the validity of the means-end chain model of consumer

THE VALIDITY OF THE MEANS-END CHAIN MODEL OF CONSUMER BEHAVIOUR Final Report

Joachim Scholderer Klaus G. Grunert

Aarhus School of Business

2

MAPP project paper no. 05/04 ISSN 0907 2101 July 2004

3

CONTENTS

PREFACE 5

EXECUTIVE SUMMARY 6

INTRODUCTION 9

The means-end chain model in consumer research: Its origins and problems 9

THEORETICAL RECONCEPTUALISATION 11

Means-End Chains as Memory Structures 12

Means-end Chains as Motivational Processes 12

Categorisation and construction of meaning: the bottom-up route 13

Motivation and goal pursuit: the top-down route 13

Testable propositions 15

OVERVIEW OF EXPERIMENTAL WORK 17

Laddering interviews 17

Priming Experiments 18

EXPERIMENT 1 19

Method 19

Results 21

Discussion 24

EXPERIMENT 2 25

Method 25

Results 26

Discussion 27

EXPERIMENT 3 29

Method 29

Results 30

Discussion 31

4

EXPERIMENT 4 32

Method 32

Results 33

Discussion 34

EXPERIMENT 5 36

Method 36

Results 37

Discussion 38

EXPERIMENT 6 40

Method 40

Results 41

Discussion 42

GENERAL DISCUSSION AND CONCLUSIONS 44

REFERENCES 47

PREFACE

The research reported here was supported by the Danish Social Science Research Council (Statens Samfundsvidenskabelige Forskningsråd) through grant number 24-01-0170, “The validity of the means-end chain model of consumer behaviour”. Correspondence should be addressed to Joachim Scholderer, MAPP, Aarhus School of Business, Haslegaardsvej 10, DK-8210 Aarhus V, Denmark. Tel.: +45 89 486487. Fax: +45 86 150177. E-mail: [email protected]. The authors would like to thank Diana Dreier, Susanne Drachmann, Martin Grotrian, Kit Hagemann Petersen, Peter Skaaning Jørgensen, Jacob Møller, Janet Ross, Heine Bech Simonsen, Anette Thorndal, Stina Tylén, and Kamilla Wingfield for their invaluable contributions as project assistants, and, of course, all those who participated in the experiments.

Aarhus, July 2004

Joachim Scholderer Klaus G. Grunert

6

EXECUTIVE SUMMARY

Despite its popularity in consumer research, means-end chain theory suffers from problems of unconfirmed validity: the nomological status of its central construct, the means-end chain, is still unknown. The aim of the research reported here was threefold: (a) to reformulate means-end chain theory in a coherent theoretical framework, (b) to derive falsifiable predictions from the framework, and (c) to test these predictions by established experimental methods. Theoretically, means-end chains can be cast as associative networks with a three-layered structure. Four postulates can be formulated that impose testable restrictions on the layered network structure: hierarchicity, automatic spreading activation, bidirectionality, and self-relevance. The predictions were tested in altogether six experiments. The basic methodology was the same in all experiments. Two sessions were held with each participant. In a pilot session, each participant completed four different laddering tasks. Each task consisted of four different consumer products varying on three different attributes. After the pilot session, the word material that participants had generated in the laddering task was entered into a database. Individualized stimulus sets were then generated from the database for use in the second session. To avoid carry-over effects, the second session was arranged after a long delay. Each participant completed a sequential priming experiment in which single-presentation lexical decision tasks were used. Experiment 1 (N = 90) was designed to test the hierarchicity and self-relevance postulates. Hierarchicity was tested by examining whether response facilitation effects were higher when primes and targets were directly associated nodes in a means-end chain (attributes and consequences, or consequences and values) than when primes and targets were indirectly associated through a mediator (attributes and values, mediated by consequences). Self-relevance was tested by examining whether response facilitation effects were stronger when primes and targets were taken from a person’s own means-end chain (as measured by the laddering method) than when taken from another person’s means-end chain or from a standardised word list. In this experiment, only bottom-up priming of means-end chains was investigated.

7

Experiment 2 (N = 91) was designed to test the bidirectionality postulate. Bidirectionality was tested by examining whether the results obtained in Experiment 1, where bottom-up priming of means-end chains had been used (attribute, consequence, value), would remain stable when the direction of priming was turned around to top-down priming (value, consequence, attribute). Experiment 3 (N = 30) was designed to ensure that Experiments 1 and 2 were internally valid. In both experiments, the word material representing means-end chains had been elicited by means of a laddering task where different foods had served as product examples. Hence, the generated material shared a common associative context. To test the alternative explanation that a generalized activation of this common associative context might have been responsible for the priming effects observed in Experiments 1 and 2, and not specific activation of particular means-end chains, Experiment 3 replicated the previous experiment, but used only stimulus materials from a food context, i.e. also in the standardised word-list conditions against which all priming effects were benchmarked. Experiment 4 (N = 120) was designed to test the automaticity postulate. Automaticity was tested by examining whether the hierarchicity and self-relevance effects observed in Experiments 1 could be replicated under conditions designed to suppress controlled information processing. In line with standard procedures, short inter-stimulus intervals and a high proportion of fillers and non-words in the word material were used for this. Experiment 5 (N = 65) was designed as a second test of the automaticity postulate. The postulate was tested by examining whether the bidirectionality effect observed in Experiment 2 could be replicated under conditions designed to suppress controlled information processing. Again, short inter-stimulus intervals and a high proportion of fillers and non-words were used to induce automatic information processing. Experiment 6 (N = 30) was designed, in analogy to Experiment 3, to ensure that Experiments 4 and 5 were internally valid. To test the alternative explanation that generalized activation of a common associative context might have been responsible for the priming effects, Experiment 6 replicated the automaticity conditions of the previous experiment, but used only stimulus material from a food context, i.e. also in the standardised word-list conditions against which all priming effects were benchmarked. Overall, only few of the predictions were met. Hierarchicity, the assumption that means-end chains have a three-layered chain structure (as opposed to a non-

8

hierarchic, single-layered network structure), could only be established in one out of six experiments. Automaticity, the assumption that spread of activation through a means-end chain would still occur when controlled information processing was suppressed, could indeed be established in three out of three experiments that induced automatic information processing. Results for bidirectionality, the assumption that bottom-up priming effects would be mirrored by top-down priming effects, were favourable. Self-relevance, the assumption that spreading-activation effects would be stronger for means-end chains generated by participants themselves than for means-end chains generated by other participants (strong self-relevance) or materials taken from a standardised word list (weak self-relevance), could partially be established. Evidence for strong self-relevance was found in two out of six experiments, whereas evidence for weak self-relevance was found in four out of six experiments. The results raise new questions concerning the theoretical foundations of means-end chains. It can be concluded that means-end chains, as conventionally measured by the laddering method, are firmly anchored in people’s memory, but not as firmly as originally hypothesized. The hierarchicity assumption appears to be particular problematic. It appears that the association structure is non-hierarchic, displaying properties of a single-layered network with high associative redundancy. Three areas were identified where additional theoretical work is needed: how means end-chains enter more complex cognitive structure, the cognitive processes leading to the elicitation of means-end chains in laddering interviews, and the spreading activation processes explaining the retrieval and use of means-end information.

9

INTRODUCTION

The means-end chain model in consumer research: Its origins and problems

Rational choice approaches to consumer behaviour rely on the assumption that consumers have well-defined goals. Faced with alternative options, consumers are thought to use these goals as subjective standards against which the properties of the options are evaluated. Since goals and their attainment are subjective in nature, the generic concept “utility” is usually used as a common scale onto which all subjective evaluations are mapped. Whilst much research on consumer decision-making appears to be satisfied with this abstraction, it does not provide much guidance to practitioners trying to find out which products to develop, how to position them, and how to communicate about them in a way which will make them attractive to consumers. In the marketing area, it has therefore become common to employ psychological approaches in an attempt to obtain deeper insight into how consumers perceive, form preferences for and make choices between alternative purchasing options. Means-end chain theory is a prominent example of such an approach. It was developed in the 1980s in the area of brand communications in marketing (Gutman, 1982; Olson, 1989; Olson & Reynolds, 1983; Zeithaml, 1988). The basic tenet of means-end chain theory is that consumers perceive utility in a product only to the extent that they expect the consumption of this product to lead to self-relevant consequences, which in turn derive their importance from the extent to which they help consumers attain personal life values. Put another way, consumers are assumed to perceive utility in a product to the extent they perceive the product to be a ‘means’ to attain an ‘end.’ The way in which consumers subjectively link characteristics of a product to their personal ends is called a means-end chain. As an example, a product characteristic like “organic” in a food may, in the mind of the consumer, be linked to the consequences “good for the environment” and “good for personal health”, which may in turn be linked to the values “responsibility” and “quality of life” (Figure 1). Research employing the means-end chain approach has mainly followed three streams. There has been methodological research dealing with ways of measuring consumers’ subjective means-end chains (Aurifeille & Valette-Florence, 1995;

10

Botschen & Thelen, 1998; Grunert & Grunert, 1995; Hofstede, Audenaert, Steenkamp & Wedel, 1998; Roehrich & Valette-Florence, 1991; Valette-Florence & Rapacchi, 1991). There has been descriptive research employing the means-end approach and its associated measures (Bredahl, 1999; Grunert, Lähteenmäki, Nielsen, Poulsen, Ueland & Åström, 2004; Nielsen, Bech-Larsen & Grunert, 1998; Nielsen, Sørensen & Grunert, 1997; Valette-Florence, Sirieix, Grunert & Nielsen, 2000).

