1. CONTEXT OF THE STUDY
The Paris Memorandum of Understanding (Paris MoU) was established in 1982 to improve safety at sea. It consists today of 27 participating maritime Administrations and covers the waters of the European coastal States and of the North Atlantic from North America to Europe. Its mission is to inspect foreign ships calling at Paris MoU ports, aiming at eliminating the operation of sub-standard vessels through a harmonized system of Port State Control (PSC). The Paris MoU has for many years defined criteria in order to inspect the most riskyFootnote 1 vessels in priority order. The so-called Target Factor is still in use within the Paris MoU as a tool for selecting ships eligible for an inspection. For many years as well, the Black-Grey-White (BGW) lists of flags are established each year in order to provide an independent categorization of flags on the basis of inspected and detained vessels registered under these flags.
In cooperation with the European Commission and to take into account what happened during 2006 at the European Community level on the proposed recast Directive on PSC, the Paris MoU will soon implement a New Inspection Regime (NIR). This will replace the 25% commitment for individual member States with an annual 100% inspection coverage for vessels operating in the Paris MoU region. But while low-risk ships will be rewarded with a 24 month interval, the High Risk Ships (HRS) in the NIR will be subject to a more rigorous inspection regime with an inspection every 6 months. Criteria aiming at characterizing HRS are described in the internal report (Paris MoU, 2006).
In this report, Paris MoU defines on one hand, the ship risk profile using a set of variables related to the ship – ship type, age, flag, company, Recognized Organization (RO), number of deficiencies and detentions within previous 36 months – and on the other hand characterizes HRS by criteria in relation with each of these variables considered one by one, with associated weighting points. Black flagged vessel is one of these criteria. More precisely, as shown in Table 1, criteria examples are: for ship type – chemical tankers, gas carriers, oil tankers, bulk carriers, passenger ships have a weighting point of 2; for ship age – ships >12 years old have a weighting point of 1; for ship flag – flags in the black list have a weighting point from 2 to 1 according to whether the flag is considered Very High Risk (VHR), High Risk (HR), Medium to High Risk (MtoHR) or Medium Risk (MR), etc. The total score is then added and the Paris MoU defines High Risk Ships as “ships which meet criteria to a total value of 5 or more weighting points” (Paris MoU, 2006).
Design of High Risk Ship in the NIR. (Source: (Paris MOU, 2006))
However, all of this additive process is based mainly on ship detentions and/or deficienciesFootnote 2 and does not embrace the concept of risk according to its technical definition (see Footnote footnote 1). In particular, the probability component of risk is missing or in other words the probability of a casualty occurringFootnote 3.
In this context, some members of the Sub-Committee on Flag State Implementation (FSI) of International Maritime Organization (IMO), in particular Turkey and New Zealand, had recommended that consideration be given to the integration of casualty and PSC data for the measurement of flag State performance, for the consideration of balanced criteria for the targeting of PSC inspections and more generally in order that a complete picture of a vessel's activities and history would be available to the maritime community (FSI 15, 2007). It is not the main intention of this paper to identify the pros and cons of the combination of casualty and PSC related data and the need for further study, if any. These are the first aspects of the terms of reference of the Correspondence Group which has been set up under the co-ordination of France by the Sub-Committee of FSI to clarify the situation (FSI 15, 2007).
The aim of this paper is to present a method similar to the one used by the Paris MoU to class flags in their BGW lists, but in which the use of casualties instead of detentions and the consideration of a multivariate approach instead of considering the flag only as the Paris MoU does, may enable the extension of the BGW detention-based lists of flags established by the Paris MoU to BGW lists of categories of vessels based on their observed casualties on a given period. It is thought that this method, if its application is proven to be feasible, would be a valuable contribution to the Correspondence Group mentioned above if this group concludes with the necessity of combining casualty data and PSC data, with a view to complement the current processing of flag State performance and the targeting criteria for ship inspections.
The remainder of this paper is organized as follows. In Section 2, we briefly recall the method used by the Paris MoU to class the flags in their BGW lists. Then we explain in Section 3 how to extend this method by a multivariate approach in order to establish BGW lists of categories of vessels as regards their observed casualties. A limited application of this approach is given in Section 4 to demonstrate the feasibility of its implementation and we discuss the limits. Finally, Section 5 presents some conclusions.
