Profiling fishing fleets requires input from several types of specialists: samplers, data-processing specialists, statistical analysts, and experts in various fields (sociologists, economists, biologists as well as fishing industry and fisheries staff). For this reason, it is very important that preparations are exhaustive and cover all of the following steps:
- Clarification of the nature of the problem and the questions posed: this step must include consultation with the specialists responsible for the profile as well as those who commission the study; this dictates the choice of methods and defines the context of the profile (the fishery concerned and information needs).
- Answering the question: "Which data is needed to provide which information?" This requires the specification of information that is necessary to answer the questions posed, from which the nature and the number of the variables to be collected can be ascertained. Each of the issues that are considered in establishing the fishing fleet's profile (such as vessel characteristics, social and economic issues, fishing activity) includes a considerable number of variables. BEWARE: "A surfeit of information is detrimental to the analysis!" Wanting to collect too much data is a common mistake. It reduces the effectiveness of the investigation and its analysis in terms of output, duration, cost, or the effectiveness of reporting. Generally, the choice of questions will be guided by the relevance of the variables to the analysis, in respect of the objective of the profile and of the costs resulting from the planned duration of the investigation. For this reason it is necessary to coordinate the planning of data-gathering along with data-processing.
- Planning the survey: the choice of sampling personnel, the development of the questionnaire, selection of sampling units (definition of the target population, sampling strategy), drafting guidelines for the investigators, validation of the questionnaire (this step often permits the revision of certain question and helps determine the time required to carry out the study, as well as contributing to the training of the sampling personnel).
- Planning the processing of the data: taking into account aspects of data-processing for the data entry, encoding, validation, and processing of data - the choice of equipment (hardware, software) and of the personnel for the task; the statistical analysis of the data (which methods for which questions and which data?) The process of planning the analytical methodology often makes it possible to review the relevance of the information and to consider the nature and attributes of the measured variables which will direct the statistical methodological choices to be made.
- Training needs: identifying the procedures to be carried out, and their timing, makes it possible to specify the expertise necessary to carry out the work and to identify the requirements for training and for external inputs. The first specifically concerns the training of the investigators and sampling personnel, and this must never be neglected as the quality of the data will depend on the quality of their work.
- Communication/information needs: fishing fleet profiling requires the acquisition of data to be precisely aligned with the component units of the fishery. The investigation of the fishers and/or the owners of the fishery units must be clearly announced in advance, specifying the objectives, the need for professional participation and the likely impact of the study on the people concerned. It is also appropriate to anticipate disseminating the results of the study to these same people, who in general have a direct interest in the questions which have warranted the need for a profile of the fleet.
- Cost-estimation: planning all the above steps makes it possible to estimate the costs associated with the process of profiling a fishing fleet, from design to utilisation of the results obtained.
The information necessary to produce a fishing fleet profile can thus be of various types: fishing effort, catch, vessel characteristics, fishing behaviour (tactics, strategies), etc. These data are generally obtained by various survey techniques; they require the effective participation of the subjects of the study, and on the application of a questionnaire by technical staff of the Fisheries Department or on a request to an agency specialized in carrying out such investigations. Irrespective of the technique used, several principles should be taken into account:
- rigorous selection of appropriate people to carry out the acquisition of the data in the field;
- preliminary training of the investigators: including sessions to explain the content of the questions to the investigators before any acquisition of data;
- holding field interviews. Regardless of the nature of information gathered, administering the questionnaire generally requires meeting with professionals (fishing, captains, ship-owners) whilst they are working. Such investigations can cover sensitive issues in the fishery and the field work often requires preliminary discussion to explain the context of the study and its objective. Biases found in the results of this type of investigation are often due to erroneous answers given in response to questions, whether it is due to unwillingness, or to lack of understanding of the questions posed. This type of field work requires some tact and perseverance by data recorders in order to provide the best psychological environment to obtain the necessary information.
- identification of local counterparts and someone to act as focal point for information.
The different aspects of fishery catch and effort data-collection will not elaborated upon in this document since this is a traditional part of the normal duties of those in charge of monitoring fishing, and fisheries management.
For industrial fisheries, this information is generally acquired by means of catch-effort return forms, or log-books, given to fishers who are obliged to complete them as a licensing requirement. Catch and effort data from these fisheries makes it usually possible to carry out an exhaustive analysis of the fishery. The compilation of the volume unloaded, by species or commercial category, and of duration of fishing for the various vessels, provides the total catch and effort of fisheries. These are the data used to establish the fisheries models used in the working groups that make recommendations for the management of industrial fisheries under the control of regional fisheries management organizations such as ICCAT. In addition, these data can provide the means for detailed analyses of fishing strategies since it is often possible to reconstitute the effort and catch by fishing trip, and the calendar for each vessel (see Section 2.2.3).
For small-scale inshore fisheries or artisanal fisheries, information is collected from surveying fish landing points, which makes it possible to obtain catch and effort estimates from complete fishing trips. Data are collected for a sample of the fishery, which is often difficult to study because of the number of fishing units and the spatial and temporal dispersal of landing points. These routine investigations are generally carried out by technical staff of government fisheries services; they must be rigorously sustained, and the work of the investigators must be encouraged and remunerated accordingly in order to ensure the long-term stability of the system.
Fishing fleets are generally the subject of a regular census, which provides an exhaustive inventory of fishing units. The issue of fishing licences provides the opportunity for an annual inventory of active units, to obtain information on the general characteristics of the vessels and to catalogue them according to activities which are subject to regulation. This census makes it possible to define a frame survey, which will be used later to provide a context for sampling the fishery. A "frame survey" identifies the whole range of accessible and countable elements, from which it is possible to take a fragment (sample) to extrapolate the state of the whole (population), for example: a list of postal addresses for the demographic census; a list of telephone numbers for a census of the population accessible by telephone; a list of vessel serial numbers for a licenced vessels in a fleet.
