m end chains and laddering p an inventory … chains and laddering: an inventory of problems and an...
TRANSCRIPT
MEANS-END CHAINS AND LADDERING:AN INVENTORY OF PROBLEMS AND
AN AGENDA FOR RESEARCH
Working paper no. 34November 1995
MAPP
CENTER FOR MARKEDSOVERVÅGNING, -VURDERING OG BEARBEJDNING TIL FØDEVARESEKTOREN
CENTRE FOR MARKET SURVEILLANCE, RESEARCH AND STRATEGY FOR THE FOOD SECTOR
•T
HE
AARHUS SCHO
OL
•
O
FBUSINESS
MEANS-END CHAINS AND LADDERING:AN INVENTORY OF PROBLEMS AND
AN AGENDA FOR RESEARCH
Klaus G. GrunertThe Aarhus School of Business
Suzanne C. GrunertOdense University
Elin SørensenThe Aarhus School of Business
EXECUTIVE SUMMARY
1. Means-end chains are a device used to understand how consumers mentallylink products to self-relevant consequences. Means-end chains are usuallymeasured by a method called laddering.
2. Means-end chains may fall short of tapping all relevant aspects of howconsumers think about products. Specifically, nonverbal imagery, episodicinformation, and procedural knowledge are not included in means-end chains.
3. A number of methodological problems can be identified in the collection ofmeans-end chain data. Major problems, which should be addressed in research,are methods to elicit the product attributes the laddering is to start with, theintegration of a usage situation in the interview, and the basic decision on howmuch to direct the respondent.
4. Concerning the coding of laddering data, a higher degree of transparency ofthe coding process would be desirable.
5. Hierarchical value maps, a major graphic device used to visualise resultsfrom laddering studies, should take care of possible non-homogeneity ofrespondents and of the intricacies of aggregation following from it.
6. A catalog of research topics is developed that can guide studies aimed atimproving the means-end chains and laddering methodology.
Introduction 1
Means-end chain theory 1
Means-end chains versus other models of cognitive structure 3
Means-end chains and the explanation/prediction of behaviour 5
Collecting laddering data 7
Elicitation techniques and product stimuli 7
Situational specificity 9
Forked answers 9
When to stop, when to go on 10
Hard versus soft laddering 11
Coding of laddering data 12
Distinction between attributes, consequences, and values 12
Finding the ‘right’ level of abstraction 13
Increasing transparency and reliability in the coding of laddering data 13
Analysis 14
What is a hierarchical value map? 15
Determination of cutoff-level 15
Homogeneity of respondents 16
The non-redundancy assumption 18
Improved algorithms for deriving hierarchical value maps 19
The validation of results from a laddering study 19
A programme for research 20
References 23
INTRODUCTION
Laddering and means-end chains are one of the most promising developmentsin consumer research within the last decade. It is an approach which takesconsumers’ individuality seriously, but nevertheless comes up with quantitativeresults. It is rooted in a cognitive approach, and allows for emotional andunconscious (or, at least, semi-conscious) factors. It is intuitively appealing tothe practitioner, but has likewise attracted academic research.
Increased acceptance and use of a new approach inevitably leads to thedetection of unresolved issues and problems. Many of these unresolved issuesare related to the collection and analysis of laddering data. However, many ofthem also point at problems of a more theoretical nature. In this paper, wewould like to present some of the issues we regard as unresolved, and suggestresearch which could help solve these problems. The major part of the paper willdeal with methodological problems of the interview technique called laddering,of coding laddering data, and of analysing the coded data. However, as we alsowill attempt to show, methodological and theoretical issues are partlyinterlinked: Resolutions of methodological problems may require theoreticalprogress, or at least a clarification of some theoretical issues. We therefore startwith a discussion of some unresolved theoretical issues.
MEANS-END CHAIN THEORY
A discussion of means-end chain theory is made difficult by the fact that theepistemological status of means-end chains (MEC) is not completely clear. MEChave been used and discussed from the viewpoint of various research traditions,ranging from an interpretivist phenomenological view to a neo-positivist nomo-logical view. For the sake of simplicity, however, we distinguish two basic views(Grunert & Grunert, in press). We call them the motivational and the cognitivestructure view.
The motivational view is that laddering and MEC are concerned with obtaininginsight into consumers’ buying motives - ie, in the way basic motives are linkedto shopping behaviour. Laddering and MEC would then be a modern variant ofmotivation research in the Dichter (Dichter, 1960) tradition. Laddering can givevaluable insight by prompting consumers to reflect on their buying motives ina way not typical for daily shopping behaviour. Such insight is necessarilyqualitative in its character, and the kind of structures derived will besituationally constructed meanings. A theory in this context would be a set ofcategories useful for structuring laddering response data in such a way that theresearcher’s understanding of the consumer is improved.
The cognitive structure view is that MEC are a model of consumers’ consump-tion-relevant cognitive structure, ie, of the way consumption-relevant know-ledge is stored and organised in human memory. MEC would then be an ele-ment in a more complete cognitive theory of the consumer based on the generalcognitive view of human beings, as depicted in figure 1. According to the cogni-tive view, human beings analyse information obtained from the environment byrelating it to information already stored in memory, and use that information todirect behaviour towards the attainment of goals (Grunert, 1994; Simon, 1990).
1
MEC would then be a model of how consumption-relevant information is storedin memory, ie, about consumption-relevant cognitive structure. When supple-mented with theories or assumptions about the analysis of input from theenvironment, activating and adding to cognitive structure, and with theories orassumptions about the formulation of output, based on cognitive structure,MEC would become part of a theory with the aim of explaining and predictingconsumer behaviour.
(Adapted from Grunert, 1994)
The literature on MEC does not take a clear stand on which of these two viewsare being endorsed. The Grey Benefit Chain (Young, 1975), one of the earlyapproaches, seems to be most in the motivation research tradition. Asselbergs(Asselbergs, 1989), while adapting schema theory and therefore a cognitiveperspective, regards MEC generated by laddering not as measures of cognitivestructure, but as a ‘reconstruction’ of relevant information, which also tendstowards the first view. Gutman, Olson, and Reynolds (Gutman, 1982; Olson,1989; Olson & Reynolds, 1983; Peter & Olson, 1993; Reynolds & Gutman, 1988)adopt a cognitive structure perspective, saying that the hierarchical value mapderived from laddering data is “an aggregate map of cognitive structure” (Olson& Reynolds, 1983).
2
information
environment
analysis ofinput
human being
memory
formulation ofoutputquestion
Figure 1. The cognitive view of human beings
These two views are related to different general views on scientific analysis inconsumer research, as indicated above. If one adopts the interpretivist pheno-menological view that all measurements are context-dependent, the notion ofmeasuring a situation-invariant cognitive structure does not make much sense.If one, on the other hand, adopts a neo-positivist nomological perspective, aswould be typical for most of cognitive psychology, measuring only meaningwhich is dependent on the interview situation - which may include stimuliprompting the respondent to recall consumption situations during the interview- would appear unsatisfactory, because such measurements could not be takenas empirical estimates of constructs that can enter nomological propositions.
We will only deal with the second view in the following. It raises two questions:Why are MEC expected to be a better model of consumers’ cognitive structuresthan other possible models of cognitive structure? And how can MEC beintegrated into a more complete cognitive theory so that it becomes possible toexplain and predict behaviour?
Means-end chains versus other models of cognitive structure
The literature on cognitive psychology is replete with models of cognitivestructure. Most of them are variants of a basic network model1, ie, a modelconsisting of two types of elements: a) concepts, also called cognitive categoriesor nodes, and b) their associations, also called links. They can be categorised onthe following dimensions.
• Positional versus distributional (Anderson, 1983; Johnson-Laird, 1984;McClelland, 1985): Is a particular piece of information stored at any oneplace in cognitive structure, or is every piece of information distributedacross the whole structure?
• Episodic versus semantic (Tulving, 1972, 1985): Is the cognitive structure astore of general, semantic information, or is it a store of information aboutevents with a time-and-place tag?
• Verbal versus imagery (Paivio, 1986; Ruge, 1988): Is the informationstructured in verbal form or in such a way that it can at least be convertedinto verbal information by a lexicon, or is the information in images?
• Declarative versus procedural (Cowan, 1988): Is it information about facts,information which can be verbalised and explained? Or is it informationabout doing things, mental programmes which can be performed, but notnecessarily verbalised?
• Hierarchical versus non-hierarchical (Chang, 1986): Are cognitive categoriesorganised in the cognitive structure on a concreteness-abstractnessdimension, or not?
3
1 Naturally, there are alternatives to network models: schemas, frames, and scripts are some of thenon-network constructs that have been used to model cognitive structures. Discussing their relative merits anddeficiencies is, however, far beyond the scope of this article.
• Types of associations (Grunert, 1982a; Norman & Rumelhart, 1975): Doesthe cognitive structure distinguish between different types of associations -like set membership, causality, or type - and is there any syntax on whatmay be associated in which way?
It is not difficult to place MEC on these dimensions. First of all, MEC are in thefamily of network models, since they consist of nodes and links. They are posi-tional, because a node refers to a specific concept. They are semantic, becausethey depict general knowledge on products, their attributes and consequences,and not knowledge about individual usage events. They are verbal, because lan-guage is used both in measurement and in presenting the resulting models ofcognitive structure. They refer to declarative knowledge, because they only in-volve knowledge about states which can be verbalised, and not proceduralknowledge. They are hierarchical, because the cognitive categories in a MEC arelisted according to abstractness. And finally, they restrict associations to a par-ticular type, namely associations expressing causality: an attribute leads to aconsequence, and a consequence to a value.
It is harder to find arguments that support the fact that MECs are the mostappropriate model of cognitive structure in the context of consumer behaviour.MECs do have face validity, but it is not difficult to find arguments for othertypes of cognitive structure as well.
