Identifying in¯uential attributes in freight route/mode choice
decisions: a content analysis
Kevin Cullinane
a,*, Neal Toy
b aDepartment of Maritime Studies, Hong Kong Polytechnic University, Hum Hom, Kowloon, Hong Kong, People's Republic of China
b
University of Plymouth, Drake Circus, Plymouth, Deron PL4 8AA, UK
Received 8 December 1998; received in revised form 8 June 1999; accepted 21 June 1999
Abstract
The adoption of Stated Preference (SP) techniques in freight route/mode choice studies requires the identi®cation of the major modal attributes which in¯uence these decisions. While there is a need to limit the number of attributes and attribute levels in order that the number of combinations (decision alterna-tives) presented to respondents is at a manageable level, it is also important that these variables are ac-curately identi®ed and speci®ed. In the majority of empirical studies, methods employed for performing this task have tended to revolve around either focus groups, interviews, unscienti®c syntheses of previous studies or even merely the feel or hypotheses of the researchers. This paper explains and presents the ap-plication of a content analysis methodology to the (mostly Western) freight route/mode choice literature, thereby providing a formal approach to the identi®cation and justi®cation of the attributes that are to be utilised within (Stated Preference) SP experiments. The implications for attribute selection in empirical studies are discussed, with particular reference to the Eastern European context. Ó 1999 Elsevier Science
Ltd. All rights reserved.
Keywords:Freight; Mode choice; Attribute selection; Content analysis; Stated preference; Decision models
1. Introduction
This paper presents and discusses the results from an application of content analysis to the (mostly Western) route/mode choice literature. Such an approach provides a formal methodology
*Corresponding author. Tel.: +852-2766-5111; fax: +852-2764-3374. E-mail address:[email protected] (K. Cullinane)
1366-5545/99/$ - see front matter Ó 1999 Elsevier Science Ltd. All rights reserved.
for the identi®cation and justi®cation of the attributes which are hypothesised as in¯uential in freight route/mode choice and which are, therefore, to be utilised within SP experiments related to this choice.
Commonly used in other areas of social science research, content analysis involves the objective and systematic identi®cation of speci®c textual characteristics, yielding simple quantitative mea-sures that facilitate the testing of hypotheses and the drawing of inferences from an existing body of literature. As such, it provides a formal and systematic means of specifying a hypothetical set of testable attributes for freight route/mode choice SP experiments.
Following a brief overview of the research methodology of content analysis, the major ap-proaches to content classi®cation and enumeration are outlined and a particular system of analysis is selected, justi®ed and applied to the freight route/mode choice literature. The results of this analysis are presented and the implications for attribute selection in future empirical studies are then discussed, with particular reference to the Central and Eastern European context.
2. Objectives of content analysis
As is often recognised in the literature relating to SP experimental design, there is a need to limit the number of attributes and attribute levels in order that the number of combinations (decision alternatives) presented to respondents is at a manageable level. Pearmain et al. (1991) propose an upper limit of six or seven variables, with Jones et al. (1986) highlighting the im-portance of precisely identifying and specifying these variables. In the majority of empirical SP studies, however, the methods employed for the de®nition of variables (attributes) and their levels have tended to be arbitrary rather than systematic.
This paper applies content analysis as a systematic, quantitative, but yet relatively simple methodology for establishing the most important factors identi®ed in the extensive freight route/ mode choice literature which has been generated over many years.
In essence, there are three objectives for the application of content analysis conducted within this paper. First, to contribute to the optimisation of an SP experimental design intended for use in the later phases of a wider research project. It is hypothesised, though impossible to test at this point, that this approach will minimise the speci®cation error component in the route/mode split module of an overall freight demand model. Second, following the conclusion of the overall research study, the results of the content analysis provide a benchmark against which the results of the empirically derived attribute rankings in Baltic freight corri-dors can be compared. Finally, as already stated, the content analysis provides a quantitative perspective on the traditional literature review. It is this last aspect which is the primary focus of this paper.
3. Content analysis de®ned
carried out in accordance with speci®c rules or guidelines, each stage of which should be su-ciently well de®ned to conform to the scienti®c prerequisite of accurate replication, an important consideration in justifying the scienti®c integrity of the method (Adams and Schvaneveldt, 1991, p. 299).