A sample from the total population is usually necessary in a survey to establish a profile of fishing fleets. This type of profiling is carried out occasionally, not routinely, and requires detailed information on the fishing units, something which is difficult to obtain for the entire fleet. An investigation into the characteristics of the fishing units can consist of many questions asked on a variety of topics: design features of the boat, fishing gear, fishing operations, operating accounts and crew. The frame survey then makes it possible to randomly choose samples from the whole population, which assures the representativity of the fleet sample.
|
In 1996, the Direction de la Marine Marchande of Morocco estimated the number of registered boats (units) in the inshore fishery to be 2 169. Each unit was described by 19 variables describing the general characteristics of the boats. Using information from the Direction des Pêches Maritimes et de l'Aquaculture on fishing licences in 1996, the active fishery in 1996 was estimated to comprise 1 777 vessels, not including artisanal fishing boats. Within the framework of the Inshore Fishery Modernisation Programme, an investigation was carried out to profile the fleet. A representative sample of 497 fishing units was selected from the list and from the descriptions of the 1 777 boats which constituted the frame. |
The databases that result from the obligatory completion of logbooks in certain fisheries make it possible to reconstruct "fishing calendars" for the various units. It is then possible to profile the fishing fleet according to fishing activities over a period of time, and to answer questions about the dynamics of exploitation. In the case of fisheries where port-sampling is carried out, these calendars can only be established by systematically pursuing supplementary data for a sample of fishing units. This type of investigation is relatively difficult to implement for it requires assiduous fieldwork to regularly make contact with units during the course of their fishing operations. The information necessary to classify fishing behaviour generally requires direct discussion with the skipper of the fishing boat. They would have to be interviewed on return to port and the frequency of contact will thus depend on the time at sea, each interview relating to one or more previous trips (a maximum of two days proceeding the day of the interview, since the quality of the data depends on the memory of the person questioned).
|
A follow-up survey of fishing units was carried out by the Centre de recherche océanographique de Dakar-Thiaroye (Senegal) in 1992. It surveyed a hundred fishing units to describing the tactics and strategies of artisanal fishing. This supplementary follow-up survey was carried out in order to help develop a model to simulate the dynamics of the fishery, which required a better understanding of the comportment of fishing units in the short- to medium-term. The investigation consisted of three parts: 1 - vessel specifications, 2 - the description of trips by regularly sampling on return from fishing, and interview with fishers, and 3 - the frequency and the nature of fishing activity during the course of the year (reason for return, fishing area, or fishing gear). |
The development of the questionnaire must be based on consensus between the various beneficiaries of the investigation. Several meetings are generally necessary:
1. to identify the topic and the nature of the questions;
2. to obtain agreement on the wording of the questions;
3. to test and validate the prototype questionnaire.
The questionnaire usually consists of several parts classified by topic (e.g.: technical characteristics, fishing activities, catch, social and economic information). Annex I provides an example of a questionnaire, implemented in Morocco for the profile of its coastal fleets, which consisted of several parts; 12 pages in total. Although rather cumbersome, this questionnaire had to be comprehensive enough to provide all of the information necessary to design a programme of modernization for the fleet and for the conversion of fishing vessels. In fact, all the information required by the questionnaire in Annex I would not have been necessary to fulfill the requirements of a fleet profile. The entire data-set resulting from the questionnaire used the Moroccan survey could be used by many researchers working on fishery development or management problems, but it should be noted that such a questionnaire is perhaps a little too lengthy, making it difficult to plan data-processing and analysis.
|
List of topics used in the typology of fleets for the Moroccan inshore fishery study (See Annex I): Vessel Technical specification (questionnaire submitted to all the sampled vessels) - Characteristics of the vessel Crew questionnaire (questionnaire submitted to a sub sample of the sampled vessels) - Fishing master |
A questionnaire generally consists of several types of question:
- numerical: quantitative (e.g. vessel size);
- nominal: qualitative value (e.g. target species name);
- scale: response on a scale of satisfaction or agreement (e.g. frequency of fishing trips: 1 = less than average, 2 = average, 3 = more than average);
- simple: only one possible answer (e.g. primary target species);
- multiple: several possible answers (e.g. the top three target species);
- simple text: answer comprising a word or a code (e.g. name of the port at which operations are based).
It is possible to distinguish between "closed" questions, which require a response to a given series of choices (for example: "type of vessel = trawler, sardine boat, longliner, mixed gear, or other (specify)". The "other" heading makes it possible in hindsight to create a new category if it proves to be relevant and numerous), and "open" questions, to be answered in an open-ended textual form (e.g. a detailed description of the holds). The answers to this last type of question can be classified and coded at a later stage in the treatment or can be left as comments available on the questionnaire forms.
The answers to the questions are either quantitative (e.g.: "horsepower of engine?") or qualitative (e.g.: "manufacturer of engine?"). In the first case it is important to explain the units of measurement used in the questionnaire, to avoid errors resulting from mistakes of scale. Generally, the questionnaire should include all of the information necessary to guide the investigator during course of the study (e.g.: "engine power: to be indicated in horsepower"; "operational capacity: to be indicated in rpm"; "position of bridge equipment: to be marked on the diagram provided"; "number of fishing operations per fishing trip; to be indicated, for a trawler as the number of deployments of the trawl, for a sardine boat as the number of deployments of the net, for longliner, in a number of sets"). There should be a brief explanatory guide to each questionnaire to remind the investigators of the guidelines for properly implementing the survey. In the case of a qualitative answer (to a question of the nominal type) it is desirable to indicate on the questionnaire the list of possible answers: (e.g. "Type of vessel: trawler, longliner, liner, seiner or other") in order to avoid errors of understanding in the question or the recording of information.