There is ample evidence that consumer behaviour is influenced by episodic infor-mation. Critical incidences, ie, previous key experiences with products and ser-vices which are clearly linked to a specific time and space, do influence intentionsto repeat a behaviour, without this information necessarily being converted intosemantic information. A single controversy with a hotel clerk, remembered clear-ly with regard to time and place, may lead to a decision never to visit the hotelagain. A spectacular airline crash makes people seek other airlines or avoid fly-ing altogether (for some time). Generosity of an outlet manager in exchanging adefective item may reinforce the decision to shop at this outlet again. A chicken-dinner induced salmonella infection may lead to avoiding chicken in the future.These are all instances of episodic information influencing consumer behaviour.
Non-verbal imagery has been shown to be a major component in how consumersstore information. Visual information enhances both encoding and recall, andalthough the issue is far from settled, much empirical evidence seems to suggestthat not all visual information is turned into a propositional form in memory,but is stored separately as mental imagery (Kosslyn, 1975; Pylshyn, 1973).Some consumer researchers have therefore advocated the use of pictorial sti-muli in researching consumers’ cognitive structures (Phillips, Olson, & Baum-gartner, in press; Ruge, 1988; Zaltman & Higie, 1993).
Consumer behaviour clearly draws on procedural knowledge. The way onereaches into a shelf to take an item which has been bought many times, theroute driven and the outlet visited in daily shopping, what one asks for at thebutcher’s counter may all be aspects of consumer behaviour which to a largedegree have become automated and hence procedural. Looking beyond shoppingactivities, procedural knowledge abounds in the use at home of the goods pur-chased: preparing meals, cleaning the house, getting rid of waste and garbageare all activities where procedural knowledge plays a large role.
4
Research on semantic memory has shown that knowledge may be organised ina non-hierarchical way (Chang, 1986). In the context of means-end chains, thiswould imply that in addition to A-C-V chains, also A-V-A-C chains or chainswith redundant links would be allowed. This issue is potentially accessible byreaction time experiments: the classic means-end chain would predict that veri-fying a statement about a link between an attribute and a value would take lon-ger than verifying a statement about a link between an attribute and a conse-quence. Contradicting results would be evidence of a non-hierarchical structure.
Finally, associations beyond subjective causality, for example about supplemen-tarity or substitutability of products, ie set membership associations, may beadded. The question is whether causality as a central guiding principle for orga-nising experience may be culture-specific, ie, mostly applicable to the Westerncivilisations.
Therefore MEC are clearly not a complete model of consumers’ cognitive struc-ture. They can be regarded as an excerpt of consumers’ cognitive structure, con-centrating on aspects which are regarded as relevant from a specific angle. Theproblem is that this angle has not been made very explicit in the literature.Theoretical research placing MEC into the broader context of models of cog-nitive structure would therefore be welcome. Such research should specify thetheoretical choices made in MEC theory, and provide arguments, founded in theliterature on cognitive psychology, that determine under which circumstancesthe specific aspects of cognitive structure captured by means-end chains are themost appropriate. Theoretical research resulting in propositions on how MECcould be broadened to include other forms of knowledge would also be helpful.
Means-end chains and the explanation/prediction of behaviour
As figure 1 shows, a model of cognitive structure cannot by itself explain orpredict behaviour; it has to be supplemented by assumptions about cognitiveprocesses (this argument is presented in more detail in Grunert & Grunert, inpress). MEC have not yet been integrated into a theory that includes suchassumptions, and therefore it has been difficult to evaluate the usefulness ofMEC as a tool to explain or predict consumer behaviour. If, however, MEC areto develop into a theoretical tool within the nomological research tradition, it isimportant to develop such a theory.
There are, of course, a lot of theories on consumer behaviour with a cognitiveorientation. If MEC are to be developed into a more full-scale theory, it is neces-sary to specify which range of behaviour the theory is supposed to explain, andwhy and how the theory is different from existing theories. Theory developmentis not the purpose of this paper, but we would like to suggest that the kind ofbehaviour to be explained is the choice between alternative courses of action,especially actions involving the purchase and/or use of products and services.We would also like to suggest that the theory of reasoned action, aimed atexplaining similar kinds of behaviour, can serve as a useful point of departureand of comparison. In the theory of reasoned action,2 the cognitive structure
5
2 We refer to the more well-known theory of reasoned action and not to its successor, the theory ofplanned behaviour, for reasons of simplicity of expositon only. See (Ajzen, 1985; Ajzen & Madden, 1986).
relevant for predicting behaviour is assumed to consist of sets of beliefs, whichare pairs of cognitive categories linked by an association. In one set of beliefs,courses of action are linked to consequences of those actions, and in another setrelevant others are linked to reactions to these courses of action. The linkbetween these two sets of beliefs and behaviour is explained by the well-knownformula
The theory of reasoned action thus takes two sets of factors as externally given:
• the excerpt of cognitive structure which is relevant for explaining/predictingthe behaviour in question, meaning the two sets of beliefs and theirstrengths (bi, nbi), and
• the motivation associated with the beliefs (ai, mci)
are assumed to be known and to be stable across the range of behaviour to beexplained. Given that these factors are known, the theory says that the inten-tion to take a particular course of action will vary with the sum of the motiva-tions weighted with the strengths of the beliefs. The theory of reasoned actionis not very explicit about which type of cognitive processes actually could bringthis about, but it has been shown that a spreading activation process in asemantic network can bring about results which correspond to this specificvariant of a linear model (Grunert, 1982b).
Relating back to the overall cognitive framework in figure 1, the theory ofreasoned action thus has two components: a model of cognitive structure and amodel of output formulation. The theory does not say why certain beliefs becomerelevant (or ‘salient’) in the context of a particular choice of courses of action,and it does not explain how the motivations determining the impact of thebeliefs on behavioural intention come about. In the description of the theory, itis explicitly acknowledged, however (Ajzen & Fishbein, 1980; Fishbein & Ajzen,1975), that characteristics of the choice situation will determine both whichbeliefs become salient and which motivations will determine behaviour. Theprocess in which the individual analyses the situation and activates a subset ofbeliefs from his/her cognitive structure is just not part of the theory. In theterms of figure 1, the theory of reasoned action has the overall structure
BEHAVIOUR = f[MEMORY, FORMULATION OF OUTPUT | ANALYSIS OF INPUT]
In MEC theory, neither the relevant excerpt from the cognitive structure nor themotivation is given. Rather, by measuring MEC a broader excerpt from cogni-tive structure is uncovered which is relevant across a large range of situations,and in any particular situation only a subset of it may become behaviourallyrelevant. Also, motivation is not assumed to be stable, but different values andconsequences may be more or less motivating in different situations. A theoryrelating MEC to behaviour thus would not only have to specify the formulationof output leading to behavioural intention once the relevant excerpt from cogni-tive structure and motivation is known, but also the analysis of input explaininghow, in a given situation and under given motivational constraints, certainparts of cognitive structure become relevant. If both types of processes would be
6
B BI w b a w nbmci i i i~ = +∑ ∑1 2
specified, it would become possible to predict behaviour contingent upon thesituation and the motivational state of the individual. We would then obtain amore complete cognitive theory, which also explains those factors which aretaken for given in the theory of reasoned action:
BEHAVIOUR = f[ANALYSIS OF INPUT, MEMORY, FORMULATION OF OUTPUT]
A few building blocks in constructing such a theory are already available. Rey-nolds (Reynolds & Perkins, 1987) has developed Cognitive Differentiation Ana-lysis, where respondents’ ratings of how well a product fits with the varioussteps in a ladder are used to explain product preference or product perception.In a similar vein, Bagozzi (Bagozzi & Dabholkar, 1994) has used regressionanalysis to relate the presence of certain links in respondents’ ladders to theirpast behaviour and to the two summary constructs of the theory of reasonedaction, that is, attitude towards the act and subjective norm. Both are morepragmatic solutions for a problem of analysis rather than attempts at realtheory building, but they can be interpreted as attempts to formalise theformulation of output (like behavioural intention or preference) based onexcerpts from cognitive structure which follow a means-end structure. Reynoldsfurthermore seems to suggest a principle for the analysis of input as well: hisresults from Cognitive Differentiation Analysis seem to suggest that preferencetasks activate the more abstract sections of means-end chains, whereasperceptual tasks seem to activate the more concrete ones.
In summing up, the theoretical problem addressed in this section is the lack ofa theory linking MEC to consumer behaviour. We have proposed that such atheory should both specify cognitive processes determining how situationalfactors lead to the activation of subsets of cognitive structure, and how thesesubsets then lead to the formation of behavioural intention. Theoreticalresearch proposing and/or testing such theories, possibly as an extension of thetheory of reasoned action, would make an important contribution.
COLLECTING LADDERING DATA
We would like to address four problems with regard to the execution ofladdering interviews: elicitation techniques and product stimuli, situationalspecificity, forked answers, and the decision when to stop probing.
Elicitation techniques and product stimuli
The first step in a laddering interview is to determine the product attributeswhich are to be the point of departure for the probing process. Depending onwhich attributes are elicited, the resulting ladders will differ.
The kinds of attributes elicited from respondents will depend on the retrievalcues provided to the respondent in the interview situation. Four major types ofcues have been used. In a free elicitation situation, the respondent is providedonly with the general product class as a retrieval cue, possibly supplemented bya usage situation (Olson & Muderrisoglu, 1979). In triadic sorting, usuallyconcrete products are presented to the respondent as cues (Reynolds & Gutman,
7
1988). In free sorting, respondents are provided with a larger number ofproducts, typically on a set of cards (Peter & Olson, 1993). In attribute selectiontasks, the respondent is provided with a list of possible attributes as cues.
This raises two questions. Will the set of attributes finally selected as thestarting point for ladders differ depending on which elicitation method is used?And if yes: which set of attributes is the ‘right’ one?