2. THE PARIS MOU BGW LISTS OF FLAGS
2.1. The calculation of flag performances
The text contained in this section is extracted from (Paris MoU annual report, 2006, pages 50–51). “The performance of each flag State is calculated using a standard formula for statistical calculations in which certain values have been fixed in accordance with agreed Paris MOU policy. Two limits are processed, based on binomial calculus: the ‘black to grey’ and the ‘grey to white’ limit, each with its own specific formula:
In the formula N is the number of inspections, p is the allowable detention limit (yardstick), set to 7% by the Paris MOU Port State Control Committee, and z is the significance requested (z=1·645 for a statistically acceptable certainty level of 95%).
Above this ‘black-to-grey’ limit means significantly worse than average, where a number of detentions below the ‘grey-to-white’ limit means significantly better than average. When the amount of detentions for a particular flag State is positioned between the two, the flag State will find itself on the grey list. The formula is applicable for sample sizes of 30 or more inspections over a 3-year period. To sort results on the black or white list, simply alter the target and repeat the calculation. Flags which are still significantly above this second target, are worse than the flags which are not. This process can be repeated, to create as many refinements as desired. (Of course the maximum detention rate remains 100%!) To make the flags' performance comparable, the excess factor (EF) is introduced. Each incremental or decremental step corresponds with one whole EF-point of difference. Thus the excess factor EF is an indication for the number of times the yardstick has to be altered and recalculated. Once the excess factor is determined for all flags, the flags can be ordered by EF. The excess factor can be found in the last column of the black, grey or white list. The target (yardstick) has been set on 7% and the size of the increment and decrement on 3%. The Black/Grey/White lists have been calculated in accordance with the above principles”.
2.2. The Paris MoU BGW lists 2006
To illustrate the application of the method used by the Paris MoU for the calculation of flag performances, Table 2 shows the Black-Grey lists 2006 established on 5 June 2007. The White list is established according to the same principles but is not shown in the Table.
The Paris MoU Black-Grey lists 2006 (5 June 2007). (Source: http://www.parismou.org/ Annual report, 2006.)
3. A MULTIVARIATE APPROACH FOR THE ESTABLISHMENT OF BLACK-GREY-WHITE LISTS OF CATEGORIES OF VESSELS WITH REGARDS TO THEIR OBSERVED CASUALTIES IN A GIVEN PERIOD
3.1. Ship variables, levels of variables and category of vessels: a multivariable approach
In the framework of this approach, each vessel will be characterized by a set of the following six variables:
• three physical variables: ship type, size and age,
• three variables related to the entities which, somehow or other, manage the ship and take account in particular of compliance to international regulation, maintenance of vessels and equipments, training and quality of personnel on board, etc: ship flag, company (or owner) and Recognized Organization (RO) (which is most often the classification society).
Other ship variables may be considered, taken away or added to the list given above without difficulty (e.g. addition of some variables related to the ship history such as change of flag, change of owner, change of RO, class withdrawn, etc.) the only constraint being to have enough ship numbers in each category of vessels.
The levels of each of these ship variables have then to be specified by considering their micro values and by grouping them adequately. This grouping process is made as a function of the values of the considered variable, having in mind to have sufficient ship numbers in each category of vessels. For example, it could be decided to define:
• for ship type: 5 ship type levels based on the IMO definitions of selected major ship types as follows: bulk carrier (including OBO); general cargo vessel and multipurpose (including RoRo, Reefer, Heavy Load, …); tanker (including oil, chemical, gas); container ship; passenger vessel (including ferries, RoRo passenger,..)
• for ship size: 9 ship size levels in gt: size 1: 300 ⩽<500; size 2: 500 ⩽<1000; …; size 9: ⩾60 000 gt.
• for ship age (which is a continuous variable): 6 ship age levels in years: age 1: 0 <⩽5; age 2: 5<⩽10; ….; age 6: >25.
• for ship flag: 4 levels by country groups (OECD, EU, Open Registries & Other countries) or by flag colours as allocated by Paris MoU as a function of the inspected and detained ships registered under this flag ( 6 levels from Black VHR to White)
• for ship company and ship RO: 4 levels by country groups (same as for flag) or by performances according to Paris MoU formula as a function of detentions/deficiencies related to the ships owned or classified under this company or this RO (4 levels from very low to high).