In parallel with designing and prototyping the questionnaire, it is necessary to anticipate computer input and data processing requirements, since these may have an influence on the coding of the information. Several computer applications (e.g. Sphynx, Question) exist for the purpose of designing and editing a questionnaire, while at the same time providing support for data input, and offering statistical processing functions to assist in the presentation of results. Such software makes it possible to anticipate the required analytical steps at the time the questions are formulated (chronological order of the questions, placing the questions in comparison with others, grouping the questions logically by type, maximum number of modalities for a nominal question), and to minimize any problems of execution or management of the investigation down the line, in particular during the recording, the validation and coding, and the compilation of data.
The questionnaire can be used both for implementation in the field and for computer input of the data. For this it must be designed in such a way that responses in the field can be made in the form required by the database software. It then consists of a section for writing the answers at the time of the interview, and a section for coding the information, putting it into a form suitable for computer input. This type of form has the advantage of minimizing transcription errors during coding, which can be carried out by the investigator between two surveys (and not by a third person), and of economizing on forms.
|
Annex II provides an example of the form used for the follow-up survey of fishing units in the Senegal artisanal fishery. This example illustrates the design of a questionnaire for the simultaneous acquisition of information in the field and its coding for the data-processing. It also demonstrates the type of questions to ask in order to profile a fishing fleet according to fishing behaviour. |
A survey which contains a large number of questions is seldom possible to implement across the whole population. In obtaining a manageable sample it is necessary to define a subset of units which provide the best possible representation of the total population (in this case, the fishing fleet). According to statistical theory, a sample is most likely to be representative if it is selected randomly, without bias in the choice of sample units by the investigator. This rigorous technique requires access to a frame survey in order to define a random sample, based on a known, non-zero, probability of each unit taking part in the investigation. This minimizes bias arising from a lack of correspondence between the investigated sample and the population onto which one seeks to extrapolate the results of the investigation. However, random sampling sometimes poses problems from the point of view of logistics and cost. It is necessary to make contact with the randomly identified fishing vessels in the sample, irrespective of their locality or their availability. A sample selected on a non-random basis, where the units of the sample are chosen according to pre-established criteria, is one of the strategies often practiced in order to cope with these constraints. In this case, one seeks to obtain a sample as representative as possible of the full range of heterogeneity observed within the fishery. Stratified sampling makes it possible moreover to maximize precision whilst minimizing effort in the acquisition of information. The stratification allows existing knowledge about the heterogeneity of the fishery to be taken into account, in particular the spatial dimensions (geographical distribution) and fishing techniques.
|
A sample of 497 fishing vessels in the Moroccan coastal fleet was selected, on a logical basis according to the availability of fishing masters, in order to cover 25-30 % of the vessels of each port and type of boat (trawler, sardine boat and longliner). The survey was guided by information on fishing licences provided by the Direction de la Marine Marchande and the Direction des Pêches Maritimes et de l'Aquaculture, which gave the number of units listed for each of the ports and types of vessels. The choice of units in the field was made with the aim of ensuring the most representative possible coverage of the diversity of the fleet, in particular from the point of view of horsepower, age, and length of boat. This technique of sampling is similar to the "quota method". The statistical population - the vessels of the coastal fleet - is stratified according to two criteria: geographical (port) and type of fishing licence (trawler, sardine boat or longliner); the total of the boats by stratum (a combination of port * type of licence) represents the whole of the active coastal fleet at a given point in time, according to the data provided by the administrative services in 1996. The completely random selection of 25-30 % of the units estimated in each stratum provides a representative sampling of the spatial heterogeneity and of all the licence types in the coastal fishery. |
Even if the sample does not allow the valid extrapolation of results to the entire statistical population (that is to say, all of the elements - here all of the vessels of the fishing fleet - from which the sample was selected according to the sampling criteria), classification is nevertheless of interest from a descriptive and qualitative point of view. Indeed, apart from the classification of fishing vessels as such, one of the major objectives of a fleet profile is to try to distinguish different classes within a heterogeneous assemblage, by highlighting variables that differentiate between the various classes. Even if the relative proportions within the total population are not respected, the process of identifying types within the sample still makes it possible to establish their characteristics, their specificities and their differences. A second, less detailed, survey can then be carried out on the whole population in order to evaluate the importance of the types highlighted by the first step. However, it is preferable to start with an adequate sample in order to target the two objectives directly, that is to say the quantification as well as the identification of the various types of units in the fishing fleet.
Once the fieldwork has been completed, the information that has been gathered must be centralized and input to the computer. Data entry is tiresome work, but it is necessary to accord it some attention because of inevitable errors in reading and entering the information, especially if it is not carried out by people accustomed to using a keyboard for data-entry. In order to minimize these errors, it is often advisable to carry out:
1. duplicate data-entry (investment in this is profitable compared to the costs resulting from errors detected later in the database) and;
2. data input masking: this makes it possible, at the point of entry on the keyboard, to limit input to the type of answers that are acceptable as responses to that question (e.g.: as a result of a mistake in reading the forms, the entry of a quantitative datum - for example "12" in a column corresponding to a qualitative question - for example "fishing licence?"- can be made impossible). Input masking can also allow the entry of quantitative values which correspond only to the range of possible answers to the question. The program can be set to prevent, for example, the entry of the value "2001" in response to the question "age of the boat"? The data entry mask is generally defined at the time that the questionnaire is designed or the data-entry module of the database is programmed.