The answer to the last question obviously depends on which aim one has withthe study as such. If we retain an overall aim of laddering studies to measureexcerpts from cognitive structure which, in the context of a larger theory, wouldbe able to predict consumer behaviour, and especially choice behaviour whenchoosing between various alternative purchases, then the ‘right’ set ofattributes is the one comprising the attributes which tend to be used in makingdecisions between alternative products or services. Such attributes may be bothintrinsic and extrinsic and may have varying levels of concreteness. They maynot be attributes of the product at all. Consumers may feel a lack of competencein making choices and leave the choice of product to an agent (as when buyingtires for a car, for example). Consumers may decide not to buy a product becauseit is only available at an outlet which is not on their usual shopping route. Inboth cases the relevant set of attributes would be attributes of an agent orretailer, not of the products. Also, consumers may base decisions on attributes ofthe production process which are not mirrored in the final product, for examplewhen buying ethical food.
A triadic sorting task, with an emphasis on visible differences between pro-ducts, favours concrete intrinsic at the expense of extrinsic and/or less concreteattributes, and may therefore lead to the generation of irrelevant attributes (forexample, about the size or colour of products), which may then result in shortand irrelevant ladders. A free elicitation task may face the opposite problem: therespondent may generate abstract product attributes or even consequences,necessitating ‘backwards laddering’ in order to obtain complete chains. Also, ina free elicitation task, when the respondent is unable to generate attributesspontaneously, s/he will aid their retrieval process by framing the problem athand, for example by imagining a certain choice situation or a certain usagesituation. Since this framing is not necessarily communicated to the inter-viewer, a set of respondents may generate various sets of attributes referring todifferent situations, without the possibility of dealing with these differences inthe subsequent analysis of the data.
The problem with regard to elicitation techniques in a laddering interview istherefore that different techniques may lead to different sets of attributes beinggenerated, leading to the measurement of different excerpts from cognitivestructure, with no a priori way of knowing which technique will lead to the‘right’ result. It would be helpful to see research looking at how the elicitationmethod affects the resulting set of attributes, and how different elicitationmethods fare with regard to the relevance of the attributes generated, whererelevance should be linked to the extent to which the attributes are actuallyused in choice situations. Research testing new ways of eliciting attributes, withthe explicit aim of ensuring relevance, would also be welcome. One possible waythat may be explored is to start with a choice or preference task, and link theelicitation of attributes to an explanation of the preferences voiced. For example,
8
instead of asking for similarities and differences between products, one couldask the respondent to rank products according to preference and then relatethese preferences to attributes.
Situational specificity
One of the strengths of a cognitive structure approach as compared to a multi-attribute approach is that the situational dependency of attribute importance isexplicitly acknowledged, and is explained by the way in which attributes arelinked to consequences and values. This may have two possible consequences forthe way the laddering interview is conducted. Either one is interested in thecognitive structure with respect to a certain usage situation only; then it shouldbe made sure that it is this situation respondents have in mind during theladdering interview. Or one is interested in cognitive structure broadly, coveringthe main usage situations relevant for the product category in question. In thiscase, it becomes important that the interviews cover a variety of usagesituations. Depending on the main purpose of the study, the treatment ofsituational specificity in the interview should therefore differ. In the first case,the situation should be explicitly explained to the respondent, and it would bethe interviewer’s task to ensure that the respondent sticks to this situationthroughout the interview. In the latter case, it may be appropriate to start withan elicitation of situations before even starting to elicit attributes (which maybe situationally dependent, as explained above). It may even be advisable todistinguish real from ideal situations, in cases where consumers have apreference for a product but usually abstain from buying it due to socialinfluences or perceived lack of control (Ajzen & Madden, 1986).
The problem discussed here is that much of the information generated during aladdering interview will depend on the usage situations the respondents havein mind, while the interview technique in its standard form provides no meansof handling this situational specificity. It would be helpful to have researchinvestigating how differnt framings in terms of situations lead to differentresults of laddering studies. Based on such results, one could come up withsuggestions on how to make usage situations an integral part of the ladderinginterview.
Forked answers
In a laddering interview, it is assumed that respondents’ answers can bestructured according to a linear sequence of cognitive categories of increasinglevels of abstraction. This causes no problems as long as, at every step in theprobing process, the respondent retrieves from his/her memory only onecognitive category that appears to be a suitable answer. However, when therespondent has a quite elaborate cognitive structure with regard to the productin question, the retrieval process may result in the retrieval of several cognitivecategories, at the same level of abstraction. As an example, the probe “why isgood taste important to you?” may lead to the retrieval of three answers: a) “itmakes me feel relaxed and joyful”, b) “my family will be pleased”, and c) “itshows that I am a good housekeeper”. The interviewer has various ways ofhandling this. When the interview is conducted in a flexible way, the interviewer
9
may record the various answers and continue to ladder from each answer oneafter the other. If the interviewer insists on pursuing one ladder and urges therespondent to concentrate on one answer, the additional categories retrievedmay linger in working memory and interfere with later answers to furtherprobes, leading to deviations from the ideal of producing a stream of cognitivecategories with increasing levels of abstraction. The further probe “why is itimportant for you to please your family” may then result in an answer like “myfamily likes it when I am a good housekeeper”, which may be regarded as anattempt, from the respondent’s side, to pick up some loose ends in his/herworking memory and, at the same time, to come up with an answer that pleasesthe interviewer.
The problem discussed here is that respondents, in a laddering interview, mayfeel a natural tendency to come up with more than one answer to any particularprobe. While this phenomenon undoubtedly exists, little is known about howfrequent it is, which consequences it has for the results, and how various waysof conducting laddering interviews could affect it. Any research shedding lighton these matters would be useful. In addition, suggestions for new techniquesallowing forked answers and making use of them would be welcome as well,especially since the sequence in which answers at a given level of abstractionare given, and their inter-response times, may contain important additionalinformation on association strengths in cognitive structure.
When to stop, when to go on
When should an interviewer stop probing? When the respondent has reachedthe level of terminal values, it seems natural to stop probing. But what if arespondent has difficulty in finding further answers already at the consequencelevel, or even abstract attributes - and keeps saying, for example “I simply likesomething that tastes good”? How long should the interviewer press therespondent for additional answers? Interviewers in laddering studies havementioned this as one of the most difficult aspects of conducting laddering(Grunert & Grunert, in press).
When the cognitive structure is especially weak - there are few and weakassociations between cognitive categories - the respondent may soon be unableto retrieve additional categories in answering questions which probe for moreabstract categories. There is ample research demonstrating what happens insituations where a respondent has difficulties in retrieving more answers to aquestion, but is pressed to do so nevertheless (Gruenewald & Lockhead, 1980;Strube, 1984). The respondent will make a new attempt at retrieval by using adifferent strategic perspective. An example will make this clear. When arespondent has stated “being healthier” as a consequence of eating a foodproduct, and the interviewer asks “why is it important to you to be healthier?”,there may be no immediate answer. In trying to find an answer, the respondentcan use different strategic perspectives. S/he can go through recent illnesses,and how life would have been without them. S/he can consider, whether s/heactually wants to be healthier, and how health may compete with other higherorder values. S/he can try to imagine future life situations, where good healthwould be especially important, like going on a ski holiday or getting through aweek of stressful worklife. In each case, quite different cognitive categories maybe retrieved.
10
The stronger the impact of strategic perspectives on the answers given by therespondent, or, put another way, the more the task changes from a retrieval taskto a problem solving task, the more doubtful it becomes whether the interviewwill lead to some kind of measurement of the respondent’s cognitive structure.At best, the measurement will be affected by unknown strategic perspectives. Atworst, the respondent may actually construct new links between cognitivecategories, that is the respondent’s cognitive structure will be changed duringthe interview.
One could argue that these associations were there but had not been retrievedinto consciousness before the laddering interview. That is, the ladderingprocedure was successful in making explicit associations which seldom becomeconscious. This interpretation is impossible to refute. If we accept, however, thatlearning associations always require conscious awareness of the pair ofelements to be learned (a view which is widely accepted, see the review byHoffmann, 1994), then retrieval of a pre-existing association should beaccompanied by a feeling of recognition, and not of surprise.
The problem addressed here is whether ‘hard probing’ increases or decreasesthe validity of the results of laddering. To some extent, this problem is amenableto research as well. Change of strategic perspectives and/or construction of newassociations will be noticeable to the attentive interviewer as pauses, breaks,unfinished sentences, etc., and can also be detected after the interview byanalysing tapes. By reanalysing laddering data, omitting answers which therespondent gave only after considerable deliberation of that kind, one could firstof all gain some insight into how ‘hard probing’ affects results. When the resultsof a laddering study are used in the context of a theory to predict behaviour, onecould further check the predictive validity of the additional informationprovided by hard probes.
Hard versus soft laddering
Most of the problems discussed in this section, namely the possible elicitation ofirrelevant attributes, unclear situational dependence of the answers given bythe respondent, forked answers, and answers which have come about only byputting heavy pressure on the respondent, could be detected and possiblycircumvented by a trained interviewer, if the interviews are conducted in amanner encouraging a natural and redundant flow of speech, based on whichthe interviewer reconstructs ladders only after the interview. We may designatethis type of interview, where the respondent’s natural flow of speech isrestricted as little as possible, as ‘soft’ laddering. In contrast, ‘hard’ ladderingthen refers to interviews and data collection techniques where the respondentis forced to produce ladders one by one, and to give answers in such a way thatthe sequence of the answers reflects increasing levels of abstraction. Datacollection techniques that do not involve personal interviews at all, like self-administered questionnaires (Pieters, Baumgartner, & Stad, 1994; Walker &Olson, 1991; Young, 1975) and computerised data collection devices, are allexamples of ‘hard’ laddering.