First used as long ago as the 18th century in religious studies (Adams and Schvaneveldt, 1991), content analysis is now commonly applied in a wide range of social sciences where it is valued highly for the unobtrusive nature of its application. A full description of the historical develop-ment of the techniques involved and their application is provided in Babbie (1995) and Jupp and Norris (1993). Basically, however, by objectively and systematically identifying speci®ed textual characteristics, content analysis yields simple quantitative measures which allow the testing of certain hypotheses, the validation of results yielded by alternative approaches and the drawing of inferences.
Holsti (1968, p. 601) de®nes content analysis as:
``. . .any technique for making inferences by systematically and objectively identifying spec-i®ed characteristics of messages''.
In essence, as Weber (1990) implies, the central intent of content analysis is that of data re-duction. More speci®cally, Singleton et al. (1993, p. 381) state:
``The basic idea is to reduce the total content of a communication to a set of categories that represent some characteristic of research interest''.
Robson (1993, p. 273) describes content analysis more generally as being ``. . .akin to structured
observation''.
The choice of documents which provide the source data for a content analysis will depend on their availability, accessibility and relevance, as well as on their personal interest to the researcher. Nearly every form of written communication, however, has the potential to be included. Within the numerous studies where documents have been subjected to content analysis, Sarantakos (1993, p. 206), has identi®ed ®ve broad groupings by type±public documents, archival records, personal documents, administrative documents and of greatest relevance to the study undertaken herein, formal studies and reports. The use of academic texts as source material is also endorsed, amongst others, by Bouma and Atkinson (1995), Bailey (1994), Adams and Schvaneveldt (1991) and Singleton et al. (1993).
4. The methodology of content analysis
4.1. Preamble
must be decided upon (Sarantakos, 1993). There are numerous alternative methodological options with no single approach or method superior overall. Optimality in experimental design tends, therefore, to be very situation-speci®c.
The ultimate objective of this content analysis is to contribute to the design of an SP ex-periment by ranking freight journey attributes according to the cumulative/collective impor-tance placed upon them within the existing literature. Because of the nature of the existing literature, this produces a paradigm of freight route/mode choice which re¯ects the conceptual thinking of the Western world. The rankings of decision attributes derived from the content analysis can be compared against those derived from the results of an empirically applied SP experiment.
4.2. Research question and hypotheses
The starting point for a content analysis is the formulation of a research question. In this case, this is the identi®cation of the most appropriate, scienti®cally (objectively) derived variables for use in an SP experiment of freight route/mode choice. These variables will re¯ect, and emerge from, the dominant content categories identi®ed through the content analysis of a sample data-base of literature.
4.3. Sampling strategy
In sampling terms, the population under consideration is the existing freight route/mode choice literature. Useful reviews of this literature are provided by Matear and Gray (1993) and Gray (1982).
Any speci®c route/mode choice issue may form the principal theme of a particular source document. Alternatively, a particular source may address a number of issues. To make the content analysis more manageable and to isolate sections of the texts that are central to the area of re-search interest, a certain degree of selectivity is required. Bearing in mind the need to ``. . .preserve
the semantic coherence of texts as units'' (Weber, 1990, p. 43), the content analysis focuses only on those sections of the relevant source texts which pertain explicitly to route and/or modal choice/ split issues. Even more speci®cally, it should focus only on those parts of a source text where the importance of various trip attributes are considered.
The actual sampling strategies commonly employed in content analysis are similar to those utilised in other areas of social research and include simple random sampling, systematic sampling and strati®ed sampling. Berg (1995) and Babbie (1995) provide examples of how these sampling procedures have been applied in content analysis. As the most widely employed method, random sampling has been selected for this analysis.
The database comprises a variety of source literature. An attempt has been made, however, to limit the analysis to refereed academic output. This includes journal papers, conference pro-ceedings, textbooks, etc. The actual size of the sample has been determined by certain practical limitations on the accessibility and/or availability of sources, as well as by the usual ®nancial and time constraints. A total of seventy-®ve publications are included (due to its length, the complete listing of sources is available from the authors upon request).
4.4. Category construction
Categories need to be devised to provide the basis for classifying textual content. Collecting data through the application of content analysis implies identifying categories as they appear in certain contexts (Sarantakos, 1993). These categories are akin to the variables within a ``normal'' research setting. Category construction is widely recognised to be the area of content analysis requiring most consideration and having the greatest in¯uence over the results achieved (Berelson, 1971). The great variation and ambiguity within the literature as to what actually constitutes ``service'' poignantly illustrates the necessity of de®ning categories very carefully.