The computerized data can be stored in tabular form in a spreadsheet (e.g. Microsoft Excel, Lotus 123) or more compactly and efficiently in a relational database management system (e.g. Access, Oracle, dBase). The design of a database requires some technical skill, but the exercise can be profitable, particularly where there are large quantities of data which need to be regularly updated and/or transferred to other information processing systems.
|
Data on the 497 fishing units was produced by a survey, carried out by the Institut National de Recherche Halieutique, (INRH) in Morocco, of the inshore fishery, and entered and stored using dBase. Eight different files resulted: 1- General characteristics + bridge equipment; Each file comprised 497 records and a number of fields corresponding to the number of variables relevant to each topic. Several key fields are common to different files, such as the port and the name of the boat. This establishes a relationship allowing later concatenation (joining) of the files. The assemblage of 8 files on the Moroccan fishing units includes a total of 601 variables, of which 550 are unique. |
The data entered into a data processing system can be subject to secondary treatment by other users. There are two types of data-processing compatibility problems: a difference in the type of computer or operating system (commonly: PC - the so-called "wintel" machines, Apple Macintosh or Sun/Unix machines) - and a difference in format of files produced by the data processing software (files produced by the Excel spreadsheet program possess by default, for example, the filename extension ".XLS" which makes it possible to identify the software which created the file. There now exist means for importing and exporting files between several different types of computer and data-processing programs, which overcomes most problems of compatibility.
|
The data in the 8 files of the INRH survey were entered using a dBase program, resulting in files with a ".DBF" extension. In order to be accessible to certain other programs, these original files were then exported to TEXT format (file extension ".TXT"). Using the SAS statistical software package, the 8 files were merged into one data file consisting of 497 different units and 550 variables (SAS data files with an ".SSD" file extension). Another conversion to TEXT format of this unified file then made it possible to import the data into it the SPAD data-analysis program to carry out the typological analysis that resulted in the profile. The profile of the Moroccan fleets thus required the use of three data-processing software packages: dBase for the basic data-entry and compilation, SAS for the management of the data and certain statistical treatments, and SPAD for the typological analysis itself. |
The data processing sequence can thus comprise a series of manipulations involving various computers and data-processing programs. The processing of data often requires the use of several data-processing tools, according to the functionality of each: management of the data (input, verification, and compilation), graphical analyses, elementary statistical analyses, multivariate data analysis. However, one person must be in charge of the original database, in order to avoid problems of revision or duplication of the data. On the other hand, since several people are often involved in the analytical steps, appropriate documentation describing the format of the data files must be available. This documentation should normally provide the following information:
- the name of the file and a description of the contents of the file;
- the name of the variable corresponding to each field (or column) in the file;
- the specific and complete meaning of each field/variable
This list includes as many lines as there are fields (variables) in the file. It is also advisable to indicate the size and the nature (quantitative, qualitative, and textual) of the variable described by each field, in order to facilitate reading the data during secondary treatment.
|
Annex III provides a description of the 8 files produced during the entry of data resulting from the survey questionnaires used in profiling Moroccan coastal fleets. This example demonstrates how each question was titled and expressed in the form of one or several fields in the computer file. |
It is also common to use existing data files arising from other sources and providing additional information about the subject of the study. The problems encountered when interfacing the data from various sources often result from the data coding system. To extract data from file "B" in order to add them to the data of file "A" requires the presence of a common field (called a "key" field), i.e. a common reference that is identical between the two files, such as a vessel name. This process requires, on the one hand, identification of the two data sources and the availability of expertise to solve the problems encountered and, on the other hand, a preliminary analysis of the two files A and B to check the adequacy of the key reference used in merging the two files.
|
For the Moroccan coastal fleet, the Direction de la Marine Marchande held a data-file of boats registered in 1996, which included entries on 2,169 boats with 19 variables describing each. In parallel, the Direction des Pêches Maritimes provided a data-file of boats operational in the coastal fishery, consisting of the holders of fishing licences in 1996. This file included 1,777 units and 14 variables. These data were used to assess how representative the "profile study" sample of 497 boats was in comparison to the registered fleet as a whole, as well as the part of the fleet licenced for fishing in 1996. It was necessary, as a starting point, to "clean" the data-files by checking the appearance of the boats in the sample with the two master files, on the basis of registration number and the name of the boat (it was necessary, for example, to add to the file the information on several boats which had been given licences after 1996, and which were therefore not listed in the master file of 1996) as well as the integrity of the lists of boats in the different files (e.g.: certain vessel names were duplicated and had to be removed from the master files). |
Whatever the nature of the data - catch and effort, fishing vessel characteristics, or fishing calendars - the data processing for a fleet profile requires the following steps:
- Validating the data: at this stage, if the data were never subject to preliminary validation, it may be possible to check the acquisition process and to correct any errors. It is possible to evaluate the rate of response to the questions, and any missing data;
- The definition of data tables: by choosing the individuals and the variables which will be subject to statistical analysis. Statistical software generally uses a data file which is a table with a number of lines or rows equal to the number of individuals (the n elements of the sample, which are the n fishing units which are the subject of the profile) and a number of columns or fields equal to the number of variables on which the statistical analysis will be carried out. At this stage, it is useful to distinguish two types of variables: principal variables and supplementary variables: The former are taken into account during the typological construction of the profile (active variables), whereas the latter take effect at a secondary level to help explain the profile that results. For example, it is not appropriate to use the variable "port" in the development of the profile (otherwise the boats will be classified according to geographical criteria, amongst other things) but to look a posteriori to see if a profile established on the basis of other variables (vessel characteristics, fishing activity schedules) has any relationship to the variable "port": the latter is then called a "supplementary variable", and is used to explain the results obtained from the "active variables".