Figure 1. Hypothetical means-end chains for the product attribute ‘organic’ And there has been considerable applied research employing the means-end approach to derive market communication strategies (Gengler & Reynolds, 1995; Jolly, Reynolds & Slocum, 1988; Reynolds & Craddock, 1988; Reynolds & Gengler, 1991; Reynolds, Gengler & Howard, 1995; Reynolds & Gutman, 1984; Reynolds & Rochon, 1991; Reynolds, 1995) and, to a lesser extent, new product concepts (Herrmann, 1996; Søndergaard, in press). All this research takes the basic assumptions of the means-end chain approach for granted. Surprisingly, these basic assumptions have never been put to an empirical test. We believe there are three reasons for this. Firstly, the means-end approach seems to have good intuitive plausibility for many, which may make an empirical test of the basic assumptions less urgent. Secondly, these basic assumptions are often formulated in relatively loose terms, which prevents rigorous empirical testing. In particular, the nomological status of the central construct, the means-end chain, is unknown (Grunert, Beckmann & Sørensen, 2001). Whilst some researchers interpret means-end chains as memory structures (Gutman, 1982; Reynolds & Gutman, 1988), others see them as models of motivation (Cohen & Warlop, 2001; Pieters, Allen & Baumgartne,r 2001). Hence, a common conceptual

Organic

Quality of lifeResponsibility

Good forpersonal health

Good for theenvironment

Personal values

Consumption consequences

Product attributes Organic

Quality of lifeResponsibility

Good forpersonal health

Good for theenvironment

Personal values

Consumption consequences

Product attributes

11

framework is lacking that would help the development of coherent and testable theory (Grunert & Grunert, 1995). Third, the methods employed by means-end chain researchers impose a presupposed structure on the data. Means-end chains are usually measured by a semi-qualitative technique called laddering (Grunert & Grunert, 1995). While this technique comes in many variants, they all share the characteristic that they force respondents to elaborate their answers in an increasingly abstract way. A particular sequence of cognitive categories is assumed (attributes-consequences-values), but earlier levels of abstraction are apparently not considered necessary for later levels of abstraction to be reached. Hence, there is no way data gathered by means of a laddering interview could disconfirm means-end chain theory (Brunsø, Scholderer & Grunert, 2004; Scholderer, Brunsø & Grunert, 2002). We believe that the time is ripe for an investigation of the validity of the means-end chain model, and we will present a first attempt at this here. Our investigation has two parts: a reformulation of the theory in more precise terms, resulting in a number of testable propositions, and a series of empirical studies testing these propositions. Such an investigation is a prerequisite for further scientific progress in the area, and the need for such an investigation has been acknowledged by major researchers in the field (Grunert, Beckmann & Sørensen, 2001; Reynolds & Olson, 2001).

THEORETICAL RECONCEPTUALISATION

As noted above, a means-end chain can either be interpreted as a memory structure or a motivational process. It can be interpreted as a memory structure, because it makes assumptions on how consumers perceive and categorise products and services, i.e., how knowledge on products and services is stored in memory. It can be interpreted as a motivational process, because it makes assumptions about what drives consumers’ desire to choose one product over another. Both interpretations are present in the means-end chain literature, but they are not usually clearly distinguished. By separating them, and by relating them to the bodies of literature existing on memory structure on the one hand and on motivational processes on the other, we can develop a more stringent theoretical framework, leading to testable propositions.

12

Means-End Chains as Memory Structures

When interpreted as memory structure, means-end chains describe conceptual associations between product characteristics, consumption consequences and personal values. Applied to the example of organic food above, this would mean that consumers’ memory might contain a top-level category like “personal values” with the concept “responsibility” as an element. On a subordinate level of abstraction, there might be a category like “ways of acting responsibly” with the concept “doing something good for the environment” as an element. On a further subordinate level, there may be a category “ways of doing something good for the environment” with the concept “consuming organic foods” as an element. Drawing a computer analogy, such a system would be part of consumer’s “mental hardware”. It would be amenable to reconfiguration (by means of associative learning), but still only provide the structures on which the actual information processing is run, depending on the needs arising from a particular situation. The situation focused on by means-and chain researchers would typically be that of a consumer being confronted with a product. If the product has a particular characteristic distinguishing it from others (e.g., “organic”), consumers are thought to categorise the product automatically according to the means-end chain associated with that characteristic (e.g., “consuming organic foods” is a way of “doing something good for the environment” is a way of “acting responsibly”).

Means-end Chains as Motivational Processes

When interpreted as motivational processes, means-end chains describe the way in which a personal life value is being enacted through consumption consequences that arise from certain product characteristics. Once consumers have committed themselves to a certain value (e.g., “it is important for me to act responsibly”), this value may become a top-level goal consumers intend to achieve in their lives. Due to its abstractness, however, such a top-level goal can only be achieved by breaking it gradually down into more concrete, manageable goals that are instrumental in the achievement of the top-level goal (e.g. one way of “acting responsibly” is “doing something good for the environment”, one way of “doing something good for the environment” is “consuming organic foods”). The “mental hardware” accessed by such a motivational process may be the same as in the memory-structure interpretation above. However, the sequence in which product characteristics, consequences and values are accessed is exactly the opposite. We can therefore also call the memory structure view the bottom-up view of means-end chains, whereas we can call the motivational process view the top-down view.

13

We can link these two views of means-end chains, which can supplement each other, to two distinct bodies of literature. Concerning means-end chains as memory structures, we can draw upon the literature on conceptual hierarchies. Concerning means-end chains as motivational processes, we can draw upon the literature on goal pursuit.

Categorisation and construction of meaning: the bottom-up route

In research on conceptual representation, the model coming closest to the hierarchical structure posited by means-end chain theory is that of conceptual hierarchies. Conceptual hierarchies are systems in which categories are organized in terms of class inclusion relationships. The highest level of such hierarchies is the most inclusive one (e.g., “dairy foods”), and each lower-level category is nested within a higher level (e.g., “cheese”, “milk”, “yoghurt”). Conceptual hierarchies have been investigated for natural categories such as foods, plants and animals (Brooks, Norman & Allen, 1991; Johnson & Mervis, 1997, 1998; Lopez, Atran, Coley & Atran, 1997; Malt, 1994; Malt & Johnson, 1992; Rips, 1989; Rosch, Mervis, Gray, Johnson & Bouyes-Braem, 1976; Ross & Murphy, 1999; Tanaka & Taylor, 1991), yielding predominantly positive evidence for the hierarchical nature of their organization (for a review see Murphy & Lassaline, 1997). A number of studies have also looked at goal-derived and script categories (such as “foods you eat for breakfast” and “snack foods”) and found them to exhibit hierarchicity, although to a lesser degree than is usually found for natural categories (Barsalou, 1983, 1985, 1991; Medin, Lynch, Coley & Atran, 1997; Ross & Murphy, 1999). Our first theoretical reformulation is therefore as follows: means-end chains are a special case of conceptual hierarchies, where the levels of the hierarchy are defined as product attributes, consequences of product use, and life values. This is in line both with the original formulation of the means-end chain concept in marketing by Gutman (1982), and its early origin in the psychology of personal constructs (Kelly, 1955).

Motivation and goal pursuit: the top-down route

In motivation research, on the other hand, the model coming closest to the hierarchical structures posited by means-end chain theory is that of goal pursuit.

14

Gollwitzer (1990, 1993, 1999) has investigated how people attain abstract, high-level goal states by breaking them down into a hierarchy of more concrete, manageable goals. The key distinction of his theory is between goal intentions and implementation intentions. Goal intentions specify a certain end point that may be either a desired action or the outcome of an action. By forming goal intentions, people translate noncommittal motives into binding goals (Gollwitzer, Heckhausen & Steller, 1990; Gollwitzer & Wicklund, 1985). Implementation intentions are subordinate to goal intentions and specify the when, where, and how of behaviors leading to goal attainment. They have the structure of “When situation X arises, I will do Y” and thus link anticipated opportunities to goal-directed behaviors. The processes on which the effects of implementation intentions are based relate to both the specified situations and the intended behaviors. This mental act is assumed to lead to the automatisation of the intended goal-directed behavior once the critical situation is encountered. By forming implementation intentions, people can strategically switch from conscious and effortful control of their goal-directed behaviors to being automatically controlled by selected situational cues. For instance, people who have formed the goal intention to become slimmer can furnish it with implementation intentions that specify when, where, and how they want to buy which sorts of foods with a low fat content. The implementation of their goal intention is thus placed under the direct control of situational cues and removed from conscious and effortful control. An impressive research program by Gollwitzer and colleagues has shown that implementation intentions are the key determinants of intention-behavior consistency (Gollwitzer & Bargh, 1996; Gollwitzer & Bayer, 1999; Gollwitzer & Brandstaetter, 1997; Gollwitzer & Moskowitz, 1996; Gollwitzer & Oettingen, 1998; Gollwitzer & Schaal, 1998). Means-end chains can therefore also be interpreted as a hierarchical system of goal and implementation intentions. Below the life value level, consequences can be interpreted as subgoals to the superordinate goal of achieving the life value, and attributes linked to these consequences can be interpreted as intentions to implement the achievement of these goals by buying products possessing these attributes.