There are certain prerequisite characteristics of a categorisation system which must be con-sidered and incorporated at the research design stage (Holsti, 1968):
· Category validity.Categories must be created which intrinsically re¯ect the conceptual
frame-work of the research. This problem is resolved through consultation and iterative ®ne-tuning during the open coding stage (Bouma and Atkinson, 1995). In this study, the chosen categories re¯ect generic collections of very speci®c attributes that are considered in¯uential in freight route/mode choice decisions.
· Categories are exhaustive and mutually exclusive. Every relevant basic recording unit must be
classi®able. Equally, ambiguity in classi®cation is avoided because each basic recording unit should ®t into only one given category (Robson, 1993).
· Transparency. To meet the scienti®c requirement that results can be replicated, it should be
clear which recording unit is allocated to which category.
Since the development of categories is an iterative process, however, it is not necessary to specify the de®nitions of each concept at the outset of the analysis (Babbie, 1995). According to Bailey (1994, p. 307):
``Only by letting the categories emerge from the documents to be analysed can the goals of mutual exclusiveness and exhaustiveness be met. Categories constructed without prior in-spection of documents would no doubt exclude many important categories and include many that are super¯uous or unnecessary.''
An initial appraisal or formal pilot survey of a sub-sample of randomly chosen texts usually provides a starting point. The use of an open coding strategy allows every possible category to be included, with the speci®cation of each category becoming increasingly ®ne-tuned as the review becomes more extensive.
Sarantakos, 1993 and Babbie, 1995). This is because independent tests of coding a single sample of text are likely to reveal a certain level of ambiguity between researchers as to which concepts actually conform to a particular category construct. An ensuing revision of the category de®ni-tions is likely to eradicate any such ambiguities upon re-testing. Sarantakos (1993) proposes that if at least 80% of the variation is agreed upon between researchers, then the system can be fully operationalised. As broadly developed through the process outlined above, the categories adopted within this research are summarised in Table 1.
4.5. Units of analysis
The recording unit is the smallest body of text in which an example of one of the content
categories appears. Frankfort-Nachmais and Nachmais (1996) classify ®ve types of recording unit;
1. Single wordsor terms.
2. Themes± Even though the boundaries can be dicult to identify, in its simplest and most
prag-matic form, this amounts to a sentence.
3. Characters ± Not relevant in this study.
4. Paragraphs± These often contain more than one theme and, therefore, are not necessarily
mu-tually exclusive.
5. Items± These can include whole books, papers or chapters although the caveat concerning
mu-tual exclusivity still applies. An overall message (latent content), however, can usually be iden-ti®ed in the text.
In this study, data were collected by utilising recording units 1 and 5 above. A focus on single words as the recording unit means that each occurrence of the word within the text was recorded. This is the most commonly used system and has several advantages (Robson, 1993). Unlike a theme-based recording unit, for instance, each word is discrete and has clear boundaries. Since
Table 1
Category constructs and their relationship to underlying attributes, variables and terms derived iteratively from the literature
Category name Variables/Terms covered by category
Cost/Price/Rate Cost, Price, Rate Service (non-speci®ed) Service (non-speci®ed) Transit time reliability Transit time reliability
Frequency Frequency
Distance Distance
Speed Speed, Transit time, Terminal time, Transhipment time
Flexibility Flexibility, Convenient schedule, Non-speci®c extras, Pick-up and delivery Infrastructure availability Infrastructure availability, Accessibility
Capability Capability, Service availability, Equipment available, Capacity
Inventory Inventory
Loss/Damage Loss/Damage, Claims
Characteristics of the goods Type, Value, Value/weight ratio, Volume, Weight, Density, Shipment size Sales per year Sales per year
Controllability/tracability Controllability, Tracability
individual words are to be identi®ed and replicated by a second coder, this process is likely to exhibit greater reliability than when alternative recording units are used.
The mechanistic process of counting the frequency of occurrence of that which is physi-cally present (termed the manifest content) is a sound approach as far as reliability and replication are concerned. If due consideration is not given to the context in which the re-cording unit appears, however, reservations will exist over the extent to which inferences may be realistically drawn from the data. According to Robson (1993), in such instances, it is necessary to consider the context unit. This is de®ned as ``. . . the largest body of text that
may be examined when characterizing a recording unit '' (Frankfort-Nachmais and Nachmais,
1996, p. 327).
