On the hierarchical nature of means–end relationships in laddering data

https://doi.org/10.1016/j.jbusres.2006.10.004Get rights and content

Abstract

Means–end analysis appeals to managers, and laddering is applicable for a large diversity of products and services. The cognitive model underlying conventional means–end analysis is hierarchical. Means serve certain ends, which, in turn, serve as means to higher-ordered ends. These ladders stretch from concrete attributes to abstract values. However, the question needs to be considered of whether or not means–end relations are really hierarchical. This article reviews the doubts with respect to this issue in the literature, and analyzes means–end datasets for two empirical studies. Here, the means–end relations turn out to be symmetrical rather than asymmetrical, that is, if respondents say that A is a means to B, they are also likely to say that B is a means to A. This contradicts the hierarchy assumption, and the conclusion is that means–end relations are not necessarily hierarchical. When means–end relations are symmetrical, rather than asymmetrical, a network representation is more adequate than a hierarchical value map (HVM). In that case the centrality of a concept in the network is the key to the prominence of a concept, rather than its level in the HVM. The implication of these findings is that in empirical applications, the hierarchy assumption underlying laddering needs testing. Failure of such tests to confirm the hierarchy assumption has consequences for the interpretation of the results and for the policy recommendations, based on the research.

Introduction

Means–end analysis is in wide use in marketing for understanding consumer behavior. The basic idea of means–end analysis is that consumer knowledge relating to product and brands exists at different levels of abstraction and that these levels relate hierarchically (Reynolds and Olson, 2001). Means–end analysis underlying notion is that decision-makers choose that course of action (e.g., the purchase of a specific brand) that is most likely to achieve desired outcomes. Means–ends analysis unravels and shows the structure of these means and ends, which provides information about why consumers do or do not like and buy specific products or brands.

Actual efforts to discover the structure of means–ends linkages that consumers use in their thinking about products follow from the seminal work of Reynolds and Gutman, 1984, Reynolds and Gutman, 1988. From the eighties on, this work became widespread, as is illustrated by the 1991 special issue of Journal of Business focusing on means–ends research. Applications of means–ends analysis in marketing are available on a large scale, with application domains varying from ski resorts (Klenosky et al., 1993), to fashion (Botschen and Hemetsberger, 1998), and beef (Ter Hofstede et al., 1998). The method is useful for benefit-based market segmentation, image management, advertising strategy, sales force motivation, fundraising, industrial marketing and financial engineering (Botschen et al., 1999, Reynolds and Olson, 2001). The results of means–end analysis tend to have a great deal of managerial appeal. However, its methodology and its theoretical underpinnings have evolved in an informal, somewhat haphazard manner (Reynolds and Olson, 2001).

The mostly unquestioned assumption of means–end analysis is that a hierarchy exists, given by the direction of the means–end relations as found in interviews and questionnaires (Cohen and Warlop, 2001). This hierarchy implies asymmetry. If A is a means to B, B is not also a means to A. However, research has never addressed the issue whether or not means–end relations are asymmetric enough to provide a sufficient basis for the assumption of a hierarchy. This question becomes the more urgent as researchers invested in the development of more advanced methods for analyzing means–end data. For example, Valette-Florence and Rappachi (1991) used methods from graph theory for deriving the Hierarchical Value Map from the raw means–end data. Ter Hofstede, Audenaert, Steenkamp and Wedel (1998) developed the Association Pattern Technique (APT), which is very useful for measuring means–end chains in large surveys. These methods do not test the assumption of a hierarchy, but take it as given.

This study questions the hierarchy assumption underlying means–end analysis. The article first explains the a-priori reasons for doubts regarding this hierarchy assumption. Subsequently, the article examines two typical means–end data sets on the extent to which these support the notion of hierarchy. The means–end relations in both data sets will turn out to be symmetrical rather than asymmetrical. This contradicts the hierarchy assumption. This article discusses how a wrongful assumption that a hierarchy exists can produce misleading marketing recommendations, and proposes an alternative (non-hierarchical) representation for situations where hierarchy does not apply.

Literature shows surprisingly little discussion about the assumption that the underlying cognitive model of means–end analysis is hierarchical in nature. Whether such a hierarchical structure exists in the consumer's mind is not investigated, but just assumed (Cohen and Warlop, 2001). However, an alternative model exists, in which consumers just have networks of concepts in their minds, systems of interlinked concepts, where one concept gets its meaning from its links with other concepts. Two elements characterize such networks: concepts, also called nodes, and their associations, also called links. In principle, means–end chains are semantic networks representing knowledge about products. Semantic networks can be hierarchical or non-hierarchical. Research on semantic memory shows how nonhierarchical semantic networks or association patterns can perfectly capture human knowledge (Chang, 1986, Grunert et al., 2001). A hierarchical goal structure is a special type of semantic network (Anderson, 1983). The means–end structures belong to this special hierarchical class of semantic networks. A basic difference between a (general) semantic network and a hierarchical means–end network is that in the traditional means–end network relations are inherently directional and, therefore, asymmetric. In a non-hierarchical semantic network, a link between A and B can just as well be a link between B and A. However, if A is a means to achieve B as the end, one cannot just turn this around and say that B is also a means to achieve A. In a network where relations are not primarily directional, centrality in the network indicates the prominence of a concept, rather than its level in the hierarchy. In this case, the focus should not be on the concepts that are high in the hierarchy, but on the concepts that are most central.