Defined within this approach is finally a category of vessels which is a set of vessels which meet a particular combination of levels of each of the sets of levels associated with the ship variables which have been chosen to characterize a vesselFootnote 4. For instance, on the basis of the six ship variables and levels defined above, a particular category of vessels (among a total of 25 920 categories) is the set of all container ships of size 4 (1500–5000 gt), age 5 (20–25 years), grey flagged, owned by a medium level company, classified by a high level RO. This notion of category of vessels is very important because it enables a multivariate analysis (of casualties, detentions, deficiencies, etc.) by considering the set of N variables that have been chosen to characterize a vessel as a whole, instead of considering these variables one by one as is done for instance in the Paris MoU targeting processFootnote 5. Some additional comments about this notion are given hereunder:
• Add1: more variables and levels per variable are considered, the higher the number of combination of levels (i.e. the number of vessel categories) and of course the less the number of vessels per category are. For example: If we consider the 3 ship variables type, size and age only with the above mentioned levels, there are 5×9×6=270 vessel categories. If we consider the flag as a 4th ship variable having 4 levels, there are 1080 categories of vessels; etc.
• Add2: for statistical reasons which will be explained in the next section, the number of vessels per category must be at least 30. This minimum number can be obtained most often by considering world merchant fleet descriptive statistics over a sufficient wide time period (e.g. 10 years assuming a minimum of 3 ships/category/year).
• Add3: if 30 vessels/category are not possible to get over the time period, a new level grouping has to be proposed by decreasing the number of variables and/or the number of levels of variables.
3.2. The calculation of the performances of categories of vessels
The same classical statistical method that is being used today by the Paris MoU to class flags in their BGW lists (see Section 2) is used here to establish the performance of each category of vessels as regards its observed number of casualties on a given time period (Degré, Reference Degré2007). The only differences with the Paris MoU are the following: casualties are considered instead of detentions; categories of vessels (as defined in 3·1) are considered instead of flags; the number n of vessels in a given category of vessels is considered instead of the number N of inspections a given flag is subject to.
To make clear the description of the method, it consists of calculating, for each category of vessels, a confidence interval around the allowable number of casualties of that category. This calculation is based on a fixed allowable probability of casualty (p), the number (n) of vessels of the considered category and the assumption that casualties occur independently, which is generally the case. In this situation, the number of casualties of each category of vessels follow a Binomial distribution B (n, p) which can be approximated by a Normal distribution N (np, np(1−p)) if n is greater than 30. It is then possible to build for each category of vessels a 95% confidence interval around the allowable number of casualties of that category (i.e. around the expected average number of casualties which value is n.p) as expressed by the relation (2) belowFootnote 6:
(where z=1·645 for a statistically acceptable certainty level of 95%).
The performance of each category of vessels may then be established by comparing the observed number of casualties of that category to the upper and lower limits of its own confidence interval (2) as follows:
• if the observed number of casualties of a category of vessels is over the upper limit of (2), this means that the observed number of casualties is statistically significantly higher than the expected average number of casualties of that category and this category of vessels belongs to the Black list (or, which is equivalent, is characterised as Black).
• If the observed number of casualties of a category of vessels is under the lower limit of (2), this means that the observed number of casualties is statistically significantly lower than the expected average number of casualties of that category and this category of vessels belongs to the White list (or, which is equivalent, is characterised as White).
• If the observed number of casualties is inside the confidence interval (2), this means that this observed number of casualties is around the expected average and this category of vessel belongs to the Grey list (or, which is equivalent, is characterised as Grey).
To determine the value of the allowable probability of casualty (p) of a certain type of casualty, one will estimate it by the total number of casualties of this type that occurred over a given time period (generally many years) out of the total world merchant fleet over this same time period.
4. APPLICATION OF THE METHOD RESTRICTED TO THREE SHIP VARIABLES
Before presenting the results obtained, a brief description of the data that have been used for this application is given hereunder.
4.1. Casualties descriptive statistics
IMO casualty records as reported to IMO by member StatesFootnote 7 have been used for this application and casualties recorded world-wide over the period [1998–2003] have been analysed. As shown in Table 3, over that 6-year period, data on 2343 very serious or serious casualties of vessels (commercial and non commercial) of all sizes was collectedFootnote 8 (390 per year on average). 299 vessels were removed from the data set, in order to analyse only those casualties involving commercial vessels of a size equal to or greater than 300 gt. The remaining 2044 serious or very serious casualties that occurred world-wide in the period [1998–2003] (i.e. an average of 340 casualties per year) were selected. A univariate and multivariate casualty analysis was then done as described in (Degré & Muller, Reference Degré and Muller2006).