- The description of each variable of interest by univariate analysis. The initial step in any statistical analysis is the systematic analysis of the variables, starting with their elementary statistics (mean, standard deviation, minimum, maximum, and mode, median). In addition to an understanding of the data and their variability, this stage often contributes to detection of errors within the table. In fact, analysis often provides the best validation of the data! Systems of double-entry and data input-masking (see 2.5) do not make it possible to detect all entry errors and it is common to find aberrations in the data files when looking at the results of analysis.
- The study of the relationships between the variables of interest through bivariate analysis. A bivariate analysis is carried out through statistical methods which make it possible to study the relationship between variables taken pairwise (two by two). These methods include graphical techniques and quantitative techniques which offer the possibility of testing the strength of the relationship between the variables.
- The study of the similarities between the individuals and between the variables of the table by multivariate analysis. These statistical methods make it possible to visualize and to quantify, on the one hand, the relations between all the variables retained following the bivariate analysis and, on the other hand, the resemblances between the individuals of the table described by the "multivariable" of the table. These methods include graphical techniques to visualize the relationships between the individuals and between the variables, and quantitative techniques which provide indices to interpret the results and, if required, to test the validity of the statistical model.
- The synopsis and restitution of the results: one of the great difficulties of statistical analyses involving a large number of variables, including the typological analyses necessary for a fishery profile, is to provide a sufficiently clear, overall summary synthesis of the successive analyses carried out on all of the variables that describe the individuals in the data tables.
Statistics offers a range of methods, the choice of which will depend on four factors:
1. the type of variables: qualitative or quantitative;
2. the status of the variables: explanatory or dependent;
3. the number of variables: one, two or multiple;
4. the type of analysis: exploratory (descriptive) or confirmatory (test).
Producing a fishing fleet profile consists of exploring the structure of the data by analyses, on the one hand, of the relationships between the variables and, on the other hand, of the similarities between the individuals described by these variables. The first type of analysis allows the selection of the most relevant variables for the profile, and brings out the combinations of values of these variables which will best characterize the different classes of fishing unit. The second type of analysis makes these classes clear by grouping the individuals that resemble each other, on the basis of the description of the variables in the data tables. Two stages are recognized:
setting up the data structure; and
interpreting the data structure.
The first stage utilizes exploratory or descriptive methods to summaries the data set in the form of statistical tables or graphs (e.g.: classes of individuals described by their means). The second stage requires the use of statistical tests to validate the relevance of the classes by highlighting the significant variables of these classes.
Variables can be summarized by several statistical indices. For quantitative variables, the average or the median is used to describe the location of n individuals within the range of the variable and its standard deviation, whilst the minimum and the maximum are used to describe their dispersion (variability). It is also interesting to make use of quantiles which correspond to the values of the variable which separate the n individuals by a given percentage. For example, the quartiles which separate the distribution into 4 equal parts: Q1 for the first 25% [Min-Q1], the median which separates the distribution in two equal parts [Q1-Med] = [Med-Q3] = 25% and Q3 for the last 25% [Q3-max]; or the (per) centiles which separate the distribution in 100 equal parts. It is particularly useful to analyse C1 and C99, since these correspond respectively to the value of the variable which defines 1% of the extreme individuals of the distribution [min-C1] = [C99-Max] = 1%.
Qualitative variables are described by the frequency, in absolute value (number) and relative value (percentage), of the individuals in the different values (modal classes) of the variable.
|
The 497 boats in the sample of the Moroccan coastal fleet were characterized by elementary statistical analysis of the quantitative variables available in the 2 master files ("Registration" and " licence") in order to assess the representativity of the sample compared to the whole fleet. The number in the sample, N indicates the number of values found in the file; it does not necessarily correspond to the number of boats of the sample, 497: indeed, there is no information on several of these boats in the files provided by the administration (there is missing data). |
||||||
|
VARIABLE NAME |
VARIABLE LABEL |
N |
Mean |
St. Deviation |
Minimum |
Maximum |
|
L_HT |
overall length |
233 |
16 1417167 |
4 8511686 |
6. 00 |
26. 76 |
|
CREUX |
draught |
274 |
2 0914964 |
0 6626511 |
0. 70 |
3. 46 |
|
TJB |
gross tonnage |
497 |
38 9603058 |
26 0614706 |
2. 33 |
133. 47 |
|
LARGEUR |
beam |
255 |
4 9809412 |
1 5224519 |
1. 76 |
8. 60 |
|
CV_MOT |
engine power |
497 |
238 4265594 |
137 6390468 |
26. 00 |
675. 00 |
|
NBANCONS |
age of boat |
340 |
15 1117647 |
10 6560262 |
2 |
68 |
|
|
year of construction |
340 |
1981.89 |
10 5660262 |
1929 |
1995 |
|
The boats in the sample have, for example, a length which varies from 6.00 to 26.76 m, with an average of 16.14 m. The standard deviation of 4.85 m indicates that the majority of the sample has a length ranging between 11.26 (16.14 - 4.84) and 20.99 m (16.14 + 4.84).At the same time, the distribution of the 497 boats by area and fishing method can be studied. The total of the lines and the columns describes the distribution of the sample between the various fishing areas used by Moroccan boats and the various fishing methods. The percentage by area, if stratified sampling has been carried out correctly, must reflect the percentage of the total fleet. This cross-tabulation makes it possible to study the relation between the two qualitative variables (area and fishing method) based on the distribution of individuals to the various cells of the table. |