The second recording unit used was theitemwhich, in this case, described the whole document under analysis. In terms of the relative emphasis placed on each category, a judgement was made as to the overall message conveyed by a source document. In this way, the latent (underlying) content complements the mechanistic analysis of the manifest content. While providing a more holistic perspective on a source documentÕs entire content, analysis at this level is performed at the cost of reliability and speci®city.
The integration of both manifestand latentcontent within the same content analysis is a rec-ommended approach in order to take advantage of the inherent strengths of each level of re-cording unit and to mitigate the eects of their respective limitations (Holsti, 1968; Berg, 1995). Also, if the coding of bothmanifest and latent content is reasonably consistent, the conclusions drawn from both methods should broadly agree.
4.6. Systems of enumeration
Bailey (1994) and Weber (1990) provide detailed coverage of the advantages and disadvantages of other systems of enumeration, as well as the following four most commonly applied systems of enumeration:
· Time-Space System ±where the relative amount of column space given to each respective
cat-egory is measured;
· Appearance ± do categories appear in the context unit at all;
· Frequency ±where every occurrence of a recording unit is counted. Probably the most widely
applied system.
· Intensity ±generally employed when dealing with attitudes and values ± scales are utilised.
The use of the frequency system involves making two important assumptions: (1) that the frequency of a word or category is a valid indicator of its importance and (2) that each individual occurrence of a word or category is of equal importance or value (Singleton et al., 1993). Its use in this study is justi®ed, however, because:
· The intention is to establish aranking of attributes in¯uential in the route/mode choice deci-sion-making process. Therefore, an assessment of the relative, rather than absolute, frequency of each category within the database of literature is more appropriate. Such an approach reduc-es the possibility of any distortion caused by the ®rst assumption.
5. Results
Tables 2 and 3 show the summary statistics derived from a manifest content analysis of seventy-®ve articles contained in a database of literature deemed relevant to the subject of freight route/
Table 3
Summary results of manifest analysis using the appearance enumeration measure
Category Article appearances Percentage appearance rate Rank
Cost/Price/Rate 74 98.7 1
Service (unspeci®ed) 55 73.3 5
Transit time reliability 64 85.3 3
Frequency 37 49.3 10
Distance 50 66.7 6
Speed 71 94.7 2
Flexibility 49 65.3 7
Infrastructure availability 32 42.7 11
Capability 48 64.0 8
Inventory 28 37.3 14
Loss/Damage 48 64.0 8
Characteristics of the goods 57 76.0 4
Sales per year 12 16.0 15
Controllability/tracability 32 42.7 11
Previous experience 29 38.7 13
Total 686
Table 2
Summary results of manifest analysis using the word or term enumeration measure
Category Number of mentions Percentage of total mentions Rank
Cost/Price/Rate 1152 23.6 1
Service (unspeci®ed) 366 7.5 5
Transit time reliability 538 11.0 3
Frequency 149 3.1 11
Distance 222 4.6 10
Speed 705 14.5 2
Flexibility 226 4.6 8
Infrastructure availability 146 3.0 12
Capability 232 4.8 7
Inventory 100 2.1 14
Loss/Damage 263 5.4 6
Characteristics of the goods 377 7.7 4
Sales per year 30 0.6 15
Controllability/tracability 146 3.0 12
Previous experience 223 4.6 9
mode choice decisions.1 As can be seen, in terms of both the absolute number of category mentions and also in terms of simply the appearance of the category in the articles analysed, there is a striking degree of agreement between the results achieved under either method of enumera-tion. Indeed, at 0.92, SpearmanÕs rank correlation coecient is extremely high.
Both approaches to the manifest content analysis yield a high level of agreement not only as to what are the ®ve most often considered factor categories in the freight route/mode choice liter-ature, but also their ranking. Only from the ®fth ranked category onwards do the measures then yield results which diverge.
The inclusion of the ``service (unspeci®ed)'' category as the ®fth most common consideration in the literature is indicative of the sometimes abstract nature of the ``service'' element which, from the perspective of each individual decision maker, often encompasses extremely varied and, quite possibly, dierent speci®c service characteristics. It is also interesting to note that, of the top ®ve categories to simultaneously emerge from both these approaches to content analysis, it is this category (or attribute) which poses the greatest diculty in developing a quanti®able measure of its various levels. Unfortunately, the development of a suitable metric is, of course, a necessary prerequisite to SP experimentation. A further interesting result is that under both systems, the secondary range of categories receiving consideration in the literature are loss and damage, ¯exibility and capability.