Distribution of casualties according to severity evaluated in this application.
4.2. World merchant fleet descriptive statistics
World merchant fleet descriptive analysis has the aim in this application to derive the number n of vessels in each category (as defined in Section 3.1). In principle, it is necessary to consider the world merchant fleet (vessels of 300gt or over) over the same years as those considered for casualties' analysis, i.e. over each year of the period [1998–2003]. Due to a lack of data, it has been decided to take the world merchant fleet of year 2000 as a reference, (which approximately, lies in the middle of the considered period) and to apply the 2000 fleet data over years 1998, 1999, 2001, 2002 and 2003 (or, in other words, to multiply by 6 the 2000 fleet data to cover the period [1998, 2003]). All the data related to the 2000 world merchant fleet were issued from (ISL, 2001) and from some specific requests to Institute of Shipping Economics and Logistics. A univariate and multivariate world merchant fleet analysis was then done as described in (Degré & Muller, Reference Degré and Muller2006).
4.3. Categories of vessels considered in this application and estimated values of the allowable probability of casualty
In this application, the calculations have been made in priority considering only the three physical ship variables type, size and age for which complete statistics on a world-wide basis were available. This was not the case for the other crucial ship variables flag, company and R.O. This is the reason why the only categories of vessels which are considered in this application are any combination of ship type, size and age levels, as defined in Section 3.1 (in total 270 categories). In this application, a fourth variable has also been considered: the type of casualty.
The estimated values of the allowable probability of a casualty (p) needed for the processing of the performance of each category of vessels with regards to its observed number of casualties (the factor p appears in the 95% confidence interval expressed by the relation (2) in Section 3.2) are contained in Table 4 as a function of casualty type. They are estimated by the total number of casualties of a given type that occurred world-wide over the considered six year period out of the total world merchant fleet over this same time period.
Values of probability of casualty as a function of casualty types.
CN=collision; FD=foundering; FX=fire/explosion; HM=Hull/machinery failure: WS=stranding.
4.4. Results on Black-Grey-White categories of vessels restricted to three ship variables
Let us recall that here only 270 categories of vessels are considered by combination of the levels of three physical ship variables i.e. type (5 levels), size (9 levels) and age (6 levels). The results when all casualty types are gathered, are shown in Table 5 but other results as a function of casualty types were analysed in (Degré & Muller, Reference Degré and Muller2006). In this matrix, each cell is a particular category of vessel. The meanings of the allocated colours Black-Grey-White to the categories of vessels (i.e. their performances) were given in detail in Section 3.2. Let us add only that two levels of the grey colour (light grey and dark grey) – corresponding both to the position of the observed number of casualties of the considered category inside the confidence interval (2) but respectively before the expected average number of casualties of that category and after the expected average number of casualties of that category – have been introduced in Table 5. But this distinction is not very important.
BGW categories of vessels with regards to their observed casualties over [1998–2003] when all casualty types are gathered.
As shown in Table 5, passenger vessels of large size (greater than size 7), old tankers (older than age 5), cargo vessels (most often of advanced age) of medium size (size 3 to 6)) and old bulk carriers (older than age 3) of medium to large size (size 5 to 9) are Black categories of vessels (or which is equivalent, belong to the Black casualty-based list).
If we compare these results to the criteria set up by the Paris MoU in the NIR for defining High Risk Ships (Section 1) on the basis of those related to type and age only, consistency is found mainly for bulk carriers (bulk carriers older than 12 years are targeted vessels by Paris MoU and the categories of bulk carriers older than 10 years of medium to large size are in the Black casualty-based list). Tankers older than 12 years are targeted vessels by Paris MoU, but only some categories of tankers older than 20 years are in the Black list as regards their observed casualties. Only the size among the three physical ship variables which were considered seems to play a role for passenger vessels in their membership of the Black casualty based list. Finally, cargo vessels are not targeted vessels by Paris MoU, but as shown in Table 5, many categories of cargo vessels older than 10 years belong to the Black casualty-based list. When considering the type of casualty, it can be shown (Degré & Muller, Reference Degré and Muller2006) that these categories of vessels are especially subject to foundering and this fact is confirmed by the (DNV, 2003) study.