11
Though the ‘soft’ approach is potentially better in handling the types of problemsdiscussed in this section, at the same time it leads to increased degrees of free-dom for the interviewer, which may introduce new biases. The interviewer musttry to make sense of the answers and to relate them to the means-end model.This requires interpretation, and, often, generalisation by the interviewer. Forexample, respondents may retrieve episodic rather than semantic information,which usually means that they start telling little stories (“last time I bought this,I noticed...”). People may give an answer and immediately start elaborating it (“Ilike bread with wrinkles. Not wrinkles generally, but, you know, the typewhich...”). People may jump back and forth between the levels of abstraction. Allthis requires cognitive processing on the interviewer side in order to distilladders: generalise from the episodic to the semantic, simplify the elaborate, andsort out levels of abstraction. The more this occurs, the more influence theinterviewer will have on the results. ‘Hard’ laddering is an attempt to avoid this.
It would be interesting to see research comparing the results of hard and softladdering. Should a test of convergent validity establish that both hard and softladdering lead to similar results by and large, one could safely conclude thathard laddering is a preferable technique, since it is easier to administer and lesscostly. Differing results would, however, call for an investigation of predictivevalidity, in the context of a larger theory as sketched earlier in this paper. Itwould be highly useful to have research that can pin down where hardladdering may be safe to perform hard laddering, and when it appearsnecessary to employ soft laddering.
CODING OF LADDERING DATA
We discuss two problem areas in coding: the distinction between attributes,consequences, and values; and the problem of finding the ‘right’ level(s) ofabstraction. This leads us to the general problem of increasing transparencyand reliability in the coding of laddering data.
Distinction between attributes, consequences, and values
The distinction between attributes, consequences, and values should, of course,be based on a conceptual definition of these terms. The laddering literature issurprisingly void of such definitions. In practice, many borderline cases turn up.When the respondent says “healthy” - is this an attribute or a consequence?“The bread is healthy” probably designates an (abstract) attribute. “I will behealthy when I eat this bread” seems to be a consequence. But, isn’t ‘health’ avalue? Making such categorisations in a uniform way is heavily dependent onthe availability of context information, ie, on the extent to which a term likehealthy is embedded in a redundant flow of natural speech. How much contextinformation is available will depend on how the laddering interview is executed.Laddering done by self-administered questionnaires or by computer (the ‘hard’form) usually provides very little context information. In laddering tasks wherethe interviewer records the answers as notes in ladder schemes, the contextinformation is available to the interviewer only and not in subsequent coding.Only when the interviews are taped and transcribed is the full context availablein coding.
12
It would be helpful, in any laddering study, to obtain information on thereliability of classifying answers as attributes, consequences, or values. It wouldbe even more helpful to see research showing how this reliability may dependon the way the laddering interview was conducted.
Finding the ‘right’ level of abstraction
The difference between two answers is rarely purely lexical. To define them assynonyms and group them into the same category, the category has to be at amore abstract level than the answers themselves. For example, ‘excellent taste’and ‘pretty good taste’ may both be sorted into ‘good taste’. ‘Good taste’ and ‘badtaste’ may both be sorted into ‘taste’. ‘No chemicals’ and ‘no preservatives’ maybe sorted into ‘only natural ingredients’. Such codings appear intuitivelyapparent, even though they obviously lead to a loss of information.
If the coding stops at that level, the resulting number of categories will usuallystill be large (easily 40-50 concepts). The implication matrix will be correspon-dingly large, the cell frequencies will be low, and it will not be possible to com-pute and draw a hierarchical value map, destroying the most appealing deviceof the laddering technique. Thus, a more radical coding is typically required toreduce the data (Gutman, 1991). This, usually, means that the level of abstrac-tion for each category has to be raised considerably. ‘No artificial colours’, ‘nopreservatives’, ‘more minerals’, ‘better ingredients’, ‘freshly milled flour’ allbecome ‘ingredients’. Many will still find this uncontroversial. At the conse-quences and value levels the real difficulties start. Can ‘joy’ and ‘not being de-pressed’ both be coded into ‘well-being’? Or ‘have experiences’ and ‘curiosity’ into‘variation’? Such rather broad categories usually have to be created, if theoutcome is to be a technically manageable implication matrix.
The problem, not specific to the coding of laddering data but common to manyforms of content analysis, is the lack of transparency of the coding process,leading to a low degree of intersubjectivity.
Increasing transparency and reliability in the coding of laddering data
It is obvious that coding is a complicated process that gives a lot of latitude to theresearcher, and much more attention should be paid to it in the typical methodo-logy discussion than has been the case so far. In particular, the process should bemade more intersubjectively accessible. Having parallel coders is of course themost common recourse used in research practice. But with laddering-type data,parallel coders may be a mixed blessing. The raw data used for the coding,usually laddering schemes or interviewer notes, already involves informationloss compared to the original interview. Context information, which may be help-ful or important when coding the data, has already been lost. The interviewer,who has conducted the interview, will be the best possible coder, because s/he willremember part of the context information (and may also better clarify mattersby referring back to a tape). A second coder, who does not have this backgroundinformation, may perform the coding in a different way, due to this lack of impli-cit context information. We would then observe a low inter-coder reliability, thereason being a difference in the context information available to the two coders.
13
The need for context to attach meaning to an answer refers to the generalproblem qualitative researchers call indexicality: For the researcher, it ispossible to understand, or make sense of a respondent’s answer only by relatingit to that respondent’s individual background. Answers must be interpretedrelative to the respondent’s background, experience, career, her/his interpre-tation of the data collection situation, etc. (Gutman, 1991; Hoffmann-Riem,1980; Hopf, 1978; Küchler, 1981; Schütze, Meinefeld, Springer, & Weymann,1973). The less one knows in advance about how respondents think about thetopic to be researched, ie, the less one knows about the product attributes,consequences, and values likely to be used by them to attach meaning to acertain product category, the more serious the indexicality problem. And themore context information available, the easier this process of assigning meaningwill be. Even though this argument is raised mostly from an interpretivistphenomenological perspective, the general issue of a valid assignment ofmeaning to data is relevant also from a nomological perspective.
Therefore it is easy to argue for the interviewers coding the data themselvesand for avoiding parallel coders. But this makes it all the more important todevise instruments which can make the coding process more transparent andreliable. Some of the experiences and tools developed within the realm ofcomputer-assisted content analysis (Grunert & Bader, 1986; Züll, 1992), whichis concerned with finding more intersubjective ways of coding text data, may behelpful here. The basic idea developed is that of iterative coding. This meansthat a first coding is performed, and the implications of this coding are madetransparent by aids like keyword-in-context lists, leftover lists, and insertion ofcodes into the text data base. Based on these aids, the coding is revised, and theimplications of the revised coding are analysed in the same way. This procedurecontinues until the coding appears satisfactory. Of course, the decision aboutwhat can be regarded as ‘satisfactory’ still rests to a large extent on face validityconsiderations and, therefore, on the judgement of the individual researcher.However, such procedures provide documentation as to how the codingproceeded, thus increasing the intersubjectivity of the process.
Applied to laddering data, one could imagine a windows-based software aid withthree parallel windows: one showing the raw text or the interview notes (for‘soft’ laddering only), one showing the ladders, and one in which synonyms aredefined, ie, where categories are formed. Both the step from raw text to laddersand from ladders to categories could be mouse-based, so that consequences of aparticular coding step become immediately visible in the context of allpreviously performed codings. The software could combine all entries into aparticular category with their text or ladder contexts for perusal, and couldprompt for coding at the discretion of the researcher.
The available software for coding laddering data is rather cumbersome, andwork leading to improved tools for this important step is called for.
ANALYSIS
The major tool in the analysis of laddering data has been the derivation of ahierarchical value map. We discuss the phenomenological status of hierarchicalvalue maps and then several technical problems in their derivation. It should
14
also be noted that alternatives to hierarchical value maps have been presentedas well: both multidimensional scaling (Aurifeille, 1991) and multiplecorrespondence analysis (Valette-Florence & Rapacchi, 1991) have beensuggested as alternatives. These techniques result in a representation wherethe cognitive categories are not linked in a network, but placed in a multi-dimensional space, where distances are used to express association. We do notdiscuss these alternative techniques further, because we believe that a networkrepresentation is more adequate to data based on a theoretical backgroundwhere cognitive structures are assumed to be modelled as networks.
What is a hierarchical value map?
The hierarchical value map, the main output from a laddering analysis, is acharacterisation of a group of respondents. There are two possible views, onemodest, one ambitious, of hierarchical value maps. The modest view is that ahierarchical value map is a device that shows the major results from a ladderingstudy of a group of respondents without having to go through all the individualladders. The more ambitious view is that the hierarchical value map is anestimate of cognitive structure for that group of respondents. While theladdering literature does not take a clear stand on which view one should adopt,the more ambitious view would be in line with much previous research onestimating cognitive structures, especially within the word associationparadigm (Deese, 1965; Szalay & Deese, 1978). The argument for the moreambitious view runs as follows: at the individual level, our data are not richenough to estimate a respondent’s cognitive structure. In a laddering study, the2-3 ladders typically obtained from an individual respondent reveal someaspects of his/her cognitive structure, but they are not an estimate of thecognitive structure itself, because the cognitive structure itself is not a collectionof single chains, but an interrelated net of associations. However, when weobtain ladders from a group of homogeneous respondents, then the set of laddersobtained from them, taken together and analysed by an appropriate algorithm,will yield an estimate of this group’s cognitive structure.
The view taken should determine which operations are regarded as admissiblewhen aggregating laddering data. In the following discussion, we assume themore ambitious view. However, the problems discussed will affect theinterpretability of hierarchical value maps also when only the modest view isadopted.
Determination of cutoff-level
Based on the implication matrix it is in principle possible to draw a map whichshows all the cognitive categories which resulted from the coding process, andin which two cognitive categories are linked whenever the corresponding cell inthe implication matrix has a non-zero entry. In practice, this is seldom possibleor desirable, because of the large number of non-zero entries in the typicalimplication matrix (often several hundreds). In practice, one tries to find ahierarchical value map that includes the ‘most important’ links. This is achievedby specifying a cutoff level.