If the concepts in a consumer's mind are not necessarily structured in a hierarchical way, why do researchers then so often find hierarchical means–end structures? The directionality observed in means–end structures can, to a large extent, be explained as an artifact of the data collection technique used. Most versions of the paper-and-pencil method of Walker and Olson (1991) suggest an order to the respondents by the arrows connecting the boxes in the questionnaire. The set-up of most laddering interviews is such that these literally “push” respondents up a hierarchy in an effort to discover hierarchically linked concepts (Cohen and Warlop, 2001). Thus, the data collection technique may be too directive, imposing on respondents a specific structure (a potential research artifact) and a sequence of responses to elicit from this structure (Bagozzi and Dholakia, 1999).

Studies reporting means–end relations where the means–end direction runs in both ways feed these doubts regarding the hierarchy assumption. For example, Pieters, Baumgartner and Allen (1995) found almost as many entries implying that physical appearance (means) leads to self esteem (end), as entries implying that self-esteem (means) leads to physical appearance (end). Bagozzi et al. (1996) found that save environment (means) leads to avoid landfills (end) and at the same time avoiding landfills (means) leads to save environment (end). Scholars also found loops. Pieters et al. (1995) report, for instance, a loop of means–end relations between the goals of “self-esteem”, “confidence” and “achievement”. Botschen et al. (1999) found a loop between “right clothing”, “feeling good” and “reduce uncertainty”. This circularity destroys the hierarchical form of a map (Eden et al., 1992). Such loops appear too often to be considered as mere incidents. In conclusion, literature gives ample reason to put the hierarchy assumption that underlies means–end analysis under closer scrutiny.

Section snippets

Method

This study investigates the directionality of means–end relations in two different data sets. The aim of this study is to examine the directionality of means–end relationship and to answer the question whether a hierarchical structure or a nonhierarchical network provides the best representation of the data.

This study carries out a conventional laddering study among a sample of respondents from the relevant population, under the assumption that a hierarchical model applies. This results in a

Conclusions

This article starts with expressing doubts if means–ends structures are really hierarchical. Two means–end datasets have shown that there is more reason to interpret the means and ends as connected nodes in a non-hierarchical semantic network, than as means and ends in a hierarchical goal structure. This is a significant departure from the theoretical model underlying means–end structures in marketing. The hierarchical view of means–end structures tends to focus on the concepts at the top of

References (27)

  • Günther Botschen et al.

    Using means–end structures for benefit segmentation. An application to services

    Eur J Mark

    (1999)
  • Tien M. Chang

    Semantic memory, facts and models

    Psychol Bull

    (1986)
  • Joel B. Cohen et al.

    A motivational perspective on means–end chains

  • Cited by (46)

    • “Misalignments between users and designers as source of inspiration: A novel hybrid method for physical new product development”

      2022, Technovation
      Citation Excerpt :

      The MUD method shares some of the limitations of the means end chain and laddering methodologies, which assume the existence of a hierarchical cognitive structure in consumers' mind, retrieved through the laddering interviews (Cohen and Warlop, 1995). It might be that such hierarchical structure is an artefact of the data collection technique used, which might “push” respondents to create, instead of retrieve, hierarchically linked concepts up to abstract values (Cohen and Warlop, 1995; Van Rekom and Wierenga, 2007). To mitigate this potential effect, we terminated the laddering procedure at the level of users’ goals (rather than continuing up to the abstract values associated to these goals).

    • The significance of avoiding household food waste – A means-end-chain approach

      2018, Waste Management
      Citation Excerpt :

      Evidentially, the means-end-chain theory is useful in understanding consumer behavior (Ter Hofstede et al., 1998; Van Rekom and Wierenga, 2007) and provides the opportunity for getting brand persuasion, resulting from feelings and purchase intentions of consumers (Reynolds et al., 1995). Thus, information about why consumers do like/do buy or do not like/do not buy products can be gathered (Van Rekom and Wierenga, 2007). The data for preparing means-end-chains are collected by the qualitative interview technique laddering (Aurifeille and Florence, 1995; Leppard et al., 2004; Barrena et al., 2015).

    • To earn is not enough: A means-end analysis to uncover peer-providers’ participation motives in peer-to-peer carsharing

      2017, Technological Forecasting and Social Change
      Citation Excerpt :

      To analyze participation motives, we employed the MEC technique because it allows for a detailed analysis of a consumer's usage motives and cognitive motive structures by linking participation relevant attributes, the resulting utility components, and underlying individual values, which can serve as a basis for market segmentation (Reppel and Szmigin, 2010). The results are quantifiable (Voss et al., 2007) and of great managerial relevance (van Rekom and Wierenga, 2007). MEC analysis has been applied to a variety of marketing areas (Reppel and Szmigin, 2010) such as consumer behavior (e.g., Bagozzi and Dabholkar, 1994; Schaefers, 2013) and strategic marketing (e.g., Herrmann and Huber, 2000).

    View all citing articles on Scopus
    1

    Tel.: +31 10 4081969; fax: +31 10 4089011.

    View full text