By applying content analysis on the basis of an item unit of analysis where each article in the database is classi®ed according to its main theme or factor to emerge as in¯uential in freight route/ mode choice decisions, the results shown in Table 4 are achieved.
Under this much more subjective approach to content analysis, the method of enumeration necessitates the analyst exercising their judgement as to which category the primary theme of an article should fall. Where this decision has proven too dicult due to the inherent ambiguity of the central theme or simply because of the even-handedness with which multiple themes have been treated, the article in question has been ignored for the purpose of this analysis.
Under this measure of content, ``Service (unspeci®ed)'' emerges as the category given the greatest overall consideration within the analysed literature. This perhaps surprising result may re¯ect the diculty in subjectively classifying complete articles to particular speci®c categories. What is interesting, however, is that although the ranking of categories has clearly changed, four of the top ®ve categories resulting from the previous two approaches to content analysis are still present under this method. By applying the item-based method of enumeration, only the ``characteristics of goods'' category has dropped out of the top ®ve (to sixth position) to be re-placed by ``infrastructure availability''. Despite this similarity of results achieved, the rankings produced by this method have only a SpearmanÕs rank correlation coecient of 0.56 with those
1
produced using the word or term enumeration measure and of 0.69 with those based on article appearances.
As a ®nal test of the general consistency and validity of content analysis as a means of deriving the attributes which underpin SP experiments, a formal ``meta analysis'' of a subset of the da-tabase has been implemented (wolf, 1990). Hence, where documents positively de®ne factors as one of the top six in¯uential in the freight route/mode choice decision (perhaps even within a calibrated model), these references count towards an appearance-based enumeration of the lit-erature content. The results of this analysis are given in Table 5.
As one might expect from such an analysis, the results re¯ect the ease with which categories can be quanti®ed and subsequently included as variables in estimated models of freight route/mode choice decision making. Hence, the top three categories are transit time variability, speed and cost/price/rate. Although these categories correspond very closely to the important categories produced by all methods of content analysis, from this point on, rankings diverge between the dierent methods of enumeration employed.
In this case, ``Loss/damage'' and ``Capability'' take the other positions in the top ®ve. This meta-analysis approach to content analysis has a SpearmanÕs rank correlation coecient of 0.70 with the word (or term) basis of enumeration and a coecient of only 0.55 with the appearance-based measure. What is interesting, however, is that the rankings produced using this method of enumeration have virtually no correlation at all (a mere 0.09) with the item-based enumeration method. This is easily explained by the fact that this correlation coecient relates the rankings produced by (arguably) the most quanti®able approach to content analysis (meta-analysis; where categories are enumerated from appearances in calibrated models of the freight route/ mode choice decision) with those of an item-based enumeration which, by de®nition, is a much Table 4
Summary results of latent analysis using the item unit of analysis (reported only for articles where a dominant theme/ in¯uence could be identi®ed)
Transit time reliability 9 15.5 3
Frequency 0 0.0 12
Characteristics of the goods 3 5.2 6
Sales per year 1 1.7 10
Controllability/tracability 2 3.5 8
Previous experience 1 1.7 10
more subjective process that intends, and is likely, to capture the more abstract considerations in such decisions.
6. Conclusions
The results of the various forms of content analysis undertaken within this study generally con®rm what most would expect to be the most often considered facts or in¯uences in freight route/mode choice decision making. Note, however, that the quoted sources within any given database of relevant literature do not constitute a truly random sample; any single source is unlikely to be truly independent of others (i.e. not referred to by others). Therefore, some cir-cularity exists in the validation of this methodology. Nonetheless, content analysis provides a more logical and scienti®c basis for the justi®cation of generic categories of in¯uence over freight mode choice decisions, in contrast to the more usual, ad hoc selection of variables (attributes) to be tested.