15
The cutoff level gives the minimum cell entry in the implication matrix neces-sary to be represented as a link in the map. Since the distribution of the cellentries is usually heavily skewed - many cell entries are very low, while only afew are high - the cutoff level is a powerful device for reducing the complexityof the map. The problem is that there are no theoretical or statistical criteria toguide the selection of the cutoff level. Thus, one usually attempts some compro-mise between retaining information on the one hand and creating a manageablemap on the other hand.
Pieters, Baumgartner and Stad (Pieters et al., 1994) have suggested to selectthe cutoff level where the concentration index is highest. The concentrationindex is defined as the percentage of all links in a given implication matrix thatare retained at a given cutoff level, divided by the percentage of cells in theimplication matrix retained. A hierarchical value map based on a cutoff levelwith the highest concentration index will represent the highest possible numberof links in the data with the lowest possible number of categories. This isintuitively appealing, but it should be noted that the outcome of such aprocedure is heavily dependent on how rigorous the coding of the data has beenperformed. A coding procedure which retains many categories and thus asmaller loss of information during coding will lead to a cutoff level where onlya small percentage of active cells and of links are retained, resulting in consider-able information loss when constructing the hierarchical value map. A codingprocedure which results in only few categories, with a large loss of informationin coding, will lead to a cutoff level where comparatively many active cells andlinks enter the hierarchical value map, which then represents most of the infor-mation in the implication matrix.
There is thus a trade-off between information loss during coding and informationloss due to the use of a cutoff level in constructing the hierarchical value map.These relationships are only poorly understood, and the choice of cutoff-levelsrests mainly on rules of thumb. It would therefore be desirable to see researchcoming up with more rigorous methods for determining cutoff-levels, possiblybased on a quantification of several dimensions of information loss involved.
Homogeneity of respondents
An interpretation of the hierarchical value map as an estimate of the cognitivestructure of a group of respondents presupposes that the group of respondentsis homogeneous, or, more precisely, that their cognitive structure can beregarded as homogeneous with regard to the excerpt we want to measure.Ideally, one should test this assumption, before attempting interpretations ofthe hierarchical value map. Such tests could be conducted on the raw data,before deriving the hierarchical value map, or they could be conducted on thehierarchical value map itself.
There is only very limited methodological research giving advice on how that canbe done, and we can only point to a number of issues. The mere fact that the indi-vidual ladders differ does, of course, not constitute evidence of a lack of homo-geneity, if we assume a measurement model in which the production of laddersin the interviews is guided by a stochastic process. What we would need is a sta-tistical test on whether the differences between individual sets of ladders would
16
be compatible with such a random process. Not having such a test, a more prag-matic solution would be to perform a cluster analysis on the existence of links inthe ladders of individual respondents. If we obtain clusters with clearly distinctsets of links, it becomes intuitively less likely that the respondents are homogene-ous, and we may take that as a face validity test of the homogeneity assumption.
Roehrich and Valette-Florence (1991) have reported an example in whichladdering data served as input to a cluster analysis. Their clustering was basedon the existence of links between categories, and the unit of analysis, that is theunits to be clustered, were ladders, not respondents. Every respondent may thenbe a member of more than one cluster, so that, interpreted on the basis ofrespondents and not ladders, a set of overlapping clusters results. However, itshould be possible to conduct a similar procedure with respondents instead ofladders as the unit of analysis.
Looking not at the raw data, but at the hierarchical value map itself, the merefact that a certain path is based only on a subset of the respondents does not byitself constitute evidence of a homogeneity problem, for the same reason as above.If, however, one would find that the hierarchical value map can be divided intosubmaps so that the paths in subset A would be based on answers from subgroupI of the respondents, and the paths in subset B on answers from subgroup II, thenthis would indicate a homogeneity problem. The conceptual consistence indexdeveloped by Roehrich and Valette-Florence (1991) may be used as a diagnosticdevice in this context. For any path in a hierarchical value map, the index is thedifference between the highest frequency of any direct link in the path and thefrequency of indirect links between the start and the end node of the path.A highindex indicates that only few or no respondents had ladders including the wholepath, which has come about mainly by aggregating respondents with differentladders. A low or zero index indicates that the whole path mirrors ladders asvoiced by the respondents. The fictive hierarchical value map in figure 2 servesto illustrate how the index could be used to shed light on the homogeneity issue.
17
A1
V1 V2
C1
A2 A3
C25
5
6 4
2
55
1 6
6
7
Figure 2. Fictive example of hierachical value map
Conceptual consistency index:
A1-C1-V1 = 6-5 = 1A2-C1-V1 = 6-5 = 1A3-C2-V2 = 7-6 = 1A1-C1-V2 = 5-2 = 3A2-C1-V2 = 5-1 = 4
If the index is high for paths A1-C1-V1, A2-C1-V1 and A3-C2-V2, but low forpaths A1-C1-V2 and A2-C1-V2, then this seems to indicate that the hierarchicalvalue map merges two distinct groups of respondents, one characterised by pathsA1-C1-V1 and A2-C1-V1, and the other characterised by path A3-C2-V2.
The problem discussed in this section is the lack of clear criteria for whether ahierarchical value map may be taken as an estimate of cognitive structure for ahomogeneous group of respondents or not. A few suggestions for applying clus-tering techniques to shed light on group differences between respondents havebeen made, but clearly rigorous methodological research on the clustering ofladdering data is called for. The foundation really missing, however, is a sto-chastic measurement theory linking the production of ladders in the interviewto an underlying cognitive structure. Based on such a theory, statistical testscould be developed that allow testing of homogeneity assumption.
The non-redundancy assumption
The existing algorithms for deriving hierarchical value maps favour long chains.This goes for the paper-and-pencil method described by Reynolds and Gutman(1988), the programme LADDERMAP by Reynolds and Gengler, and the graph-theoretical method proposed by Valette-Florence and Rapacchi (1991). In terms ofcognitive structure theory, favouring long chains is identical with assuming non-redundancy of the cognitive structure. Non-redundancy means that, if category 1at abstraction level A is linked to category 2 at abstraction level B, which againis linked to category 3 at abstraction level C, then there should not be a direct linkbetween categories 1 and 3, because such a link would be redundant. Thisassumption, which is also called the economy-of-storage assumption, has beendebated for a long time in research on semantic memory (see Chang, 1986, for anoverview). There is experimental evidence both for and against it. But it is anassumption which is used in many successful models of semantic memory, andusing it in the context of means-end chain research may thus be defended. Prob-lems only arise in connection with non-homogeneity of respondents. If respondentI has a ladder 1-2-3, and respondent II has a ladder 1-3, then both underlyingcognitive structures may, at the individual level, conform to the principle of non-redundancy. But when both ladders enter the same hierarchical value map, aproblem arises: If, at the aggregate level, there is a link 1-2-3 (categories 1 and 3are linked indirectly), then there will be no link 1-3, even if such links wereobserved at the individual level. This can have rather astonishing results (thefollowing example is presented in more detail in Grunert & Grunert, in press).Imagine that, out of a sample of 30 respondents, 25 have the ladder good taste -well-being, and 5 have the ladder good taste - function better - well-being. In theaggregate map, the whole group of respondents would then be characterised bygood taste - function better - well-being, because a direct link between good tasteand well-being would be redundant. The map consequently communicates theerroneous impression that, for this group of respondents, good taste leads to thepersonal consequence that they can function better, with various marketing im-plications. Actually, most of the respondents just said that they enjoy a good taste.
The non-redundancy assumption is therefore a problem only when homogeneityis a problem as well. Whenever the homogeneity of the respondents is unclear,the non-redundancy assumption can lead to misleading characteristics of the
18
hierarchical value map. It may then be advisable to allow redundanthierarchical value maps.
Improved algorithms for deriving hierarchical value maps
In trying to develop better algorithms for the analysis of laddering data, the firststep should be an explicit stand on what a hierarchical value map is supposed todo. If it is meant as an estimate of cognitive structure, then the next step wouldbe to spell out clearly the assumptions made about the nature of that structure,especially about non-redundancy/economy-of-storage. The technical problem to besolved then is to aggregate only respondents whose cognitive structures can be re-garded as reasonably homogeneous with regard to the product in question. Thisshould be achieved by applying clustering methods to the laddering data beforeaggregation. If, on the other hand, the more modest view is adopted, ie, a hierar-chical value map is only a graphical device for the purpose of summarising mainresults from a laddering study, then the non-redundancy assumption should be re-laxed, and algorithms that allow redundant links should be developed. This maymake the maps a little harder to read, but it may avoid misinterpretations. Im-proved algorithms living up to these requirements, as well as software implemen-ting them, would be welcome additions to the tool box of laddering researchers.
THE VALIDATION OF RESULTS FROM A LADDERING STUDY
Laddering, being a qualitative data collection technique, is usually employedwith small to medium sample sizes. In many application contexts, the questionarises whether the hierarchical value map derived can be generalised to a lar-ger population, that is whether the results have external validity. A fewattempts have been made to validate results from a laddering study by quanti-tative data collection techniques.
Valette-Florence (Roehrich & Valette-Florence, 1991; Valette-Florence &Rapacchi, 1990) has used a card-sorting task, where first respondents wereshown a pile of attribute cards and had to select the most important attributefor the product in question. They were then shown a pile of consequence cardsand had to select the most important consequence following from the attribute.Eventually they were shown a pile of value cards and had to select the valuefollowing from the consequence. The procedure can be repeated with the second-most and third-most important attribute, if desired. It has the advantage thatrespondents produce ladders just as in a ‘real’ laddering interview, and theanalysis of the data by means of implication matrices and hierarchical valuemaps can proceed just as in a normal laddering study.
Vanden Abeele (1990) has presented whole chains, in verbalised form, to respon-dents (for example, “Milk is nutritious and full of vitamins and nutrients. It helpsstaying healthy, and healthy people can live a long and rich life”). Respondentshad to rate how well they thought the chain fitted the product in question. Forthose chains which were rated as best fitting, the various component links of thechain (“Milk is nutritious and full of vitamins and nutrients”/”Eating vitaminsand necessary nutrients helps staying healthy”/ ”Good health ensures a long andrich life”) were then rated for credibility.