||||||
| (CHAL: trawler, CHPA: trawler-longliner, CHSA: trawler-sardine boat, DIV: various, SARD: sardine boat, SECH: seiner-trawler, SEPA: seiner-longliner, PASA: longliner-sardine boat, PALA: longliner). |
|||||||||||
| TYPE |
|||||||||||
| CHAL |
CHPA |
CHSA |
DIV |
PALA |
PASA |
SARD |
SECH |
SEPA |
TOTAL |
% |
|
| REGION |
|
|
|
|
|
|
|
|
|
|
|
| 1. NADOR |
3 |
. |
8 |
. |
11 |
12 |
12 |
2 |
4 |
52 |
10.46 |
| 2. AL HOCEINA |
3 |
1 |
7 |
. |
10 |
9 |
16 |
. |
. |
46 |
9.26 |
| 3. TANGIER |
15 |
. |
3 |
1 |
35 |
13 |
21 |
. |
2 |
90 |
18.11 |
| 4. LARACHE |
2 |
. |
. |
. |
6 |
. |
14 |
. |
. |
22 |
4.43 |
| 5. KENITRA |
3 |
. |
5 |
1 |
1 |
2 |
2 |
. |
. |
14 |
2.82 |
| 6. CASABLANCA |
17 |
. |
5 |
1 |
8 |
9 |
19 |
. |
. |
59 |
11.87 |
| 7. SAFI |
20 |
3 |
4 |
4 |
12 |
1 |
8 |
. |
. |
52 |
10.46 |
| 8. AGADIR |
35 |
1 |
. |
. |
10 |
. |
30 |
. |
. |
76 |
15.29 |
| 9. LAAYOUNE |
20 |
. |
2 |
. |
5 |
. |
11 |
. |
. |
38 |
7.65 |
| 10.TAN-TAN |
17 |
3 |
3 |
2 |
7 |
2 |
14 |
. |
. |
48 |
9.66 |
| TOTAL |
135 |
8 |
37 |
9 |
105 |
48 |
147 |
2 |
6 |
497 |
100% |
| % |
27.16 |
1.61 |
7.44 |
1.81 |
21.13 |
9.65 |
29.58 |
0.40 |
1.21 |
|
100% |
| This table shows that the greatest number of boats is based in Tangier, with longliners and sardine boats being in the majority. The cross-tabulation illustrates a relationship between the area and the fishing method: in the north of Morocco there is a predominance of sardine boats (except in Tangier), whereas south of Casablanca, the coastal fleet is dominated by trawlers. |
|||||||||||
Statistical tables can be associated with graphs which make it possible to visualize the distribution as well as the relationship between variables; for example, it is possible to distinguish:
The histogram which represents the number of individuals, by class, for a qualitative (or semi-quantitative) variable;
The box plot which represents the quantiles of a quantitative variable;
The scatter diagram which plots the position of each individual according to two quantitative variables (a group of points whose shape indicates the degree of relationship between the two variables);
The straight line or curve which illustrates the function which connects two variables.
The graph makes it possible to explore the structure of the data quickly and to compare several data-sets. It is also used at the conclusion of a presentation to summarize and illustrate the values of a statistical table.
|
In the illustration below, the length of boat by licence-type, in a sample of the Moroccan coastal fleet, is compared on the basis of quantiles, and illustrated by box plots. The vertical axis corresponds to boat-length. Each box is delimited by the quartiles Q1 and Q3, whose variation Q3-Q1 corresponds to 50% of the vessels that are longer than the centre of the distribution. The horizontal line in the centre of the box represents the median value: if this line is in the middle of the box, it indicates that the distribution of the variable is symmetrical. The two ends of the vertical bars correspond to the values of the first and last percentiles (C1 and C99) and delimit 98 % of the sample distribution (= 497 boats); the points below or above the C1 and C99 percentiles correspond to the 1 % of the boats which have a value for this variable outside the distribution (e.g.: PALA or CHAL). The two extremes indicate the minimum and maximum values of the variable. For example, in this representation it can be seen that the "longline" fishing boats are smaller than the trawlers or the sardine boats, and that the seiner-trawlers are the most homogeneous group of boats in length (the box plot is small).
|
|
The comparison of the "type of fishing licence" frequency distributions between the sample and the total fleet (from which the sample is taken = boats in the file " licence 96") is illustrated by the frequency distribution histograms of the two data files (i.e. the 497 fleet sample individuals and 1,777 total fleet individuals partitioned amongst the 9 classes of the qualitative variable "fishing method"). The comparison between the two graphs shows that the sample over-estimates the sardine boats and underestimates the longliners compared to the information available in the master-file on licences operational in 1996. It would therefore be necessary to "rectify" or account for this bias in the sampling when conclusions are extrapolated to the entire coastal fishing fleet. Fleet Sample |
|
The evolution over a period of time of the number of boats that have acquired electronic equipment is illustrated by a cumulative frequency curve. Comparing the shapes of these curves for the various types of equipment makes it possible to illustrate the progress of modernization in Moroccan coastal fleets. It shows an acceleration in the 1990s of the acquisition of basic equipment (compass, VHF and sounder) - undoubtedly a result of incentive programmes - and the introduction of new equipment, such as the GPS, since 1995. |
The relationship between two variables, whether quantitative, qualitative or
mixed (1 quantitative and 1 qualitative), can be tested using statistical methods
(hypothesis testing). To interpret the results of the classification, and therefore,
to find the variables which explain significant differences between
the classes in the profile, 3 methods are normally used:
1. The Chi squared (c2) test, which makes it possible to see if there is a significant relationship between two qualitative variables, or to compare two distributions (for example, to compare the distribution of fishing methods, between the sample and the total population in the master file of licences from 1996);
2. The t-test of comparison between two averages, which makes it possible to compare the average of a quantitative variable between two groups;
3. and the Analysis of Variance test which makes it possible to compare the averages of two groups or more (an extension of the t-test).