15

Testable propositions

Fed by the two streams of psychological theory described above, we would now like to discuss four propositions that could be cornerstones of a reformulated theory of means-end chains. These propositions concern hierarchicity, automaticity, birectionality, and self-relevance.

Proposition 1: Hierarchicity

Both theoretical reformulations discussed above imply hierarchicity. In both cases means-end chains are conceived as a system of cognitive categories with a layered structure (Cohen, 2000; Murphy & Lassaline, 1997; Ross & Murphy, 1999). More specifically we propose that the structure is three-layered, consisting of attribute (A), consequence (C) and value (V) nodes that are hierarchically linked. A is directly associated with C, C is directly associated with V, but A is only indirectly associated with V through the mediator C. The assumption of hierarchicity has consequences for the cognitive processes operating on the system of cognitive categories. Retrieval and use of information in networks of cognitive categories is commonly modelled by spreading activation theory (Collins & Loftus, 1975), and when the spreading activation model is applied to a three-layered hierarchical structure, it would follow that when A is activated then activation spreads from A to C and then to V, but never directly from A to V. As a consequence, when the first node of a directly associated attribute-consequence (attribute-consequence) or consequence-value (consequence-value) pair is activated, the time needed for the activation to spread to the second node should be shorter than the time needed to spread from the first to the last node in an only indirectly associated attribute-value (attribute-value) pair. Such predictions can be tested by priming experiments, as we will show later (McNamara ,1992).

Proposition 2: Automaticity

Subsequent nodes in an means-end chain can be activated by means of different cognitive processes. These can be strategic, consciously controlled ones like active reasoning about the functional consequences of a certain attribute, or they can be automatic, unconscious ones, like spreading activation. Both theoretical reformulations discussed above imply a certain degree of automaticity. Categorization is commonly assumed to be an automatic process, and a central element in the theory of goal pursuit is that the formation of implementation intentions creates a degree of automaticity in relating certain cues in the environment to the goal to be attained.

16

Under normal circumstances, strategic and automatic processes co-occur and are therefore operationally confounded. However, strategic processes can be suppressed by certain experimental manipulations (Bargh, 1992; Bargh & Chartrand, 2000; Shiffrin & Schneider, 1977). If means-end chains are indeed associative networks in human memory with the layered structure hypothesized above, spreading activation should still occur when strategic, consciously controlled processes are experimentally suppressed.

Proposition 3: Bidirectionality

If means-end chains are associative networks with a layered structure, and if we look at the two theoretical reformulations above in conjuncture, there is no reason to assume that activation will not spread from both ends in the same way as in other hierarchical memory structures (Altmann & Trafton, 2002; Bargh, Gollwitzer, Lee-Chai, Barndollar, & Troetschel, 2001; Conway 1990; Shah & Kruglanski, 2002, 2003). Based on such reasoning, Scholderer et al. (2002) have reformulated means-end chain theory as a dual-process model. Their model assumes that activation can spread through the same associative network on asymmetric routes. The bottom-up route, where activation spreads from attributes over consequences to values, is similar to the hierarchical categorization model proposed by Gutman (1982). The top-down-route, where activation spreads from values over consequences to attributes, is similar to the hierarchical goal structures and motivation models proposed by Cohen and Warlop (2001) and Pieters et al. (2001). If means-end chains are indeed bidirectional structures, automatic spreading activation should occur in the same way when means-end chains are primed in a bottom-up direction as when they are primed in a top-down direction. Thus, also this proposition can be investigated by priming experiments, as we will show below.

Proposition 4: Self-Relevance

The basic tenet of means-end chain theory is that consumers perceive utility in a product only to the extent that its attributes are associated with self-relevant consequences, which in turn derive their importance from associations with personal goals. Hierarchical functional relationships are a pervasive feature of the associative and semantic structure of natural language (Murphy and Lassaline 1997; Ross and Murphy 1999; Zacks and Tversky 2001), and a major difference between a person’s means end chains and many other hierarchical relationships

17

should therefore be that means end chains are hierarchical associative network structures with high self-relevance. If means-end chains elicited from a person are indeed associative network structures with high-self-relevance to that person, automatic spreading activation through the nodes in that person’s own means-end chains should be stronger than for means-end chains not elicited from that person, or for associative or semantic relationships between concept in natural language (Shah and Kruglanski 2002, 2003). This, as well, can be tested using a priming paradigm.

OVERVIEW OF EXPERIMENTAL WORK

In the following, we will present a series of six experiments that were conducted to test the four propositions just presented. All experiments follow a common format in that they consist of two sessions, with a considerable delay in between to avoid carryover effects. In the first session, a laddering interview was conducted with the aim to elicit respondents’ means end chains. The second session consisted of a priming experiment, where the results from the laddering interview were used as stimulus material.

Laddering interviews

Laddering is a semi-qualitative interview technique that has been commonly used in studies based on the means-end approach (Reynolds & Gutman, 1988). The method consists of two steps. The first step is to find relevant product attributes that are important to a given consumer when he or she chooses between varieties of the product class under investigation. This can be done in different ways, including direct questioning, triadic sorting, or ranking of products (Bech-Larsen & Nielsen, 1999). The second step is to have the consumer elaborate on the subjective meaning of the attribute, asking him or her questions like “Why is (the attribute elicited) important to you?” and, when he or she answers, continuing with a question like “Why is that important to you?” This is continued until the respondent does not produce any new elaborations. The resulting tree (“ladder”) is regarded as an exhaustive representation of the consumer’s means-end-chain for the product class under investigation. Laddering data by itself cannot be used to test assumptions of means-end theory, because the laddering methods forces the structure assumed by means-end theory

18

upon the data. We will therefore only use the laddering technique to derive individualised stimulus sets for the subsequent experimental session.

Priming Experiments

Priming experiments are widely used in cognitive psychology (Bargh & Chartrand, 2000), but their application to questions of consumer behaviour has been sparse. The materials in a priming experiment typically consist of a number of word pairs (“primes” and “targets”, half of them semantically related, half of them semantically unrelated) plus a number of meaningless character strings (“non-words”) that are consecutively presented on a computer screen (see Figure 2). In a first step, the prime stimulus is shown. After a short inter-stimulus-interval, the target stimulus appears on the screen, and the participant is asked to decide as soon as possible whether the stimulus is a word or a non-word (“lexical decision task”). The time it takes the respondent to make this decision is called the decision latency. Decision latencies are affected by the semantic relationship between the prime and the target – if the prime and the target are related (apple – fruit) the respondent is quicker to decide that the target is actually a word than when they are unrelated (apple – chair) (e.g., Higgins, Bargh & Lombardi, 1985). This effect is called a response facilitation effect.

Figure 2. Typical priming experiment The type of semantic relationship investigated in priming experiments is typically a class inclusion relationships among concepts (apple – fruit). When the prime is a direct subcategory of the target, this is called a direct priming experiment. In a mediated priming experiment, a low-level category is used to prime a high-level

organic

0 ms 300 ms 600 ms 900 ms

responsibility

Presentation ofprime stimuluson the screen

Presentation of target stimuluson the screen

Stimulus onset asynchrony (SOA)

chair

Presentation of unrelated wordon the screen

19

category without presentation of the intermediate-level category (Tuborg – drink, without mentioning the intermediate category ‘beer’). Intermediate priming effects provide strong evidence for chain-like structures in memory (Balota & Lorch, 1986), which are the main interest in means-end chain theory. We will employ priming experiments to test the four propositions presented. Both direct and mediated priming will be used. The individualised stimulus sets elicited in the laddering interviews will be used as related stimulus pairs. Two control conditions will be introduced against which response facilitation effects due to related stimulus pairs can be tested: (a) semantically unrelated stimulus pairs constructed from normed word lists, (b) stimulus pairs that were elicited in a pilot session with another participant, but were not included in the stimulus material elicited from the focal participant.

EXPERIMENT 1

The first experiment was designed to test the hierarchicity and self-relevance postulates. Hierarchicity was tested by examining whether response facilitation effects were higher when primes and targets were directly associated nodes in a means-end chain (attributes and consequences, or consequences and values) than when primes and targets were indirectly associated through a mediator (attributes and values, mediated by consequences). Self-relevance was tested by examining whether response facilitation effects were stronger when primes and targets were taken from a person’s own means-end chain (as measured by the laddering method) than when taken from another person’s means-end chain or from a standardised word list. In this experiment, only bottom-up priming of means-end chains was investigated.

Method

Participants

Altogether 90 students from the Aarhus School of Business, Aarhus University, Aarhus Technical College, and Aarhus Business College participated for monetary compensation (DKK 75 per participant). All participants were native Danish speakers and had normal or corrected-to-normal vision. The mean age was 23.02 years (SD = 2.73), 60.9% were female.