Another result of this analysis is that the attributes most often included as in¯uential variables in SP experiments of freight route/mode choice were found to be most strongly con®rmed through the application of the more mechanistic, less subjective and rather more easily implemented ap-proaches to content analysis, such as those based on simple frequency counts of words or terms. The rather more abstract methods of content analysis that were tested did not correlate well with the simpler methods. It may be that the inherent diculty of quantifying abstract in¯uences means they are not easily incorporated into predictive models of route/mode choice decision-making behaviour. This does not negate their potential for explaining the disparities which Table 5
Summary results of meta analysis based on appearances of quanti®able factors and other model inputs
Category Number of appearances in models
Percentage of total appearances Rank
Cost/Price/Rate 54 14.3 3
Service (unspeci®ed) 1 0.3 14
Transit time reliability 85 22.6 1
Frequency 6 1.6 10
Distance 4 1.1 11
Speed 59 15.6 2
Flexibility 18 4.8 7
Infrastructure availability 0 0.0 15
Capability 41 10.9 5
Inventory 2 0.5 12
Loss/Damage 43 11.4 4
Characteristics of the goods 11 2.9 9
Sales per year 2 0.5 12
Controllability/tracability 17 4.5 8
Previous experience 34 9.0 6
invariably exist between the output of predictive models of freight route/mode choice and the actual practice of industrial decision makers.
In a study of freight route/mode choice decisions within the geographical context of the Baltic states (see Cullinane and Toy, 1998; Toy and Cullinane, 1998a and Toy and Cullinane, 1998b), initial research based on focus groups and interviews has suggested that in¯uences over freight route/mode choice decisions are likely to vary (at least in the weights attached to dierent attri-butes) from what is considered standard in the existing corpus of largely western literature. The relative in¯uence of those attributes identi®ed from a content analysis of the existing literature can be compared to the results derived from an East European case study to highlight any notable similarities or disparities and to assess any geographic context-speci®c considerations in modelling freight route/mode choice.
Acknowledgements
The authors are grateful to Professor W.G. Waters II and anonymous referees for helpful comments on an earlier draft of this paper.
References
Adams, G.R., Schvaneveldt, J.D., 1991. Understanding Research Methods, 2nd ed. Longman, London. Babbie, E., 1995. The Practice of Social Research, 7th ed. Wadsworths Publishing Co., New York. Bailey, K.D., 1994. Methods of Social Research, 4th ed. The Free Press, New York.
Berelson, B., 1971. Content Analysis in Communications Research. Hafner, New York.
Berg, B.L., 1995. Qualitative Research Methods for the Social Sciences, 2nd ed. Allyn and Bacon, Boston.
Bouma, G.D., Atkinson, G.B.J., 1995. A Handbook of Social Science Research, 2nd ed. Oxford University Press, United Kingdom.
Cullinane, K.P.B., Toy, N.R., 1998. Planned road network developments in the baltic sea region. Transport Reviews 18 (1), 35±55.
Frankfort-Nachmais, C., Nachmais, D., 1996. Research Methods in the Social Sciences. Arnold, London. Gray, R., 1982. Behavioural approaches to freight transport modal choice. Transport Reviews 2 (2), 161±184. Holsti, O.R., 1968. Content Analysis. In: Lindzey, G., Aronson, E. (Eds.), The Handbook of Social Psychology.
Addison-Wesley, Reading, MA.
Jones, P., Ferguson, D., Bradley, M., Ampt, E., 1986. Combining the Household Activity Approach with Stated Preference as a Tool for Policy Making: The Case of Peak Spreading in Adelaide, Ref. 343, Transport Studies Unit, University of Oxford.
Jupp, V., Norris, C., 1993. Traditions in Documentary Analysis, In: Hammersley, M. (Ed.), 1993. Social Research: Philosophy, Politics and Practice. Open University Press, United Kingdom.
Matear, S., Gray, R., 1993. Factors in¯uencing freight service choice for shippers and freight suppliers. International Journal of Physical Distribution and Logistics Management 23 (3), 25±35.
Pearmain, D., Swanson, J., Kroes, E., Bradley, M., 1991. Stated Preference Techniques: A Guide to Practice, 2nd ed. Steer Davies Gleave and Hague Consulting Group, The Hague.
Robson, C., 1993. Real World Research: A Resource for Social Scientists and Practitioner Researchers. Blackwell, Oxford.
Sarantakos, S., 1993. Social Research. Macmillan, Australia.
Toy, N.R., Cullinane, K.P.B., 1998a. The use of content analysis in attribute identi®cation for Freight Route/Mode choice SP experiments in Eastern Europe. Proceedings of the Universities Transport Studies Group Conference, University of Dublin, January.
Toy, N., Cullinane, K.P.B., 1998b. The use of content analysis in attribute identi®cation for Freight Route/Mode choice stated preference experiments in Eastern Europe. Proceedings of the International Symposium on Marine Technologies and Management, Ovidius University, Constanta, Romania, pp. 21±23.
Weber, R.P., 1990. Basic Content Analysis, 2nd ed. Sage Publications, London.