4.5. Discussion of the results and limits of the application of the method
This application has been described with the aim to demonstrate that the implementation of the method is feasible. This seems promising if we refer to the results obtained by considering the three physical ship variables type, size, age with a casualty type as a fourth variable. For all these variables, statistics were available on a world-wide basis and although some simplifications were made and a rather short period of analysis (six years) used, the results obtained seem to be in line with logic and intuition. Of course these results have to be checked and amplified due to a rather short period of analysis and simplifications made. Moreover, a completion of this work, by adding to the three physical ship variables considered in this study the three other crucial ship variables that are flag, company, Recognized Organisation – i.e. all entities which, somehow or other, manage the ship and take account in particular of compliance with international regulation, maintenance of vessels and equipment, training and quality of personnel on board, etc. – has to be done, the only constraint being to have enough vessels in each vessel category over the period of analysis. For this reason, ten years of collection of data on world-wide casualties and on world merchant fleet seems to be a minimum. However, one of the major issues regarding the collection of data is the lack of completeness of the records of marine casualties: the main reason being that casualty reports are usually made by flag States but on a voluntary basis. Of course very serious and serious casualties are reported more systematically than less serious casualties but the non-exhaustiveness of the casualty data base remains today the main limit of the application of the method which has been described in this paper.
5. CONCLUSIONS AND PERSPECTIVES
A method, similar to the one used by the Paris MoU to class flags in their BGW lists, but in which the use of casualties instead of detentions and the consideration of categories of vessels (multivariate approach) instead of considering the flag only, may enable the extension of the BGW detention-based lists of flags established by the Paris MoU to BGW lists of categories of vessels based on their observed casualties on a given period.
Due to a lack of statistical data, the method has been applied considering three ship variables only but it has been proven feasible, knowing that the main limit of the application of this method is today the non-exhaustiveness of the casualty data baseFootnote 9. However, it remains that:
• The notion of vessel category is interesting because it enables a multivariate analysis, considering the set of variables characterizing the vessels as a whole, instead of considering these variables one by one, as is done for example in the Paris MoU targeting process
• The method to process the performances of each category of vessels (leading to the Black-Grey-White casualty-based lists of categories of vessels) is as simple as the one used by Paris MoU to class flags, but here casualties are considered, instead of detentions.
• The performances allocated by this method to categories of vessels (Black, Grey, White) are statistically significant: categories of vessels that are Black (respectively White) are observed to have a statistically higher (respectively lower) number of casualties than the expected average number of casualties of that category.
• The Paris MoU main objective is to eliminate sub-standard ships, but ultimately, its objective is to improve safety at sea and consequently to decrease risk of casualties. For this reason, the approach described in this paper, which is risk-oriented, is not in opposition to the one chosen by the Paris MoU but must be considered as complementary.
Indeed, as already noted in the introduction of this paper, most of the criteria set up in the NIR by the Paris MoU to target ships for their inspections are functions of ship detentions and/or deficiencies.
S. Knapp in her thesis (Knapp, Reference Knapp2007) tried in particular to answer the following question: are detentions/deficiencies estimators of a high probability of accident proneness (before the rectification of the deficiencies)?Footnote 10 But she did not show the evidence, on the basis of the data she analysed, of such correlations (most of the factors related to these explicative variables on the probability of a casualty were not significant in the various logistic regression models she implemented in her work). One of the main reasons for this absence of correlation is certainly due, in the opinion of the author of this present paper, to the fact that the origin of most casualties is due to human factors and that it is very difficult, during an inspection onboard a vessel anchored in a harbour, to check and detect all the human aspects which may lead to a casualty.
Soon, the Maritime Labour Convention (MLC, 2006), when ratified and implemented, will probably improve the task of the PSC inspectors to enforce these controls. But meanwhile, and as long as the above mentioned correlations would not have been proven, it seems necessary, in the opinion of the author, to combine casualty and PSC data for the measurement of flag State performance and for defining balanced criteria for the targeting of PSC inspections. For example, Black categories of vessels may be added to the Paris MoU criteria established in the NIR.
It is the hope of the author that the methodology which has been described in this paper, whose implementation requires to be expanded in the futureFootnote 11, may contribute positively to the work of the Correspondence Group which has been set up by the Sub-Committee on Flag State Implementation of IMO for studying these questions.
ACKNOWLEDGEMENTS
This research has been funded by the European Commission (DGTREN) and conducted under the MarNIS research project.