In profiling fishing fleets, determining the structure of the profile consists of separating out those classes of vessels which are similar to each other from those which are different. Interpretation of the structure therefore consists of using statistical tests to find the variables which illustrate significant differences between classes. For qualitative variables, we compare the frequency distribution of the individuals, in the various modes of each variable, between the class and the whole sample. For quantitative variables, we compare the averages observed for the class and for the whole sample. These various indicators (frequencies and averages) are included in the tables representing the results of the typological analysis in order to summarize the variables that are characteristic of the classes. It is through the study of the values and the significant modes of the classes that it is possible to provide an interpretation of the class and to thereby validate the relevance of the typological profile.
|
A comparison of the distribution of fishing methods in the sample of 497 boats, and the target population of 1 777 boats licenced, was carried out using a c2 test. The results confirm that there is a significant difference between the sample and the population, due to an under-estimate of the number of longliners compared to sardine boat. Concurrently, we may compare the lengths of the boats in the total population with those in the sample, for each fishing method, in order to see if the sample is significantly different from the total population with respect to the size of the boat. The comparison of this quantitative variable between the various groups of boats - 7 types of fishing boats (the seiners are included in the miscellaneous category) in 2 files (licences 96 and sample = 14 groups) is carried out through an analysis of variance with two factors. Factor 1 is the group of boats associated with each of the two files (population/survey), and factor 2 is the fishing method. The statistical analysis shows that there is a significant difference, on the one hand, between the two datasets - based on the analysis of the probability that the value of the test is higher than a theoretical value called F (this probability must be lower than 5 % to demonstrate a significant difference between the compared groups). In this particular case, this significant difference between the population and the sample of boats is due to the relative importance of the sardine boats and the longliners. On the other hand, a significant difference is found between fishing methods; this confirms the differences suggested from the box plots in the previous graphical analysis of the data. However there is no significant interaction between the groups of boats in the two files and the fishing method (this interaction is labeled "method*file"), since the probability that the value of the test is higher than F is 0.2203. This probability is greater than 5 %, indicating that the sizes of the boats classified by fishing method in the sample are similar to the sizes of boats by fishing method in the total population. The analysis of variance model is significant overall: it explains 66 % of the total variability of the lengths observed for the boats for the various fishing methods, and the different files. The value of R2 makes it possible to evaluate the goodness of fit of the statistical model. RESULTS OF THE ANALYSIS OF VARIANCE FOR TWO FACTORS: Variable: LHT length |
|||||
|
Sources of |
degrees |
Sum of Squares |
Mean square |
F |
Pr > F |
|
|
|
|
|
|
|
|
Model |
13 |
112.22883042 |
8.63298696 |
172.53 |
0.0001 |
|
Error |
1151 |
57.59382612 |
0.05003808 |
|
|
|
Total |
1164 |
169.82265654 |
|
|
|
|
|
|
|
|
|
|
|
|
R-squared |
|
|
|
|
|
|
|
|
|
|
|
|
Sources of |
degrees |
Sum of Squares |
Mean square |
F |
Pr > F |
|
|
|
|
|
|
|
|
FILE |
|
3.32892686 |
3.32892686 |
66.53 |
0.0001 |
|
FISHING METHOD |
6 |
108.48631299 |
18.08105217 |
361.35 |
0.0001 |
|
METHOD*FILE |
6 |
0.41359056 |
0.06893176 |
1.38 |
0.2203 |
"Data Analysis" is a term for the array of statistical methods used for multidimensional (or multivariate) descriptive analysis. For the typology of fishing fleets, we use two types of methods: factorial analyses and automatic classification.
These methods, which are based primarily on a geometrical approach, make it possible to measure the resemblance, or the distance, between individuals and between variables, and to establish their degree of similarity. These similarities are visualized either by plotting a "cloud" of individuals (or variables) on a factorial plot, or by the shape of a dichotomous tree (dendrogram), whose success junctions illustrate the grouping of individuals. By slicing across the tree, the total population of the individuals under analysis can be partitioned, and interpreted according to the variables that are used in the analysis (active variables) and to the variables that are external to the analysis (additional variables). These partitioned groups, when interpreted, constitute the result of the typological profile: namely, the identification and the description of the elements within the various classes.
There are various methods of factorial analysis and classification, the choice of which depends on the characteristics of the data set being analysed, in particular the quantitative or qualitative nature of the data, and of the criterion which will be used to measure the relationship between the individuals or variables. To choose a method judiciously requires at least some knowledge of the theoretical basis of the techniques of data analysis.
Figure 2 shows the various stages in the exploration of the structure of the data by these methods. For the first stage, it is advisable to carry out a factorial analysis to explore the structure of the data, by studying the relationships between the variables and the resemblance between the individuals that are the subject of the typology.
Figure 3 illustrates, as an example, the result obtained from a factorial analysis designed to study the relationship between some qualitative variables (Multiple Correspondence Analysis). Four qualitative variables are used to classify strategies in a multispecies fishery: 1 - the target (signified by the catch profile of the fishing vessels), 2 - the period, 3 - the gear, and 4 - the fishing grounds. The factorial technique makes it possible to visualize the proximities between the various modalities of the variables (8 targets, 12 months, 3 gears and 28 fishing grounds).
In the same way, it is possible to analyse the plot of the individuals and to visualize the position of the fishing units on the plot of the variables.