19
Grunert (in press) has employed an extension of conjoint analysis to validatemeans-end chains. As in traditional conjoint analysis, respondents are presen-ted product profiles, which are systematically varied with regard to productattributes. These product profiles are then rated, not on a uni-dimensional pre-ference or purchase intention scale as in traditional conjoint analysis, but withregard to those consequences and values which previous qualitative researchhas shown to be associated with the product attributes. As a result, a covariancematrix of attributes, consequences, and values is obtained, and a hierarchicalvalue map can be estimated by structural equation estimation techniques.
The three examples presented are obviously preliminary research. They show aneed and some suggestions for a quantitative validation of studies employingthe laddering method. Research employing new methods or comparing thevirtues of these existing suggestions would be beneficial and will be importantcornerstones when the predictive validity of means-end estimations is to betested in the context of a broader cognitive theory, as suggested earlier.
A PROGRAMME FOR RESEARCH
In this paper we have tried to summarise a number of problems, boththeoretical and methodological ones, which have been identified with regard tomeans-end chain theory and the laddering method. The problems and theresearch questions they lead to are summarised in table 1. We conclude with afew general observations.
Firstly, the discussion has clearly shown how theoretical problems andmethodological problems are interrelated. Questions like attribute elicitation inthe interview or the optimal derivation of the hierarchical value map cannot besolved unless the theoretical status of both individual ladders and hierarchicalvalue maps is clarified. The underlying cognitive model of consumer behaviourhas to be spelled out, and a stochastic measurement model relating individualcognitive structures, ladders produced in an interview, and the hierarchicalvalue map as an estimate of cognitive structure has to be developed. This wouldalso allow for a more rigorous statistical treatment of many of the matterswhich are solved in ad-hoc ways now, like the clustering of respondents, and thedetermination of cutoff-levels.
Secondly, none of the problems discussed is unresearchable or inherently unsolv-able. Some of them can be addressed by straightforward empirical research, likethe comparison of different forms of attribute elicitation, the impact of situationalspecificity, or the differences between hard and soft laddering. Others require thedevelopment or adaptation of statistical techniques, or the development oradaptation of cognitive theories. If researchers could be attracted in sufficientnumbers to these issues, considerable progress seems possible within a few years.
Finally, we believe that the area of laddering and means-end chains, in spite ofthis catalogue of problems, has the potential of making a very significant con-tribution to consumer research. It could evolve into the cognitive theory of con-sumer behaviour, with a tool box of measurement devices developed to bridgethe gap between construct and reality.
20
21
Table 1. Problems and research topics in laddering and means-end chains
Topic area Problem Research task
Means-end Unclear which excerpts of Theory development placing MECchain theory cognitive structure is tapped into braoder model of cognitive
MEC and why this excerpt structure; specifying under whichis important circumstances MEC cover the most
relevant aspects of cognitive structure
MEC cannot be used to Development of theory combiningexplain/predict consumer MEC model with model of processesbehaviour for analysis of situational inout and
formulation of output
Collection of Choice of eliciation tech- Analysis of impact of type ofladdering data niques will affect the outcome elicitation task on set of attributes,
of a laddering study on relevarnce of attributes, on resulting ladders; development ofnew elicitation techniques
Usage situation respondents Analysis of how framing in termshave in mind affects outcome of situation affects results; develop-of study, but is not controlled ment of methods to treat situations
more systematically in laddering
Respondent may give forked Analysis of frequency of forkedanswers answers and their impact on results;
effect of various types of ladderingon forked answers
Unclear when interviewer Comparison of results based on moreshould stop probing and less extensive probing, predictive
validity of results
‘Hard’ versus ‘soft’ laddering Tests of convergent and predictivevalidity of results obtained by ‘hard’ and ‘soft’ laddering
Coding of Distinction between attri- Analysis of reliability of these classi-laddering data butes, consequences and fications and how they are affected
values not always clear by different types of laddering
Difficult to find ‘right’ level Develop methods to increase trans-of abstraction in coding parency and reliability of coding
process based on techniques ofcomputer-assisted content analysis
22
Topic area Problem Research task
Analysis Non-homogeneous Better methods for clusteringrespondents may lead to respondents based on laddering data;misleading results when development of stochastic measure-integrated into one HVM ment theory for laddering
No stringent criteria for Development of algorithm for choice of cutoff level in optimization of cutoff level basedconstruction of HVM on information loss functions
Algorithms favouring long Development improved algorithmspaths (non-redundancy for derivation of HVM includingeconomy of storaage) may options for non-redundancyproduce misleading resultsin HVM with non-homo-geneous respondents
Validation of External validity of results Comparison of approaches to results from from qualitative laddering quantify ladders; development ofladdering interviews difficult to assess new approaches for quantitative
by quantitative surveys surveys
Table 1. Problems and research topics in laddering and means-end chains, continued
REFERENCES
Ajzen, I. (1985). From intentions to actions: A theory of planned behaviour. In: J. Kuhl &J. Beckmann (Eds.), Action control: From cognition to behavior, pp. 11-39. Berlin: Springer.
Ajzen, I. & Fishbein, M. (1980). Understanding attitudes and predicting behavior.Englewood Cliffs, NJ: Prentice-Hall.
Ajzen, I. & Madden, T. J. (1986). Prediction of goal-directed behavior: Attitudes,intentions, and perceived behavioral control. Journal of Experimental Social Psychology,22, 453-474.
Anderson, J. R. (1983). The architecture of cognition. Cambridge, MA: Harvard UniversityPress.
Asselbergs, P. (1989 ). Competitive advertising - New developments in qualitativepositioning research: Meaning structure analysis. Rotterdam: IPM.
Aurifeille, J.- M. (1991). Contribution of ‘instrumental values’ to means-end chainsanalysis and to advertising conceptualization. In: K. G. Grunert & P. Valette-Florence(Eds.), Workshop on value and lifestyle research in marketing. Brussels: EIASM.
Bagozzi, R. P. & Dabholkar, P. A. (1994). Consumer recycling goals and their effects ondecisions to recycle: A means-end analysis. Psychology & Marketing, 11.
Chang, T. M. (1986). Semantic memory: Facts and models. Psychological Bulletin, 99, 199-220.
Cowan, N. (1988). Evolving conceptions of memory storage, selective attention, and theirmutual constraints within the human information processing system. PsychologicalBulletin, 104, 163-191.
Deese, J. (1965). The structure of associations in language and thought. Baltimore: JohnHopkins.
Dichter, E. (1960). The strategy of desire. New York, NY: Doubleday.
Fishbein, M. & Ajzen, I. (1975). Belief, attitude, intention and behavior. Reading, MA:Addison-Wesley.
Gruenewald, P. J. & Lockhead, G. R. (1980). The free recall of category examples. Journalof Experimental Psychology: Human learning and memory, 6(3), 225-240.
Grunert, K. G. (1982a). Informationsverarbeitungsprozesse bei der Kaufentscheidung: Eingedächtnispsychologischer Ansatz. Frankfurt: Lang.
Grunert, K. G. (1982b). Linear processing in a semantic network: An alternative view ofconsumer product eveluation. Journal of Business Research, 10, 31-42.
Grunert, K. G. (1994). Cognition and economic psychology. In: H. Brandstätter & W. Güth(Eds.), Essays in economic psychology, pp. 91-108. Berlin: Springer.
Grunert, K. G. (in press). Estimating consumers’ cognitive structures by conjoint analysis.MAPP Working Paper. Aarhus: The Aarhus School of Business.
Grunert, K. G. & Bader, M. (1986). Die Weiterverarbeitung qualitativer Daten durchcomputerunterstützte Inhaltsanalyse. Marketing - ZFP, 4, 238-247.
Grunert, K. G. & Grunert, S. C. (in press). Measuring subjective meaning structures bythe laddering method: Theoretical considerations and methodological problems.International Journal of Research in Marketing.
23
Gutman, J. (1982). A means-end chain model based on consumer categorizationprocesses. Journal of Marketing, 46(2), 60-72.
Gutman, J. (1991). Exploring the nature of linkages between consequences and values.Journal of Business Research, 22(2), 143-149.
Hoffmann, J. (1994). Unbewußtes Lernen: Gedanken zur weiteren Forschung.Psychologische Rundschau, 45(1), 43-46.
Hoffmann-Riem, C. (1980). Die Sozialforschung einer interpretativen Soziologie – DerDatengewinn. Kölner Zeitschrift für Soziologie und Sozialpsychologie, 32, 339-372.
Hopf, C. (1978). Die Pseudo-Exploration – Überlegungen zur Technik qualitativerInterviews in der Sozialforschung. Zeitschrift für Soziologie, 7, 97-115.
Johnson-Laird, P. N., Herrmann, D. J. & Chaffin, R. (1984). Only connections: A critiqueof semantic networks. Psychological Bulletin, 96, 292-315.
Kosslyn, S. M. (1975). Information representation in visual images. Cognitive Psychology,7, 341-370.
Küchler, M. (1981). Kontext – Eine vernachlässigte Dimension empirischerSozialforschung. In: J. Matthes (Ed.), Lebenswelt und soziale Probleme, pp. 344-354.Frankfurt: Campus.
McClelland, J. L. & Rumelhart, D. E. (1985). Distributed memory and the representationof general and specific information. Journal of Experimental Psychology: General, 114,159-188.
Norman, D. A. & Rumelhart, D. E. (1975). Explorations in cognition. San Francisco:Freeman.
Olson, J. C. (1989). Theoretical foundations of means-end chains. Werbeforschung &Praxis(5), 174-178.
Olson, J. C. & Muderrisoglu, A. (1979). The stability of responses obtained by freeelicitation: Implications for measuring attribute salience and memory structure. In: W.L. Wilkie (Ed.), Advances in consumer research, pp. 269-275. Ann Arbor, MI: Associationfor Consumer Research.