20

Procedure

Two sessions were held with each participant. In a pilot session, each participant completed four different laddering tasks. The product categories investigated were beef, milk, canned tomatoes, and margarine. Each task consisted of four different consumer products varying on three different attributes. Attributes were counterbalanced across the different product categories to avoid confounding. In each task, participants were asked to (a) rank the four consumer products according to personal preference, (b) identify the salient attributes governing their choice, and (c) generate consequences and values for each attribute. This way, 12 means-end chains were elicited from each participant, each consisting of an attribute-consequence-value triple. In addition, each participant completed a lengthy questionnaire on organic foods, which provided the cover story. After the pilot session, the word material that participants had generated in the laddering task was entered into a database. Individualized stimulus sets were then generated from the database for use in the second session. To avoid carry-over effects, the second session was arranged after a relatively long delay (on average 55.92 days, SD = 35.78). Each participant completed two blocks of a sequential priming experiment, a practice block and a main block. In both blocks McNamara and Altarriba’s (1988) single-presentation lexical decision task was used. Prime and target stimuli were presented sequentially on the computer screen. Separate lexical decisions were made for each of the apparently unpaired stimuli. Participants were asked to decide quickly but accurately if the stimulus presented on the screen was a word or a non-word and respond by pressing either key “1” or key “2” on the numeric block of the keyboard. The stimulus remained on the screen until either of the keys was pressed. The practice block consisted of 40, the main block of 160 such trials. A long inter-stimulus-interval (ISI) of 750 msec was used throughout the experiment. Reaction times (RTs) were measured from the onset of each target stimulus and automatically logged. In line with standard procedures (Bargh & Chartrand, 2000), implausible RTs below 250 msec or above 2000 msec were excluded from the analysis. Results from two participants had to be completely discarded because of corrupt log files.

Word material

The word material in the main block included three broad classes of prime-target pairs: (a) 12 word pairs constructed from the participant’s own means-end chains, which had been elicited in the pilot interview, (b) 12 word pairs constructed from

21

another participant’s means-end chains, checked for non-overlap with the own means-end chain stimuli, and (c) 32 word pairs constructed from a standardised word list, serving as a benchmark. Each of these classes contained a number of sub-classes differing in terms of the associative distance between the prime-target pairs. Of the 12 own means-end chain pairs, four were directly associated attribute-consequence pairs, four were directly associated consequence-value pairs, and four were indirectly associated attribute-value pairs. Likewise, of the 12 other-participant means-end chain pairs, four were directly associated attribute-consequence pairs, four were directly associated consequence-value pairs, and four were indirectly associated attribute-value pairs. Of the 32 word pairs constructed from the standardised word list, eight pairs were directly associated, eight were indirectly associated, and 16 were unassociated. In addition, the word material for each participant contained 16 fillers and 32 non-words, serving as distractors. The distribution of word length and word type in the standardised list was matched to the distribution of word length and type in the pooled means-end chain material. During the main trial, word pairs were presented in randomized order. Means and standard deviations of RTs under each experimental condition are shown in Table 1, along with the total number of responses (i.e., cumulated over participants) on which the estimates are based.

Results

Data were analyzed by means of repeated-measures ANOVA, using the generalised least squares estimator for random-effects models in STATA 8 (procedure XTREG). Experimental condition was defined as a 9-level within-subjects factor with random effects. The levels were (1) own means-end chain, direct association attribute-consequence, (2) own means-end chain, direct association consequence-value, (3) own means-end chain, indirect association

22

Table 1. Means and standard deviations of reaction times under each condition in Experiment 1.

Source Prime-target pair Mean RT (msec)

Standard deviation

Total N of responses

Attribute – consequence 654.25 257.92 331

Consequence – value 627.51 242.09 317

Participant’s own MECs

Attribute – value 645.90 270.99 317



Other participant’s MECs


Directly associated 737.37 291.05 683

Indirectly associated 779.16 315.53 682

Standardised word list

Unassociated 721.09 293.26 2055

Note. Total number of participants = 88. Implausible RTs below 250 msec or above 2000 msec excluded from analysis.

attribute-value, (4) other means-end chain, direct association attribute-consequence, (5) other means-end chain, direct association consequence-value, (6) other means-end chain, indirect association attribute-value, (7) standardised word list, direct association, (8) standardised word list, indirect association, and (9) standardised word list, no association between the two words. Level (9) was defined as the reference category. Hence, estimates of the first eight parameters can directly be interpreted as priming effects (in msec metric), i.e. the response facilitation effect that was observed when a target word was preceded by an associated word, as compared to the benchmark condition of a target word that was preceded by an unassociated word. Compared to this benchmark assessment (average RT = 724.68 msec, S.E. = 12.96), all own means-end chain conditions and all other means-end chain conditions yielded significant priming effects, whereas the two standardised word-list conditions did not. Parameter estimates are presented in Table2.

23

Table 2. Parameter estimates: priming effects obtained in Experiment 1 (bottom-up priming, controlled information processing conditions).

Source Prime-target pair b SE(b) z p

Attribute – consequence -67.580 15.584 -4.340 .000

Consequence – value -93.792 15.886 -5.900 .000


Attribute – value -75.077 15.888 -4.730 .000




Attribute – value -52.862 15.342 -3.450 .001

Directly associated 17.525 11.620 1.510 1.000

Indirectly associated 58.425 11.624 5.030 1.000


Unassociated 0 (ref) n.a. n.a. n.a.

(Constant) 724.678 12.961 55.910 .000

Note. Random-effects model; parameters estimated by means of generalised least squares. All p-values one-tailed.

To test the hierarchicity postulate, the two own means-end chain conditions with direct associations (attribute-consequence and consequence-value) were simultaneously contrasted against the own means-end chain condition with indirect association (attribute-value). However, the contrast did not yield a significant result (chi-square [2] = 1.69, p > .42).Contrary to expectation, activation did not appear to spread faster between directly associated elements in a means-end than between those that were separated by a mediator. To test the self-relevance postulate, the three own means-end chain conditions were contrasted against the two standardised word-list conditions with associations, simultaneously testing directly and indirectly associated ones against their respective counterparts. The contrast yielded a highly significant result (chi-square [2] = 101.80, p < .001). As expected, priming effects of attribute-consequence and consequence-value pairs from participants’ own means-end chains were significantly higher than priming effects of directly associated words from the standardised word list. Likewise, priming effects of attribute-value pairs from participants’ own means-end chains were significantly higher than priming effects of indirectly associated words from the standardised word list.

24

A second, conceptually stronger test was conducted by contrasting the three own means-end chain conditions with their respective other means-end chain counterparts. A simultaneous test of the three RT differences failed to reach significance (chi-square [3] = 1.99, p > .57). Contrary to expectation, word pairs from participants’ own means-end chains did not show higher priming effects than word pairs from other participants’ means-end chains.

Discussion

The experiment did not confirm our expectation on the hierarchicity effect, but the results give rise to some speculation about the nature of the spreading activation process. It is clear that the consequence-value pair resulted in the strongest response facilitation, and there was also a (weaker) tendency that the attribute-value pair resulted in stronger response facilitation than the attribute-consequence pair. A possible explanation of this could follow from the nature of values: values are defined as personal end goals of high relevance (Schwartz, 1992), and it is therefore plausible to assume that they have a higher start activation than attributes or consequences. This not only could explain the higher response facilitation for pairs involving values, it could also mean that the hierarchicity assumption may still be valid, but is confounded here with different levels of start activation: the prediction would be that the C-V pair has the highest response facilitation (which it has), and that the sequence of the two other pairs would depend on the relative strength of the difference in start activation on the one side and the slowing down in spreading activation caused by the indirect link on the other side. We could confirm that the means-end based pairs have higher self-relevance than pairs of standardised words, but that there is no difference between ‘own’ and ‘other’s’ chains. A possible interpretation of this finding relates to the question of what is actually measured in a laddering interview. While such interviews are often taken as a direct measure of respondents’ means-end structure, an alternative view which has been advanced (Grunert & Grunert, 1995) is that we should rather interpret a laddering interview as a stochastic process, where means-end chains are being retrieved from a larger and more complex structure that may possibly involve other, related means-end chains. It follows that while a chain elicited from one respondent can be assumed to be part of that respondent’s cognitive structure, the opposite, namely that a chain elicited from somebody else is not part of the cognitive structure, is not necessarily true. Since all respondents were interviewed on the same products and the samples were reasonably homogeneous, most of the chains may have been relevant for most of the respondents.

25

EXPERIMENT 2

The second experiment was designed to test the bidirectionality postulate. Bidirectionality was tested by examining whether the results obtained in Experiment 1, where bottom-up priming of means-end chains had been used (attribute, consequence, value), would remain stable when the direction of priming was turned around to top-down priming (value, consequence, attribute).

Method

Participants

91 students were recruited in the same way as above. All were students at the Aarhus School of Business, Aarhus University, Aarhus Technical College or Aarhus Business College and participated for monetary compensation. Like before, all participants were native Danish speakers and had normal or corrected-to-normal vision. The mean age was 22.69 years (SD = 2.63), 56.8% were female.

Procedure

The procedure was the same as in Experiment 1. The average time lag between the first and the second session was 90.70 days (SD = 99.04). Again, implausible RTs below 250 msec or above 2000 msec were excluded from the analysis.

Word material

The word material was constructed in the same way as in Experiment 1, apart from one important modification: in this experiment, the direction of priming was turned around. All means-end chain stimulus pairs were primed in a top-down direction. The 12 own means-end chain pairs were now four consequence-attribute pairs, four value-consequence pairs, and four value-attribute pairs. Likewise, the 12 other means-end chain pairs were now four consequence-attribute, four value-consequence, and four value-attribute pairs. Means and standard deviations of RTs under each experimental condition are shown in Table 3, along with the total number of responses (i.e., cumulated over participants) on which the estimates are based.