The second step in typological analysis then consists of grouping the individuals by means of a classification algorithm or automatic partition. There are many classification algorithms, and their choice depends on the principle of agglomeration, and thus of resemblance between the individuals typifying the classes. Again, the choice of method requires some knowledge of the principles underlying the method.
|
The profile of the 497 boats of the Moroccan coastal fleets was carried out using four methods of multidimensional analysis:
Classification is carried out based on the factorial co-ordinates of the individual vessels on the principal factorial axes, in order to smooth the variability of the data and to obtain a classification tree (dendrogram) of the separate classes. Slicing the dendrogram makes it possible to define a partition of the individual vessels in a particular number of classes. This partition is then optimized by the moving centers algorithm which makes it possible to adjust, a posteriori, the individual boats in the classes in order to minimize within-cluster variability and to maximize between-clusters variability. This whole procedure, the factorial analysis, followed by classification, then partition, helps to reveal the underlying structure of the data in the table. |
Figure 2: Exploring the structure of the data by Data Analysis methods
Data-set
Factorial analysis
Classification -
Partition
Figure 3: Example of a factorial map
resulting from a Multiple Correspondence Analysis applied to qualitative
variables to carry out a typological analysis of fishing strategies (artisanal
fisheries of Kayar in Senegal in 1992)
The stages of analysis can be summarized in the form of a flow chart representing the different steps of the process, the data sets on which the analyses are carried out, with their size (i.e.: the number of individuals and variables), and the methods selected for carrying out the analyses. An example is presented below showing 3 processes of analysis of the fishing fleet profile, in respect of the tactics of fishing (Figure 4), the technical characteristics of the boats (Figure 5) and the strategies of exploitation (Figure 6).
|
Figure 4 illustrates the approach taken with the catch and effort data in order to produce a fleet profile based on fishing tactics, corresponding to the processes applied to the artisanal fisheries of Senegal and the trawl fisheries of the Celtic Sea. "Fishing tactics" refers to the choices made during a fishing trip or a particular fishing set or haul, where the fishing takes place, the duration, the fishing effort (gear used for a certain time) and the target species. The flowchart summarizes the sequence of methods used to carry out two successive classifications: the results of the first classification, based on target species, are used to build a second table of figures which gather together various variables of interest in order to identify fishing tactics. If all the trips by the vessels of the fleet turn out to be classifiable by their tactics, it is then possible to construct a third table which will illustrate the time spent by each boat utilizing various fishing tactics (a fishing calendar or schedule). |
|
Figure 5 illustrates the process of analysis used in producing a profile based on the technical characteristics of the boats of the Moroccan coastal fleets. The first stage consists of testing how representative is the sample of 497 boats used in the investigation compared to the population in the master-file covering the whole Moroccan fleet (i.e.: the 2 169 boats of the file "Armament" provided by the Direction de la Marine Marchande and the 1 777 boats of the file "Licence96" provided by the Direction des Pêches Maritimes et de l'Aquaculture. The second stage consists of describing each variable according to its basic statistics. The third stage is the fleet profile itself. All the variables involved in the analysis are quantitative, and the process of establishing the structure of the data set is accomplished by a Principal Components Analysis followed of an automatic classification based on the technical characteristics variables (117 variables from the first 4 files of the database: 1 - general characteristics and bridge equipment, 2 - deck equipment, 3 and 4 - fishing gear: trawl, seine, net, other). Structural interpretation consists of making statistical calculations for each class identified by the typological study, for all the variables of the data table. |
|
Figure 6 illustrates the second analysis applied to the 497 boats of the Moroccan fleet, this time in order to establish the relative importance of the various fishing strategies. File 5 of the database, entitled "Fishing operations", included 124 variables describing up to 4 different fishing operations carried out by each boat during 1995. Each operation in the survey questionnaire covers a fishing campaign described by the gear used, the fishing ground visited, the species captured and the period (the months at the beginning and end of the fishing season). A campaign includes all similar fishing trips from the point of view of gear, target species and fishing zone. By analysing the combination of variables [gear * species * zone * period] it is possible to see whether boats involved in different fishing campaigns use several strategies over the course of time, in particular the general-purpose boats with multipurpose licences. The analysis of exploitation strategies is based on the study of fishing campaigns and carried out in 3 successive stages: 1. A profile of fishing operations by campaign-type The fishing operations file is modified in order to produce records by fishing operation (one boat carrying out 4 different operations thus generates 4 records): a table is obtained with 1064 lines, one for each fishing operation. As each operation is described by 29 variables according to different criteria (catch, place, period), it is necessary as a precondition to balance the weight of these different criteria (weight function of the number of variables describing the criterion) before classifying the fishing operations. Each of the three criteria is finally expressed as a qualitative variable constructed from 3 successive classifications carried out on the 1 064 fishing operations. For each classification, the operations are described by the variables relevant to the topic considered, that is to say: - the list of the captured species, for the study of targets; Following these first 3 classifications, each operation is described by a species category, a zone category and a period category. The initial multivariate table is then synthesized, with 3 qualitative variables resulting from the 29 initial variables. The fishing campaign is described by the combination of these 3 new variables: zone fished, period, target, and by the variable gear. A Multiple Correspondence Analysis can be used to analyse this kind of table in order to determine resemblances (amongst the 1 064 operations) on the basis of the relationship between the modes (classes) of these nominal variables. A new classification makes it possible to clarify the "fishing-type" campaigns. 2. Construction of a profile of fishing schedules: Elucidating the exploitation strategies then consists of describing each boat according to its fishing schedule, that is to say the time spent on the different campaign-types identified in the preceding stage. The new matrix of data is based on the 497 boats, described by the number of months associated with the different campaign types. This table is subjected to a factorial analysis (PCA) and a new profile is built to highlight the classes of boats that are engaged in the same fishing activities during the course of the year. 3. Interpretation of exploitation strategies based on the variables in the other data files: Classes of boats using similar strategies are then analysed in relation to the entire available data set, particularly the technical specification of each boat. |
Figure 5: Process of analysis of the
profile of the Moroccan fleet based on technical specifications
Figure 6: Process of analysis of the
profile of the Moroccan fleet based on exploitation strategies