Olson, J. C. & Reynolds, T. J. (1983). Understanding consumers cognitive structures:Implications for advertising strategy. In: L. Percy & A. G. Woodside (Eds.), Advertisingand consumer psychology, pp. 77-90. Lexington, MA: Lexington Books.
Paivio, A. (1986). Mental representations: A dual coding approach. Oxford: OxfordUniversity Press.
Peter, J. P. & Olson, J. C. (1993). Consumer behavior, 3rd ed. Homewood, IL: Irwin.
Phillips, D. M., Olson, J. C. & Baumgartner, H. (in press). Consumption visions inconsumer decision making. In: F. Kardes & M. Sujan (Eds.), Advances in consumerresearch Provo, UT: Association for Consumer Research.
24
Pieters, R., Baumgartners, H. & Stad, H. (1994). Diagnosing means-end structures: Theperception of word-processing software and the adaptive-innovative personality ofmanagers. In: J. Bloemer, J. Lemmink, & H. Kasper (Eds.), 23rd EMAC conference 17-20May, 2 , pp. 749-763. Maastricht: European Marketing Academy.
Pylshyn, Z. W. (1973). What the mind’s eye tells the mind’s brain: A critique of mentalimagery. Psycolochical Bulletin, 80(1), 1-24.
Reynolds, T. J. & Gutman, J. (1988). Laddering theory, methods, analysis, andinterpretation. Journal of Advertising Research(2), 11-31.
Reynolds, T. J. & Perkins, W. S. (1987). Cognitive differentiation analysis: A newmethodology for assessing the validity of means-end hierarchies. In: M. Wallendorf & P.Anderson (Eds.), Advances in consumer research Provo, UT: Association for ConsumerResearch.
Roehrich, G. & Valette-Florence, P. (1991). A weighted cluster-based analysis of directand indirect connections in means-end chains: An application to lingerie retail. In: K. G.Grunert & P. Valette-Florence (Eds.), Workshop on values and lifestyle research inmarketing. Brussels: EIASM.
Ruge, H.- D. (1988). Die Messung bildhafter Konsumerlebnisse. Heidelberg: Physica.
Schütze, F., Meinefeld, W., Springer, W. & Weymann, A.. (1973). GrundlagentheoretischeVoraussetzungen methodisch kontrollierten Fremdverstehens. In: A. B. Soziologen(Eds.), Alltagswissen, Interaktion und gesellschaftliche Wirklichkeit, pp. 433-495.Reinbek: Rowohlt.
Simon, H. A. (1990). Invariants of human behaviour. Annual Review of Psychology, 41, 1-19.
Strube, G. (1984). Assoziation. Heidelberg: Springer.
Szalay, L. B. & Deese, J. (1978). Subjective meaning and culture: An assessment throughword associations. Hillsdale, NJ: Erlbaum.
Tulving, E. (1972). Episodic and semantic memory. In: E. Tulving & W. Donaldson (Eds.),Organization of memory, pp. 381-403. New York: Academic Press.
Valette-Florence, P. & Rapacchi, B. (1990). A cross-cultural means-end chain analysis ofperfume purchases. In: N. E. Synodinos, C. E. Keown, K. G. Grunert & T. E. Muller (Eds.),Proceedings of the third symposium on cross-cultural consumer and business studies,pp.161-172. Honolulu, HI: University of Hawaii.
Valette-Florence, P. & Rapacchi, B. (1991). Improving means-end chain analysis usinggraph theory and correspondence analysis. Journal of Advertising Research, 31(1), 30-45.
Vanden Abeele, P. (1990). A means-end study of dairy consumption motivation No. ECRegulation 1000/90-43ST). EC.
Walker, B. A. & Olson, J. C. (1991). Means-end chains: Connecting products with self.Journal of Business Research, 22(2), 111-119.
25
Young, S. & Feigin, B. (1975). Using the benefit chain for improved strategy formulation.Journal of Marketing, 39(3), 72-74.
Zaltman, G. & Higie, R. A. (1993). Seeing the voice of the customer: The Zaltman metaphoreliciation technique (Working paper No. 93-114). Marketing Science Institute.
Züll, C. &. M., Peter Ph. (1992). Textanalyse. Opladen: Westdeutscher.
26
MAPP PUBLICATIONS
MAPP working papers
No. 1: Grunert, K. G. & Baadsgaard, A. Market-based process and productinnovation in the food sector: A Danish research programme, January 1992.
No. 2: Thøgersen, J. Fødevareinnovation og emballage - miljøkonsekvenser ogforbrugerreaktioner, Marts 1992.
No. 3: Bonke, J. Choice of foods - allocation of time and money, householdproduction and market services, September 1992.
No. 4: Grunert, K. G. & Ellegaard, C. The concept of key success factors: Theory andmethod, October 1992.
No. 5: Harmsen, H. Determinanter for produktinnovationssucces, November 1992.
No. 6: Grunert, K. G., Nissen, L. & Wildenhoff, L. Do Danish food companies analysetheir competitors, February 1993.
No. 7: Bech-Larsen, T. Overvejer forbrugerne emballagens funktions- ogmiljøegenskaber, når de køber fødevarer? Februar 1993.
No. 8: Lassen, J. Food quality and the consumer, March 1993.
No. 9: Bonke, J. Choice of foods - allocation of time and money, householdproduction and market services, PART II, September 1993.
No. 10: Plichta, K. Technological opportunities and paths of development, September1993.
No. 11: Kvistgaard, M., Plichta, K. & Rasmussen, O. Den danske brødindustri -struktur, teknologi, forskningsbehov, Oktober 1993
No. 12: Grunert, K.G., Brunsø, K. & Bisp, S. Food-related life style: Development of across-culturally valid instrument for market surveillance, October 1993
No. 13: Hansen, J. K. & Sørensen, H. C. The importance of the price for the sale ofecological products, November 1993.
No. 14: Grunert, K. G. & Grunert, S. C. A comparative analysis of the influence ofeconomic culture on East and West German consumers' subjective productmeaning, November 1993.
No. 15: Langhoff, T. N. The internationalization of the firm in an interculturalperspective, November 1993.
No. 16: Grunert, K. G. & Brunsø, K. Market surveillance for the food industry,November 1993.
No. 17: Harmsen, H. Tendencies in product development in Danish food companies -Report of a qualitative analysis, February 1994
No. 18: Martensen, A. A model for marketing planning for new products, February1994.
No. 19: Bech, A. C., Engelund, E., Juhl, H. J., Kristensen, K. & Poulsen, C. S. QFood- Optimal design of food products, March 1994.
No. 20: Juhl, H. J. A sensory analysis on butter cookies - An application ofGeneralized Procrustes Analysis, March 1994.
No. 21: Andersen, E. S. The evolution of credence goods: A transaction approach toproduct specification and quality control, May 1994.
No. 22: Jensen, N. N., Grunert, K. G, Baadsgaard, A., Gede, M. P. Sales determinantsof canned pork products: A world-wide study, June 1994.
No. 23: Plichta, K. Technological development in the fruit and vegetable industry: Aproperty rights perspective, August 1994.
No. 24: Land, B. Ways of life analysis and food cultures, September 1994.
No. 25: Plichta, K., Rossen, L. & Skovboe, P. Den danske frugt- og grøntbranche,Struktur, teknologi, forskningsbehov, December 1994.
No. 26: Askegaard, S. & Madsen, T. K. European food cultures: An exploratoryanalysis of food related preferences and behaviour in European regions,September 1995
No. 27: Juhl, H. J., Poulsen, C. S., Kristensen, K., Bech, A. C. & Engelund, E.Structuring latent consumer needs using LISREL, October 1995
No. 28: Bech, A. C, Juhl, H. J, Kristensen, K. & Poulsen, C. S. Sensorisk analyse irelation til marekdsorienteret produktudvikling af fødevarer, Oktober 1995.
MAPP monographs
Søgaard, V. Farmers, cooperatives, new food products, May 1994.
MAPP conference papers
Baadsgaard, A., Gede, M. P., Grunert, K. G. & Jensen, N. N. Lagged life cycle structuresfor food products: Their role in global marketing, their determinants, and someproblems in their estimation, First International Multidisciplinary Conference on FoodChoice, Brussels, July 27-30, 1992.
Grunert, K. G. Towards a concept of food-related life style, First InternationalMultidisciplinary Conference on Food Choice, Brussels, July 27-30, 1992.
Jelsøe, E., Land, B. & Lassen, J. Understanding consumer perceptions and priorities withrelation to food quality, First International Multidisciplinary Conference on FoodChoice, Brussels, July 27-30, 1992.
Brunsø, K. Market surveillance systems and food related life style. The Sixth Colloquiumfor Doctoral Students in Marketing, Barcelona, May 23-25, 1993.
Plichta, K. & Harmsen, H. Studies of key success factors of product development success:A reinterpretation of results, 22nd Annual Conference of the European MarketingAcademy, Barcelona, May 23-28, 1993.
Thøgersen, J. Emballagens miljøbelastning. En mål-middelanalyse af forbrugernesproblemopfattelse og løsningsstrategier, Nordisk Netvärkskonferens: BusinessStrategy and the Environment, Gothenburg, October 7-8, 1993.
Lassen, J. & Jelsøe, E. Food production, food quality and the consumer. 16th AnnualScientific Meeting, AGEV Working Association for Nutrition Behaviour. Potsdam,October 14-16, 1993.
Grunert, K. G. & Brunsø, K. Food-related life style: results from a cross-culturalinvestigation. 16th Annual Scientific Meeting, AGEV Working Association forNutrition Behaviour. Potsdam, October 14-16, 1993.
Askegaard, S. & Madsen, T. K. European food cultures: An exploratory analysis of foodconsumption in European regions. Fourth Symposium on Cross-Cultural Consumer &Business Studies, Kahuku, Hawaii, December 15-18, 1993.