26



Standard deviation


Consequence – attribute 705.44 326.43 327

Value – consequence 664.32 260.47 324


Value – attribute 728.29 307.58 317








Unassociated 787.93 343.68 1988


Results

The same model as in Experiment 1 was estimated based on the data from Experiment 2. Compared to the benchmark assessment provided by unassociated word pairs (average RT = 793.18 msec, S.E. = 17.12), all own means-end chain conditions and all other means-end chain conditions, but only one of the two standardised word-list conditions yielded significant priming effects. Parameter estimates are presented in Table 4. To replicate the hierarchicity test under top-down priming conditions, the two own means-end chain conditions with direct associations (consequence-attribute and value-consequence) were simultaneously contrasted against the own means-end chain condition with indirect association (value-attribute). This time, the contrast yielded a significant result (chi-square [2] = 7.51, p < .05). However, a very strong priming effect observed for value-consequence pairs was responsible for this, whereas consequence-attribute and value-attribute pairs did not differ from another.

27

Table 4. Parameter estimates: priming effects obtained in Experiment 2 (top-down priming, controlled information processing conditions).


Consequence – attribute -85.418 17.423 -4.900 .000

Value – consequence -126.869 17.505 -7.250 .000


Value – attribute -64.958 17.677 -3.670 .000




Value – attribute -39.994 17.244 -2.320 .010

Directly associated -23.305 13.714 -1.700 .045

Indirectly associated -16.381 13.740 -1.190 .116



(Constant) 793.180 17.123 46.320 .000


To replicate the self-relevance test under top-down priming conditions, the three own means-end chain conditions were contrasted against the two standardised word-list conditions with associations. As in Experiment 1, the contrast yielded a highly significant result (chi-square [2] = 30.38, p < .001). Again, priming effects within participants’ own means-end chains were significantly higher than priming effects of directly associated and indirectly associated words from the standardised word list. However, the second, conceptually stronger self-relevance test failed again (chi-square [3] = 195, p > .57). Word pairs from participants’ own means-end chains did not show higher priming effects than word pairs from other participants’ means-end chains, even when the direction of priming was turned around.

Discussion

The results are largely parallel to Experiment 1, and therefore the same interpretations can be invoked. The V-C pair resulted in the highest response facilitation, in line with the C-V pair in Experiment 1. The results are thus

28

compatible with the same post-hoc interpretation we gave for experiment 1, namely that the hierarchicity effect may be confounded with different start activations for values and other categories. Also for the results for self-relevance the same interpretation as for Experiment 1 can be invoked.

29

EXPERIMENT 3

The third experiment was designed to ensure that Experiments 1 and 2 were internally valid. In both experiments, the word material representing means-end chains had been elicited by means of a laddering task where different foods had served as product examples. Hence, the generated material shared a common associative context. To test the alternative explanation that a generalized activation of this common associative context might have been responsible for the priming effects observed in Experiments 1 and 2, and not specific activation of particular means-end chains, Experiment 3 Experiment 2, but used only stimulus materials from a food context, i.e. also in the standardised word-list conditions against which all priming effects were benchmarked.

Method

Participants

30 students were recruited in the same way as above. Like before, all participants were native Danish speakers and had normal or corrected-to-normal vision. The mean age was 23.20 years (SD = 2.25), 46.7% were female. The average time lag between the first and the second session was 122.13 days (SD = 110.60).

Procedure

The procedure was the same as in Experiment 2. Again, implausible RTs below 250 msec or above 2000 msec were excluded from the analysis.

Word material

The word material was the same as in Experiment 2, apart from a modification of the word-list context. Whilst the standardised word list in Experiments 1 and 2 included words from random contexts, the standardised word list in Experiment 3 included only words from a food context. Again, the distribution of word length and word type in this list was matched to the distribution of word length and word type in the pooled means-end chain material. Means and standard deviations of RTs under each experimental condition are shown in Table 5, along with the total number of responses (i.e., cumulated over participants) on which the estimates are based.

30



Standard deviation













Unassociated 772.12 343.15 662


Results

To test whether the results obtained in Experiments 1 and 2 would remain stable when all stimuli shared a common associative context, the same model as before was re-estimated using the data from Experiment 3. Although all effects were still in the same direction, most of them dropped considerably in size. Compared to the benchmark assessment provided by unassociated words from the new standardised word list (average RT = 782.59 msec, S.E. = 26.22), only two of the other means-end chain conditions (consequence-attribute and value-consequence) and one of the standardised word-list conditions (indirectly associated words) yielded significant priming effects. Parameter estimates are shown in Table 6. Under the common associative-context conditions imposed by this experiment, the hierarchicity test failed. A contrast simultaneously comparing the two own means-end chain conditions with direct associations to the own means-end chain condition with indirect association did not yield a significant result (chi-square [2] = .68, p > .70).

31

Table 6. Parameter estimates: priming effects obtained in Experiment 3 (top-down priming, controlled information processing conditions, common associative context).





Value – attribute -17.038 32.325 -0.530 .249




Value – attribute -19.060 32.177 -0.590 .277





(Constant) 782.588 26.222 29.850 .000


The weak self-relevance test, where the three own means-end chain conditions were simultaneously contrasted against the two standardised word-list conditions with direct and indirect associations, did not reach significance either (chi-square [2] = .57, p > .75). The strong self-relevance test, contrasting the three own means-end chain conditions with their respective other means-end chain counterparts, failed as well (chi-square [3] = 3.52, p > .31).

Discussion

Given that the words in the standardized list now all had some semantic relationship, the mere fact that the effect sizes fall is to be expected, and the significances have also to be seen in the light of the smaller sample size compared to the two previous experiments. Still, we have to note that there is not even a tendency that the means-end based stimulus material resulted in generally higher response facilitation than the standardized list words. The V-C pair still has the highest response facilitation effect, which replicates a main finding of the previous two experiments.

32

EXPERIMENT 4

The fourth experiment was designed to test the automaticity postulate. Automaticity was tested by examining whether hierarchicity and self-relevance could be observed under conditions designed to suppress controlled information processing. In line with standard procedures (Bargh and Chartrand 2000), short inter-stimulus intervals and a high proportion of fillers and non-words in the word material were used for this.

Method

Participants


Procedure

The overall procedure was as in Experiment 1, with two important modifications. This time the main block consisted of 256 trials, and throughout the block a short inter-stimulus-interval (ISI) of 250 msec was chosen to induce automatic information-processing conditions. Like before, implausible RTs below 250 msec or above 2000 msec were excluded from the analysis. Data from three participants had to be discarded completely because of corrupt log files.

Word material

The word material was the same as in Experiment 1, apart from a heavily increased number of unassociated word pairs, fillers and non-words. In addition to the 12 own means-end chain pairs (four attribute-consequence, consequence-value, attribute-value, respectively), the 12 other means-end chain pairs (four attribute-consequence, consequence-value, attribute-value, respectively), and the 16 directly associated and indirectly associated pairs from the standardised word list, the word material for each participant contained 24 unassociated word pairs, 48 fillers and 96 non-words. The words in the standardised word list were taken from random contexts and were therefore comparable to those used in Experiments 1 and 2, but not Experiment 3. Like before, the distribution of word length and word type in this list was matched to the distribution of word length and word type in the pooled

33



Standard deviation













Unassociated 715.58 267.73 4591


means-end chain material. Means and standard deviations of RTs under each experimental condition are shown in Table 7, along with the total number of responses (cumulated over participants) on which the estimates are based.

Results

The same model as in Experiments 1, 2 and 3 was estimated again, now using data obtained under conditions of automaticity and bottom-up priming. Compared to the benchmark assessment provided by unassociated word pairs (average RT = 717.09 msec, S.E. = 8.23), all own means-end chain conditions, all other means-end chain conditions, and the two standardised word-list conditions with direct and indirect association yielded highly significant priming effects. Parameter estimates are presented in Table 8. To test the hierarchicity postulate under automaticity conditions, the two own means-end chain conditions with direct associations (attribute-consequence and consequence-value) were simultaneously contrasted against the own means-end chain condition with indirect association (attribute-value). As in Experiments 1 and 3, the contrast yielded an insignificant result (chi-square [2] = 1.70, p > .42).

34

Table 8. Parameter estimates: priming effects obtained in Experiment 4 (bottom-up priming, automatic information processing conditions).





Attribute – value -114.896 11.184 -10.270 .000




Attribute – value -84.124 11.105 -7.580 .000





(Constant) 717.087 8.229 87.140 .000


To replicate the weak self-relevance test under automaticity and bottom-up conditions, the three own means-end chain conditions were contrasted against the two standardised word-list conditions with associations. As in Experiments 1 and 2, the contrast yielded a significant, but somewhat weaker result (chi-square [2] = 5.19, p < .05). This time, the strong self-relevance test yielded a significant result as well (chi-square [3] = 9.08, p < .05). Word pairs from participants’ own means-end chains showed higher priming effects than word pairs from other participants’ means-end chains or a standardised word list when controlled information processing was suppressed.

