefficient
P-value
Marginal probability
x120 x121 x122 x123 x124 x125 x126 x127 x128 x129 x130 x131 x132 x133 x134 x135 x136
0.19 4.87 1.60 1.11 2.33 0.74 0.08 0.80 0.60 1.02 1.17 12.69 1.05 0.95 0.87 0.15 1.14
0.720 0.329 0.002 0.069 0.000 0.022 0.907 0.018 0.093 0.001 0.018 0.182 0.003 0.163 0.012 0.730 0.003
0.007 0.194 0.064 0.044 0.093 0.030 0.003 0.032 0.024 0.041 0.047 0.506 0.042 0.038 0.035 0.006 0.045
x137 x138 x139 x140 x141 x142 x143 x144 x145 x146 x147 x148 x149 x150 x151 x152 x153
1.18 0.26 0.64 0.07 0.09 1.73 1.11 0.12 0.95 5.93 0.19 0.43 2.03 1.54 1.33 2.40 0.89
0.008 0.668 0.145 0.908 0.887 0.000 0.000 0.804 0.010 0.315 0.591 0.258 0.000 0.012 0.165 0.000 0.018
0.047 0.010 0.026 0.003 0.003 0.069 0.044 0.005 0.038 0.236 0.008 0.017 0.081 0.061 0.053 0.096 0.035
K.X. Li et al. / Transportation Research Part A 66 (2014) 75–87
87
References Bergantina, A., Marlow, P., 1998. Factors influencing the choice of flag: empirical evidence. Marit. Policy Manag. 25 (2), 157–174. Cariou, P., Mejia Jr., M.Q., Wolff, F.C., 2008. On the effectiveness of port state control inspections. Transp. Res. E 44 (3), 491–503. Cashman, J.P., 1977. Analysis of world merchant ship losses 1967–1975, paper no. 2 in Safety At Sea, Proceedings, London, 23–27 May. Chang, Y.H., Yeh, C.H., 2004. A new airline safety index. Transp. Res. B 38 (4), 369–383. Clark, P.D., 1991. Vessel classification isn’t insurance policy, appeals court rules. Sea Law, 3, http://navlaw.com/articles/index.html. Clark, P.D., 2009. An American admiralty law viewpoint on the changing role of classification societies. Sea Law, 5, http://navlaw.com/articles/index.html.. Cramer, M., Franses, P.H., Slagter, E., 1999. Censored Regression Analysis in Large Samples with Many Zero Observations. Econometric Institute Research Report 9939/A. Faragher, W. E., Pizzo, J. T., Reinburg, L. R., Reiter, A. D., Graham, W. C., 1979. Deep water ports approach/exit hazard and risk assessment. Planning Research Corporation. McLean, VA. U.S. Coast Guard, Report No. CG-D-6-79. Forsyth, C.J., 1991. Factors affecting tanker safety. Marit. Policy Manag. 18 (3), 313–319. Gaarder, S., Rognstad, K., Olofsson, M., 1997. The impact of human element in maritime risk management. In: Guedes Soares, C. (Ed.), Advances in Safety and Reliability. Pergamon, Oxford, pp. 857–868. Grabowski, M., Ayyalasomayajula, P., Merrick, J., Mccafferty, D., 2007. Accident precursors and safety nets: leading indicators of tanker operations safety. Marit. Policy Manag. 34 (5), 405–425. Green, W.H., 2003. Econometric Analysis, 5th ed. Prentice Hall, Engle, Cliffs, NJ. Hockaday, S.L.M., Chatziioanou, A., 1986. An analytical method for aircraft collision risk estimation. Transp. Res. B 20 (5), 415–428. Jin, D., Thunberg, E., 2005. An analysis of fishing vessel accidents in fishing areas off the northeastern United States. Saf. Sci. 43 (8), 523–540. Jin, D., Kite-Powell, H.L., Thunberg, E., Solow, A.R., Talley, W.K., 2002. A model of fishing vessel accident probability. J. Saf. Res. 33 (4), 497–510. Knapp, S., Franses, P.H., 2007a. Econometric analysis on the effect of port state control inspections on the probability of casualty. Can targeting of substandard vessels for inspections be improved? Marine Policy 31 (4), 550–563. Knapp, S., Franses, P.H., 2007b. A global view on port state control – econometric analysis of the differences across port state control regimes. Marit. Policy Manag. 34 (5), 453–483. Kristiansen, S., 2005. Maritime Transportation: Safety Management and Risk Analysis. Elsevier, Oxford. Li, K.X., Cullinane, K., 2003. An economic approach to maritime risk management and safety regulation. Marit. Econ. Logist. 5 (3), 268–284. Li, K.X., Wonham, J., 1999. Who is safe and who is at risk: a study of 20-year-record on accident total loss in different flags. Marit. Policy Manag. 26 (2), 137– 144. Li, K.X., Zheng, H., 2008. Enforcement of law by the Port State Control (PSC). Marit. Policy Manag. 35 (1), 61–71. Li, K.X., Wang, Y., Min, J., 2009. Quantitative analysis of materiality in maritime insurance. Marit. Policy Manag. 36 (5), 437–455. Li, K.X., Yin, J., Bang, H.S., Yang, Z., Wang, J., 2014. Bayesian network with quantitative input for maritime risk analysis. Transportmetrica A 10 (2), 89–118. Lloyd’s Fairplay, 1993–2008. World casualty statistics. Annual Report. Lloyd’s Fairplay, 2008. World Shipping Encyclopedia. Version 2008. Mostafa, M.M., 2004. Forecasting the Suez Canal traffic: a neural network analysis. Marit. Policy Manag. 31 (2), 139–156. Pronce, P.V., 1990. An analysis of maritime total losses worldwide and for selected flags. Marit. Technol. 27 (2), 114–116. Roberts, S.E., Marlow, P.B., 2002. Casualties in dry bulk shipping (1963–1996). Marine Policy 26 (6), 437–450. Romer, H., Petersen, H.J.S., Haastrup, P., 1995. Maritime accident frequencies: review and recent empirical results. J. Navigation 48 (3), 410–424. Rose, N.L., 1990. Profitability and product quality: economic determinants of airline safety performance. J. Polit. Econ. 98 (5), 944–964. Soares, C.G., Teixeira, A.P., 2001. Risk assessment in maritime transportation. Reliab. Eng. Syst. Saf. 74 (3), 299–309. Talley, W.K., 2001. Determinants of the property damage costs of bulk barge accidents. Marit. Policy Manag. 28 (2), 175–186. Uncles, M.D., 1987. A beta-logistic model of mode choice: goodness of fit and intertemporal dependence. Transp. Res. B 21 (3), 195–205. Wang, J., Pillay, A., Kwon, Y.S., Wall, A.D., Loughran, C.G., 2005. An analysis of fishing vessel accidents. Accid. Anal. Prev. 37 (6), 1019–1024. Yadlin-Weintraub, M., 1991. Some procedures for the statistical and empirical analyses of probabilistic discrete decisions models. Transp. Res. B 25 (4), 215– 236. Yap, W.Y., Lam, J.S.L., 2013. 80 Million-TEU container port? Sustainability issues in port and coastal development. Ocean Coast. Manag. 71, 13–25. Yin, J., Fan, L., Yang, Z., Li, K.X., 2014. Slow steaming of liner trade: its economic and environmental impacts. Marit. Policy Manag. 41 (2), 149–158.