Harmsen, H. Improving product development practices: An action-research basedapproach. Meeting the Challenges of New Product Development. Manchester Schoolof Management, May 7-10, 1994.
Sørensen, E. & Grunert, K. G. Identification of managers’ perceived key success factors.The Second International Managerial and Organisational Cognition Workshop.EIASM, Brussels, May 26-27, 1994.
Juhl, H. J., Kristensen, K., Poulsen, C. S. Bech, A. & Engelund, E. Developing foodproducts in accordance with customer demands. The Second International ProductDevelopment Conference on New Approaches to Development and Engineering,EIASM, Gothenburg, May 30-31, 1994.
Harmsen, H. Product development practice in medium-sized food processing companies:Increasing the level of market orientation. The Second International ProductDevelopment Conference on New Approaches to Development and Engineering,EIASM, Gothenburg, May 30-31, 1994.
Brunsø, K. & Grunert, K. G. Identifying food-related life style segments by a cross-culturally valid scaling device. The IAREP/SABE conference, Rotterdam, July 10-13,1994.
Thøgersen, J. Recycling as moral behaviour. The IAREP/SABE conference, Rotterdam,July 10-13, 1994.
Ölander, F. & Thøgersen, J. Understandning of consumer behaviour as a prerequisite forenvironmental protection. The 23rd International Congress of Applied Psychology,Madrid, July 17-22, 1994.
Andersen, E. S. Innovation and quality standardisation: The evolution of complexindustrial systems and complex product designs. International Schumpeter SocietyConference, Münster, August 17-20, 1994.
Kristensen, P. S. & Holmen, E. Identification of a supplier network through QualityFunction Deployment. 10th IMP Conference, Groningen, September 30-October 1,1994.
Brunsø, K. & Grunert, K. G. Development and testing of a cross-culturally validinstrument: Food-related life style. The ACR conference, Boston, October 20-23, 1994.
Askegaard, S. & Madsen, T. K. Homogeneity and heterogeneousness in European foodcultures: An exploratory analysis. 24th EMAC Conference, Paris, May 16-19, 1995.
Thøgersen, J. German consumers’packaging preferences. A conjoint analysis. 24th EMACConference, Paris, May 16-19, 1995.
Bouchet, D. Embarrassment as a key to understanding cultural differences. Basicprinciples of cultural analysis. The Second Conference on The Cultural Dimension ofInternational Marketing, Odense University, Denmark, May 27-31, 1995.
Grunert, K. G., Sørensen, E., Johansen, L. B. & Nielsen, N. A. Analysing food choice froma means-end perspective. The ACR European Conference, Copenhagen, CopenhagenBusiness School, June 14-17, 1995
MAPP reprints
Askegaard, S. (1993). A European regional analysis of selected food consumptionstatements, In: W. F. Van Raaij & G. J. Bamossy (Eds.). European advances inconsumer research, Vol. 1. pp. 410-415. Provo, UT: Association for Consumer Research.
Askegaard, S. & Ludvigsen, H. H. (in press). European food cultures in a macro and microperspective: Implications for the marketing of Asian food products: In: J. Cote & S. M.Leong (Eds.), Asia-Pacific advances in consumer research, Vol 1. Provo, UT: Associationfor Consumer Research.
Askegaard, S. & Madsen, T. K. (1993). Fødevareforbrug - et regionernes Europa?Perspektiver for danske virksomheder. Ledelse og Erhvervsøkonomi, 57 (4), 233-248.
Bech, A. C., Engelund, E., Juhl, H. J., Kristensen, K. & Poulsen, C. S. (1993). Optimaltdesign af fødevarer, Ledelse og Erhvervsøkonomi, 57 (4), 199-211.
Bech-Larsen, T. (1993). Indgår fødevareemballagens funktions- og miljøegenskaber iforbrugernes købsovervejelser? Ledelse og Erhvervsøkonomi, 57 (4), 223-232.
Biemans, W. & Harmsen, H. (1995). Overcoming the barriers to market-oriented productdevelopment. Journal of Marketing Practice: Applied Marketing Science, 1(2), 7-25.
Bonke, J. (1993). Husholdningernes valg af fødevarer - strategier for rationel adfærd.Ledelse og Erhvervsøkonomi, 57 (4), 249-259.
Bouchet, D. (1995). Tvetydigheden i interkulturel kommunikation. Ledelse i Dag 4(3), 70-76.
Bouchet, D. (1995) Marketing and the redefinition of ethnicity. In: J. A. Costa & G. J.Bamossy (Eds.), Marketing in a multicultural world: Ethnicity, nationalism, andcultural identity. Thousand Oaks, CA: Sage Publications.
Bouchet, D. (1995). Det pinlige som nøgle til opfattelse af kulturelle forskelle. Ledelse iDag 17(1), 85-96.
Brunsø, K. & Grunert, K. G. (1993). Fødevarerelateret livsstil: et instrument tilmarkedsovervågning i fødevareindustrien, Ledelse og Erhvervsøkonomi, 57 (4), 213-221.
Brunsø, K. & Grunert, K. G. (in press). Development and testing of a cross-culturallyvalid instrument. Food-related life style. In: F. Kardes & M. Sujan (Eds.), Advances inconsumer research, Vol 22. Provo, UT: Assication for Consumer Research.
Godt, J., Kristensen, K., Poulsen, C. S., Juhl, H. J. & Bech, A. C. (in press). A consumerstudy of Danish entire male pigs. Fleischwirtschaft, 75.
Grunert, K. G. (1993). Towards a concept of food-related life style. Appetite 21, 151-155.
Grunert, K. G. (1994). Subjektive Produktbedeutungen: Auf dem Wege zu einemintegrativen Ansatz in der Konsumforschung. In: Forschungsgruppe Konsum undVerhalten (Eds.), Konsumentenforschung, pp. 215-226. Munich: Vahlen.
Grunert, K. G. (1995). Konkurrentenanalyse. In: R. Köhler, B. Tietz & J. Zentes (Eds.).Handwörterbuch des Marketing. pp 1226-1234. Stuttgart: Poeschl.
Grunert, K. G. (1995). Food quality: A means-end perspective. Food Quality andPreference.6, 171-176
Grunert, K. G., Bisp, S. & Brunsø, K. (1995). Competitor intelligence: Erhvervslivet kundehos CIA? Ledelse i Dag 4(4), 24-32.
Grunert, K. G., Brunsø, K. & Bisp, S. (in press). Food-related life style: Development of across-culturally valid instrument for market surveillance. In: L. Kahle & C.Chiagouris (Eds.). Values, lifestyles, and psychographics. Hillsdale, NJ: Erlbaum.
Grunert, K. G. & Ellegaard, C. (1993). The concept of key success factors: Theory andmethod. In: M. J. Baker (Ed.). Perspectives on marketing management, pp. 245-274.Chichester: John Wiley & Sons Ltd.
Grunert, K. G. & Grunert, S. C. (in press). Measuring subjective meaning structures bythe laddering method: Theoretical considerations and methodological problems.International Journal of Research in Marketing.
Grunert, K. G., Grunert, S. C., Glatzer, W. & Imkamp, H. (in press). The changingconsumer in Germany. International Journal of Research in Marketing, 12(4).
Grunert, K. G. & Brunsø, K. (1995). Identifying food-related life style segments inGermany by a cross-culturally valid scaling device. Schriftenreihe der AGEV, 10, 132-136.
Grunert, S. C., Grunert, K. G. & Kristensen K. (1994). Une méthode d'estimation de lavalidité interculturelle des instruments de mesure: le cas de la mesure des valeurs desconsommateurs par la liste des valeurs LOV. Recherches et Application en Marketing,5-28.
Grunert, S. C. (1995). Symbols and schemas in emotional eating. Ernährungs-Umschau42, 121-125.
Harmsen, H. (1994). Improving market oriented product development in Danish foodcompanies. In: P. C. de Weerd-Nederhof, I. C. Kerssens-van Drongelen & R. Verganti(Eds.), Managing the R&D process. Twente: Twente Quality Centre.
Jelsøe, E., Land, B. & Lassen, J. (1993). Do consumers have influence on food production.In: U. Kjærnæs, L. Holm, M. Ekström, E. L. Fürst & R. Prättälä (Eds.). Regulatingmarkets regulating people - On food and nutrition policy, pp. 123-136. Oslo: NovusForlag.
Kristensen, K., Kanji, G. K & Dahlgaard, J. J. (1992) On measurement of customersatisfaction, Total Quality Management, 3 (2), 123-128.
Kristensen, P. S.(1992). Flying prototypes: Productions departments' direct interactionwith external customers, International Journal of Operations & ProductionManagement, 12 (7/8), 197-212.
Kristensen, P. S. (1992). Product development strategy in the Danish agriculturalcomplex: Global interaction with clusters of marketing excellence, The Journal of Foodand Agribusiness Marketing (4) 3, 107-118.
Martensen, A. (1993). A model for marketing planning for new products. Marketing andResearch Today (November), 247-267.
Søgaard, V. (1994). Power-dependence relations in federative organizations, Annals ofPublic and Cooperative Economics (1), 103-125.
Thøgersen, J. (1994). A model of recycling behaviour. With evidence from Danish sourceseparation programmes. International Journal of Research in Marketing, 11, 145-163.
Thøgersen, J. (in press). The behavioural aspect of recycling. The roots of support andparticipation. In: P. Vogas (Ed.). Recycling: All the latest evolution, pp. XXX. Athens:Papazisis S. A.
Thøgersen, J. (1994). Recycling of consumer waste. A behavioural science approach toenvironmental protection policy. In: B. Bürgenmeier, (Ed.). Economy, environment andtechnology: A socioeconomic approach, pp. 51-73. Armonk: M. E. Sharpe, Inc.
Thøgersen, J. (in press). Wasteful food consumption - Trends in food and packaging waste,The Scandinavian Journal of Management.
Furthermore there are a number of project papers, which are not available to the public.