Discussion

The introduction of the automaticity condition did not change the basic pattern of results. Hierarchicity as hypothesized does not materialize, but we replicate once more the finding that the C-P pairs result in the highest response facilitation, which can be interpreted in the same way as we did when discussing Experiment 1. The

35

major difference to the parallel Experiment 1 is that the strong self-relevance test now is significant. In discussing Experiment 1, we argued that a possible explanation of the failure of the strong relevance test is that means-end chains in a laddering interview are retrieved from a more comprehensive cognitive structure, which may have included some of those chains elicited from other respondents. On the other hand, we may also argue that those chains actually elicited may have had a higher accessibility for the respondent. Under conditions of automaticity, one can expect that accessibility plays a larger role for response facilitation than under conditions where strategic processing can play a larger role in redirecting spreading activation patterns. This could explain that the strong self-relevance test becomes significant only under the automaticity condition.

36

EXPERIMENT 5

The fifth experiment was designed as a second test of the automaticity postulate. The postulate was tested by examining whether the bidirectionality effect observed in Experiment 2 could be replicated under conditions designed to suppress controlled information processing. Again, short inter-stimulus intervals and a high proportion of fillers and non-words were used to induce automatic information processing.

Method

Participants

65 students were recruited in the same way as above. Again, all participants were native Danish speakers and had normal or corrected-to-normal vision. The mean age was 23.53 years (SD = 2.75), 54.5% were female. The average time lag between the first and the second session was 85.44 days (SD = 114.27).

Procedure

The procedure was the same as in Experiment 4. Like before, implausible RTs below 250 msec or above 2000 msec were excluded from the analysis. Results from one participant had to be discarded because of corrupt log files.

Word material

The word material was the same as in Experiment 4, apart from one modification: the direction of priming was turned around, in the same way as Experiment 2 used the opposite direction of priming as Experiment 1. The 12 own means-end chain pairs were now consequence-attribute, value-consequence, and value-attribute pairs. Likewise, the 12 other means-end chain pairs were now consequence-attribute, value-consequence, and value-attribute pairs. Means and standard deviations of RTs under each experimental condition are shown in Table 9, along with the total number of responses (cumulated over participants) on which the estimates are based.

37



Standard deviation













Unassociated 742.12 288.81 2474


Results

The same model as in Experiments 1 through 4 was estimated once more, now using data obtained under conditions of automaticity and top-down priming. Compared to the benchmark assessment provided by unassociated word pairs (average RT = 747.00 msec, S.E. = 14.91), all own means-end chain conditions, all other means-end chain conditions, and the two standardised word-list conditions with direct and indirect association yielded highly significant priming effects. Parameter estimates are presented in Table 10. To test the hierarchicity postulate under automaticity and top-down conditions, the two own means-end chain conditions with direct associations (consequence-attribute and value-consequence) were simultaneously contrasted against the own means-end chain condition with indirect association (value-attribute). As in Experiment 2, the contrast yielded a significant result (chi-square [2] = 11.63, p < .01). However, the effect was in the opposite direction: indirectly associated value-attribute pairs yielded higher priming effects than directly associated consequence-attribute or value-consequence pairs.

38

Table 10. Parameter estimates: priming effects obtained in Experiment 5 (top-down priming, automatic information processing conditions).





Value – attribute -134.646 16.901 -7.970 .000




Value – attribute -78.120 16.739 -4.670 .000





(Constant) 746.996 14.905 50.120 .000


To replicate the weak self-relevance test under automaticity and top-down conditions, the three own means-end chain conditions were again contrasted against the standardised word-list conditions. Again, the contrast yielded a significant but relatively weak result (chi-square [2] = 5.04, p < .05). The strong self-relevance test, contrasting the three own means-end chain conditions with their respective other means-end chain counterparts, yielded a significant but somewhat weak result as well (chi-square [3] = 7.42, p < .05).

Discussion

For other respondents’ MECs, we replicate once more the finding that the pair consisting of values and consequences results in the highest response facilitation. However, for respondents’ own MECs, we find the highest response facilitation for the V-A pairs. We can not presently come up with a plausible interpretation of this finding.

39

We do replicate the finding from the previous experiment that, under conditions of automaticity, the strong self-relevance test became significant. The interpretation can therefore be the same as in the previous experiment.

40

EXPERIMENT 6

The sixth experiment was designed, in analogy to Experiment 3, to ensure that Experiments 4 and 5 were internally valid. To test the alternative explanation that generalized activation of a common associative context might have been responsible for the priming effects, Experiment 6 replicated the automaticity conditions of the previous experiment, but used only stimulus material from a food context, i.e. also in the standardised word-list conditions against which all priming effects were benchmarked.

Method

Participants


Procedure

The procedure was the same as in Experiment 5. Again, implausible RTs below 250 msec or above 2000 msec were excluded from the analysis. Data from two subjects had to be discarded completely because of corrupted log files.

Word material

The word material was the same as in Experiment 5, apart from a modification of the word-list context similar to the one used in Experiment 3. Whilst the standardised word list in Experiments 4 and 5 included words from random contexts, the standardised word list in Experiment 6 included only words from a food context. Again, the distribution of word length and word type in this list was matched to the distribution of word length and type in the pooled means-end chain material. Means and standard deviations of RTs under each experimental condition are shown in Table 11, along with the total number of responses (cumulated over participants) on which the estimates are based.

41



Standard deviation













Unassociated 676.56 279.96 1231


Results

To test whether the results obtained under automaticity conditions in Experiments 4 and 5 would remain stable when all stimuli shared a common associative context, the same model as before was re-estimated using the data from Experiment 6. As already observed under in Experiment 3, all effects dropped considerably in size. Compared to the benchmark assessment provided by unassociated words from the new standardised word list (average RT = 678.50 msec, S.E. = 19.42), only two of the own means-end chain conditions (consequence-attribute and value-attribute) and two of the other means-end chain conditions (consequence-attribute and value-consequence) yielded significant priming effects. Parameter estimates are presented in Table 12. As in Experiment 3, the hierarchicity test failed under the common associative-context conditions imposed by this experiment. The contrast simultaneously comparing the two own means-end chain conditions with direct associations to the own means-end chain condition with indirect association did not yield a significant result (chi-square [2] = 2.75, p > .25).

42

Table 12. Parameter estimates: priming effects obtained in Experiment 6 (top-down priming, automatic information processing conditions, common associative context).





Value – attribute -85.929 25.972 -3.310 .001




Value – attribute -11.254 24.623 -0.460 .324

Directly associated 28.520 14.352 1.990 1.000

Indirectly associated 20.202 14.190 1.420 1.000



(Constant) 678.495 19.421 34.940 .000


The weak self-relevance test, on the other hand, where the three own means-end chain conditions were simultaneously contrasted against the two standardised word-list conditions with direct and indirect associations, yielded a highly significant result (chi-square [2] = 23.80, p < .001). The strong self-relevance test, contrasting the three own means-end chain conditions with their respective other means-end chain counterparts, reached significance as well (chi-square [3] = 7.72, p < .05). Under conditions of automaticity and a common associative context, word pairs from participants’ own means-end chains showed higher priming effects than word pairs from other participants’ means-end chains or the standardised word list.

Discussion

The results of experiment 6 are parallel to those of Experiment 5. We find again that, for respondents’ own MECs, the V-A pair results in the highest response

43

facilitation, which is difficult to explain. For other respondents’ MECs, we once again replicate the finding that the pair consisting of values and consequences resulted in the highest response facilitation. We also replicate the significance of the strong self-relevance criterion which we found in all experiments under the automaticity condition.

44

GENERAL DISCUSSION AND CONCLUSIONS

The aim of the research reported here was to reformulate means-end chain theory in a coherent theoretical framework, derive empirically falsifiable predictions from the framework, and test these predictions by established experimental methods. Theoretically, means-end chains were cast as associative networks with a three-layered structure: attributes, consequences, and values. Four postulates were formulated that imposed testable restrictions on the layered network structure: hierarchicity, automatic spreading activation, bidirectionality, and self-relevance. The predictions were tested in altogether six experiments, using lexical decision tasks designed to measure spreading-activation processes in associative networks. Overall, only few of the predictions were met. Hierarchicity, the assumption that means-end chains have a three-layered chain structure (as opposed to a non-hierarchic, single-layered network structure), could only be established in one out of six experiments. Automaticity, the assumption that spreading activation through an means-end chain would still occur when controlled information processing was suppressed, could indeed be established in three out of three experiments that induced automatic information processing. Results for bidirectionality, the assumption that bottom-up spreading activation processes would be mirrored by top-down spreading activation processes, were rather favourable. However, under controlled information processing conditions, the pattern of results was relatively similar for bottom-up and top-down conditions, whereas under automatic information processing, the patterns were less similar. Self-relevance, the assumption that spreading-activation effects would be stronger for means-end chains generated by participants themselves than for means-end chains generated by other participants (strong self-relevance) or materials taken from a standardised word list (weak self-relevance), could partially be established. Evidence for strong self-relevance effects was found in two out of six experiments, whereas evidence for weak self-relevance effects was found in four out of six experiments. In discussing what these results mean for the means-end chain model, we first have to note that we have not tested the means-end chain model, because such a model does not exist, as we noted in the beginning. Rather, we have tested propositions which were derived from our reinterpretation of the means-end chain model.

45

While we believe we have made a fair attempt to reconceptualize the means-end approach, staying close to the original ideas on one hand while linking the approach to well-established bodies of cognitive psychology, our attempts to make sense of the results of our experiments also showed that our attempt was only a first step and will need further elaboration. Our experiments were useful in pointing out where such further elaboration is necessary. We can pinpoint three areas which need more theoretical attention – and subsequent experimental testing. The first one relates to how means-end chains form part of a bigger cognitive structure. Traditional means-end approaches, and also our reconceptualization, mostly look at sets of chains which characterize respondents, but not at their interconnection or how they relate to other knowledge. If we interpret means-end chains as cognitive structures, then we should expect them to be interrelated, and we should expect them to be associatively related to a host of other knowledge in the mind of consumers. This has consequences for our four propositions, and especially for the hierarchicity proposition. The more we regard means-end chains as part of a larger cognitive structure, the more redundancy we will probably also have in the overall network, and the more difficult it becomes to make predictions about strict hierarchicity. We may retain the assumption that consumers have cognitive categories at different levels of abstraction, interlinked in such a way that paths are created from product attributes to values, but the network structure need not be strictly hierarchical. This issue is interlinked with the next one, namely the processes resulting in the elicitation of chains in a laddering interview. When we abandon the somewhat naïve idea that consumers have a set of chains in their head which they reveal to the interviewer in a laddering interview, we need to acknowledge that means-end chains are constructed on the basis of a larger, more complex cognitive structure in a laddering interview. We know relatively little, both in terms of relevant theory and empirical insight, about how that is being done. We can assume that some paths from attributes to values are more accessible than others, and these may be the most likely ones to be elicited in a laddering interview. But a laddering interview is a situation with high situational involvement and ample strategic processing (Grunert & Grunert, 1995), so respondents can redirect their thinking in many different ways. We found results in our experiment which are compatible with this view. The fact that the contrast between own and other’s means-end chains was significant only under conditions of automaticity is compatible with the interpretation that respondents could find most of the overall pool of chains in their own mind when they were given the opportunity to strategically redirect their thinking.

46

The third issue deals with the process of spreading activation, or, more generally, with the processes involved in retrieving and using means-end information from memory. The classical means-end model is a structure model and says very little about how that structure is retrieved and used in forming attitudes or directing behaviour (Grunert et al., 1995). We have in our reconceptualization invoked spreading activation theory for that purpose, because it is the most straightforward way to model information retrieval and especially priming effects in an associative network. But the implications of spreading activation theory in a cognitive structure compatible with the basic assumptions of the means-end approach still need to be spelled out. In interpreting the results of our experiments, we have already advanced the hypotheses that values (almost by definition) may have a higher start activation than consequences and attributes, with implications for pattern of spreading activation in a means-end related network. Likewise, it follows from the basic hierarchical model of the means-end approach that values, consequences and attributes differ in their indegrees and outdegrees and hence in their centrality in the overall cognitive network. Taken together with the fact that values, again by definition, are relatively few in number, whereas attributes may be many, implications for the spreading activation process begin to emerge: activation of an attribute may be more likely to reach a value than activation of the same value may be to reach the same attribute, thus modifying the bidirectionality proposition. Such issues can only be resolved by more elaborate theorizing, preferably in the form of a mathematical model that would allow to make more concrete predictions about patterns of spreading activation. We did, in our experiments, find plenty of evidence for response elicitation effects, indicating that the elements of means-end chains are interconnected in the minds of consumers. Thus, means-end chains, as conventionally measured by the laddering method, are firmly anchored in people’s memory, but not as firmly as originally hypothesized. On the basis of better theorizing and accompanying empirical testing, we should be able to build a revised means-end model that sheds more light on how consumers link product attributes to consequences and values.

47

REFERENCES

Altmann, E. M. & Trafton, J. G. (2002). Memory for goals: An activation-based model. Cognitive Science, 26, 39-83.

Aurifeille, J.-M. & Valette-Florence, P. (1995). Determination of the dominant means-end chain: A constrained clustering approach. International Journal of Research in Marketing, 12, 267-278.

Balota, D. A. & Lorch, R. F. (1986). Depth of automatic spreading activation: Mediated priming effects in pronunciation but not in lexical decision. Journal of Experimental Psychology: Learning, Memory, and Cognition, 12, 336-345.

Bargh, J. A. (1992). The ecology of automaticity: Towards establishing the conditions needed to produce automatic processing effects. American Journal of Psychology, 105, 181-199.

Bargh, J. A. & Chartrand, T. (2000). The mind in the middle: A practical guide to priming and automaticity research. In H. T. Reis & C. M. Judd (Eds.), Handbook of research methods in social and personality psychology (pp. 253-285). Cambridge: Cambridge University Press.

Bargh, J. A., Gollwitzer, P. M., Lee-Chai, A., Barndollar, K. & Troetschel, R. (2001). The automated will: Nonconscious activation and pursuit of behavioral goals. Journal of Personality & Social Psychology, 81, 1014-27.

Barsalou, L. W. (1983). Ad hoc categories. Memory & Cognition, 11, 211-227.

Barsalou, L. W. (1985). Ideals, central tendency, and frequency of instantiation as determinants of graded structure in categories. Journal of Experimental Psychology: Learning, Memory, and Cognition, 11, 629-649.

Barsalou, L. W. (1991). Deriving categories to achieve goals. In G. H. Bower (Ed.), The psychology of learning and motivation: Advances in research and theory (Vol. 27, pp. 1-64). New York: Academic Press.

Bech-Larsen, T. & Nielsen, N. A. (1999). A comparison of five elicitation techniques for elicitation of attributes of low involvement products. Journal of Economic Psychology, 20, 315-341.

Botschen, G. & Thelen, E. (1998). Hard versus soft laddering: Implications for appropriate use. In: I. Balderjahn, C. Mennicken & E. Vernette (Eds.), New developments and approaches in consumer behaviour research, pp. 321-339. Stuttgart: Schäffer-Poeschl.

48

Bredahl, L. (1999). Consumers' cognitions with regard to genetically modified food. Results of a qualitative study in four countries. Appetite, 33, 343-360.

Brooks, L. R., Norman, G. R., & Allen, S. W. (1991). Role of specific similarity in a medical diagnostic task. Journal of Experimental Psychology: General, 120, 278-287.

Brunsø, K., Scholderer, J. & Grunert, K. G. (2004). Closing the gap between values and behavior: A means-end theory of lifestyle. Journal of Business Research, 57, 665-670.

Cohen, G. (2000). Hierarchical models in cognition: Do they have psychological reality? European Journal of Cognitive Psychology, 12, 1–36.

Cohen, J. B., & Warlop, L. (2001). A motivational perspective on means-end chains. In T. J. Reynolds & J. C. Olson (Eds.), Understanding consumer decision making: The means-end approach to marketing and advertising strategy. Mahwah, NJ: Lawrence Erlbaum.

Collins, A. M., & Loftus, E. F. (1975). A spreading-activation theory of semantic processing. Psychological Review, 82, 407-428.

Conway, M. A. (1990). Associations between autobiographical memories and concepts. Journal of Experimental Psychology: Learning, Memory, & Cognition, 16, 799-812.

den Heyer, K., Briand, K., & Dannenbring, G. (1983). Strategic factors in a lexical decision task: Evidence for automatic and attention-driven processes. Memory & Cognition, 11, 374-381.

Gengler, C. E. & Reynolds, T. J. (1995). Consumer understanding and advertising strategy: Analysis and strategic translation of laddering data. Journal of Advertising Research(July/August), 19-33.

Gollwitzer, P. M. (1990). Action phases and mind-sets. In E. T. Higgins & R. M. Sorrentino (Eds.), The handbook of motivation and cognition: Foundations of social behaviour (Vol. 2, pp. 53-92). New York: Guilford Press.

Gollwitzer, P. M. (1993). Goal achievement: The role of intentions. European Review of Social Psychology, 4, 141-185.

Gollwitzer, P. M. (1999). Implementation intentions: Strong effects of simple plans. American Psychologist, 54, 493-503.

Gollwitzer, P. M. & Bargh, J. A. (1996). The psychology of action: Linking cognition and motivation to behaviour. New York: Guilford Press.

Gollwitzer, P. M. & Bayer, U. (1999). Deliberative versus implemental mindsets in the control of action. In S. Chaiken & Y. Trope (Eds.), Dual-process theories in social psychology (pp. 403-422). New York: Guilford.

49

Gollwitzer, P. M. & Brandstaetter, V. (1997). Implementation intentions and effective goal pursuit. Journal of Personality and Social Psychology, 73, 186-199.

Gollwitzer, P. M., Heckhausen, H. & Steller, B. (1990). Deliberative vs. implemental mind-sets: Cognitive tuning toward congruous thoughts and information. Journal of Personality and Social Psychology, 59, 1119-1127.

Gollwitzer, P. M. & Moskowitz, G. B. (1996). Goal effects on action and cognition. In E.T. Higgins & A. W. Kruglanski (Eds.), Social psychology: Handbook of basic principles (pp. 361-399) New York: Guilford Press.

Gollwitzer, P. M. & Oettingen, G. (1998). The emergence and implementation of health goals. Psychology and Health, 13, 687-715.

Gollwitzer, P. M. & Schaal, B. (1998). Metacognition in action: The importance of implementation intentions. Personality and Social Psychology Review, 2, 124-136.

Gollwitzer, P. M. & Wicklund, R. A. (1985). The pursuit of self-defining goals. In J. Kuhl & J. Beckmann (Eds.), Action control: From cognition to behaviour (pp. 61-85). Heidelberg: Springer-Verlag.

Grunert, K. G., Beckmann, S. C. & Sørensen, E. (2001). Means-end chains and laddering: An inventory of problems and an agenda for research. In T. Reynolds & J. C. Olson (Eds.), Understanding consumer decision-making: The means-end approach to marketing and advertising strategy (pp. 63-90). Mahwah, NJ: Erlbaum.

Grunert, K. G., & Grunert, S. C. (1995). Measuring subjective meaning structures by the laddering method: Theoretical considerations and methodological problems. International Journal of Research in Marketing, 12, 209-226.

Grunert, K. G., Lähteenmäki, L., Nielsen, N. A., Poulsen, J. B., Ueland, O., & Åström, A. (2001). Consumer perceptions of food products involving genetic modification: Results from a qualitative study in four Nordic countries. Food Quality and Preference, 12(8), 527-542.

Grunert, K. G., Sørensen, E., Bredahl, L., & Nielsen, N. A. (1995). Analysing food choice from a means-end perspective. In F. Hansen (Ed.), European Advances in Consumer Research (pp. 366-371). Provo, UT: Association for Consumer Research.

Gutman, J. (1982). A means-end chain model based on consumer categorisation processes. Journal of Marketing, 46, 60-72.

Herrmann, A. (1996). Wertorientierte Produkt- und Werbegestaltung. Marketing - ZFP, 18, 153-163.

Higgins, E. T., Bargh, J. A. & Lombardi, W. (1985). Nature of priming effects on categorisation. Journal of Experimental Social Psychology, 11, 59-69.

50

Hofstede, F. t., Audenaert, A., Steenkamp, J.-B. E. M. & Wedel, M. (1998). An investigation into the association pattern technique as a quantitative approach to measuring means-end chains. International Journal of Research in Marketing, 15, 37-50.

Johnson, K. E. & Mervis, C. B. (1997). Effects of varying levels of expertise on the basic level of categorisation. Journal of Experimental Psychology: General, 126, 248-277.

Johnson, K. E. & Mervis, C. B. (1998). Impact of intuitive theories on feature recruitment throughout the continuum of expertise. Memory & Cognition, 26, 382-401.

Jolly, J. P., Reynolds, T. J. & Slocum, J., John W. (1988). Application of the means-end theoretic for understanding the cognitive bases of performance appraisal. Organizational Behaviour and Human Decision Processes, 41, 153-179.

Kelly, G. A. (1955). The psychology of personal construct. New York, NY: Norton.

Lopez, A., Atran, S., Coley, J. D., Medin, D. L. & Smith, E. E. (1997). The tree of life: Universal and cultural features of folkbiological taxonomies and inductions. Cognitive Psychology, 32, 251-295.

Malt, B. C. (1994). Water is not H2O. Cognitive Psychology, 27, 41-70.

Malt, B. C., & Johnson, E. C. (1992). Do artifact concepts have cores? Journal of Memory and Language, 31, 195-217.

McNamara, T. P. (1992). Theories of priming: I. Associative distance and lag. Journal of Experimental Psychology: Learning, Memory, and Cognition, 18, 1173-1190.

McNamara, T. P. & Altarriba, J. (1988). Depth of spreading activation revisited: Semantic mediated priming occurs in lexical decisions. Journal of Memory & Language, 27, 545-59.

Medin, D. L., Lynch, E. B., Coley, J. D., & Atran, S. (1997). Categorisation and reasoning among tree experts: Do all roads lead to Rome? Cognitive Psychology, 32, 49-96.

Murphy, G. L., & Lassaline, M. E. (1997). Hierarchical structure in concepts and the basic level of categorisation. In K. Lamberts & D. Shanks (Eds.), Knowledge, concepts, and categories (pp. 93-131). London: Psychology Press.

Nielsen, N. A., Bech-Larsen, T. & Grunert, K. G. (1998). Consumer purchase motives and product perceptions: A laddering study on vegetable oil in three countries. Food Quality and Preference, 6, 455-466.

Nielsen, N. A., Sørensen, E. & Grunert, K. G. (1997). Consumer motives for buying fresh or frozen plaice - a means end chain approach. In: J. B. Luten, T. Børresen & J. Oehlenschläger (Eds.), Seafood from producer to consumer - Integrated approach to quality, pp. 31-43. Amsterdam: Elsevier.

51

Olson, J. C. (1989). Theoretical foundations of means-end chains. Werbeforschung & Praxis, 5, 174-178.

Olson, J. C. & Reynolds, T. J. (1983). Understanding consumers’ cognitive structures: Implications for advertising strategy.” In L. Percy & A. G. Woodside (Eds.), Advertising and consumer psychology (pp. 77–90). Lexington, MA: Lexington Books.

Pieters, R., Baumgartner, H. & Allen, D. (1995). A means-end chain approach to consumer goal structures. International Journal of Research in Marketing, 12, 227–244.

Reynolds, T. J. (1995). Applying laddering data to communications strategy and advertising practice. Journal of Advertising Research(July/August), 9-17.

Reynolds, T. J. & Craddock, A. B. (1988). The application of the MECCAS model to the assessment of advertising strategy. Journal of Advertising Research, 29(April/May), 43-54.

Reynolds, T. J. & Gengler, C. (1991). A strategic framework for assessing advertising: The animatic vs. finished issue. Journal of Advertising Research, 31(5), 61-72.

Reynolds, T. J., Gengler, C. E. & Howard, D. J. (1995). A means-end analysis of brand persuasion through advertising. International Journal of Research in Marketing, 12, 257-266.

Reynolds, T. J. & Gutman, J. (1984). Advertising is image management: Translating image research to image strategies. Journal of Advertising Research, 24(1), 27-36.

Reynolds, T. J. & Gutman, J. (1988). Laddering theory, methods, analysis, and interpreta-tion. Journal of Advertising Research, 29, 11-31.

Reynolds, T. J. & Olson, J. C. (Eds.) (2001). Understanding consumer decision-making: The means-end approach to marketing and advertising strategy. Mahwah, NJ: Erlbaum.

Reynolds, T. J. & Rochon, J. P. (1991). Means-end based advertising research: Copy testing is not strategy assessment. Journal of Business Research, 22, 131-143.

Rips, L. J. (1989). Similarity, typicality, and categorisation. In S. Vosniadou & A. Ortony (Eds.), Similarity and analogical reasoning (pp. 21-59). Cambridge: Cambridge University Press.

Roehrich, G. & Valette-Florence, P. (1991). A weighted cluster-based analysis of direct and indirect connections in means-end chains: An application to lingerie retail. Workshop on values and lifestyle research in marketing, Brussels, EIASM.

Rosch, E., Mervis, C. B., Gray, W. D., Johnson, D. M., & Boyes-Braem, P. (1976). Basic objects in natural categories. Cognitive Psychology, 8, 382-439.

Ross, B. H. & Murphy, G. L. (1999). Food for thought: Cross-classification and category organisation in a complex real-world domain. Cognitive Psychology, 38, 495-553.

52

Scholderer, J., Brunsø, K. & Grunert, K. G. (2002). Means-end theory of lifestyle: A replication in the UK. Advances in Consumer Research, 29, 551–557.

Schwartz, S. H. (1992). Universals in the content and structure of values: Theoretical advances and empirical tests in 20 countries. In M. P. Zanna (Ed.), Advances in experimental social psychology (Vol. 25, pp. 1-65). San Diego, CA: Academic Press.

Shah, J. Y. & Kruglanski, A. W. (2002). Priming against your will: How accessible alternatives affect goal pursuit. Journal of Experimental Social Psychology, 38, 368-83.

Shah, J. Y. & Kruglanski, A. W. (2003). When opportunity knocks: Bottom-up priming of goals by means and its effects on self-regulation. Journal of Personality & Social Psychology, 84, 1109-22.

Shiffrin, R. M. & Schneider, W. (1977). Controlled and automatic human information processing: II. Perceptual learning, automatic attending, and a general theory. Psychological Review, 84, 127-190.

Søndergaard, H. A., (in press). Market oriented new product development – How can a means-end chain approach affect the process? European Journal of Innovation Management.

Tanaka, J. W., & Taylor, M. (1991). Object categories and expertise: Is the basic level in the eye of the beholder? Cognitive Psychology, 23, 457-482.

Valette-Florence, P. & Rapacchi, B. (1991). Improvements in means-end chain analysis: Using graph theory and correspondence analysis. Journal of Advertising Research, 31(1), 30-45.

Valette-Florence, P., Sirieix, L., Grunert, K. G. & Nielsen, N. A. (2000). Means-end chain analyses of fish consumption in Denmark and France: A multidimensional perspective. Journal of Euromarketing, 8, 15-27.

Zacks, J. M. & Tversky, B. (2001). Event structure in perception and conception. Psychological Bulletin, 127, 3-21.

Zeithaml, V.A. (1983). Consumer perceptions of price, quality, and value: A means-end model and synthesis of evidence. Journal of Marketing, 52, 2-22.

the validity of the means-end chain model of consumer

Documents