Improving means-end-chain studies by using a ranking method to construct hierarchical value maps

https://doi.org/10.1016/j.foodqual.2003.09.001Get rights and content

Abstract

There is a need for more methodological guidance for researchers using the means-end-chain theory to investigate consumers' motivations for product choice. Particularly, there is a need to improve the quantitative analysis of the qualitative data that constructs the main output of laddering studies: the hierarchical value maps. Existing heuristics do not always successfully address issues associated with: (a) large datasets; (b) variation in response frequencies across levels of abstraction (attributes, consequences, values); (c) differences in response frequencies across groups (e.g. culture); and (d) variation in laddering administration (e.g. “hard” vs “soft” laddering). A new approach is described that uses a “top-down cut-off” strategy, which is driven by the data itself and accommodates the issues described above. The advantages of the method are demonstrated with an example using real data from two cultures (n=244).

Introduction

The means-end-chain (MEC) theory of consumer behaviour is a framework for understanding the associations that groups of consumers make between products' attributes and more personally relevant and abstract consequences and values. The most commonly used technique to elicit information within this framework is termed “laddering”. Laddering has been used widely in recent years to understand consumers' cognitive structures surrounding various products, including meat, genetically modified yoghurt and beer, fish, apples, vegetable oil, fashion stores and women's clothing (Audenaert & Steenkamp, 1997; Botschen & Thelen, 1998; Grunert, 1997; Nielsen, Bech-Larsen, & Grunert, 1998; Valette-Florence, Sirieix, Grunert, & Nielsen, 2000).

The results of a laddering study are usually displayed in hierarchical value maps (HVM). The HVM pictorially displays the linkages made by a group of consumers between product attributes, the consequences for the consumers from consuming these products and subsequently important values. The linkages between elements on the HVM provide both an understanding of some of the drivers of food choice and marketing “hooks” for the promotion of products (Audenaert & Steenkamp, 1997). Researchers have indicated that HVMs are useful for developing marketing strategies (Olson & Reynolds, 2003) by applying them in conjunction with techniques such as the MECCAS (Reynolds & Craddock, 1988). Other researchers have used HVMs to compare cognitive structures between groups of consumers in different countries (Bredahl, 1998; Grunert, 1997; Grunert et al., 2001). HVMs have also been used to understand brand persuasion (Reynolds, Gengler, & Howard, 1995) and are seemingly useful for making new products by disentangling the so-called consumer black box of exciting and often unexplored new benefit areas (Russell et al., in press). Despite widespread application and some detailed discussion of theoretical and practical considerations of laddering, there is a dearth of published guidance for the researcher for many aspects of the method, particularly the quantitative analysis of the qualitative laddering data that is the main determinant of the construction of the HVM. The interpretations of laddering studies are dependent upon the construction and composition of HVMs; therefore it is crucial to have rigorous, justifiable, and transparent methods for their construction.

The HVM is constructed from frequencies of linkages (connections between the elements of the levels of abstraction) that are summarised in a structural implication matrix (SIM) (Reynolds & Gutman, 1988). Two elements are displayed on a HVM when the number of links between them in the SIM exceeds a specified value, termed the cut-off level. The cut-off level specifies the threshold for the number of links before any connection is included on the HVM. There is a particular problem for investigators to determine at what frequency a connection between two levels of abstraction is termed “significant” or “important” enough to appear on the HVM. Researchers continuously find themselves in a quandary of trying to find a balance between providing just enough detail on the HVM to afford practical as well as meaningful results. The appropriate determination of the cut-off level is important as it will largely determine the shape and information portrayed on the HVM and thus the conclusions drawn from the study as well as the applicability of the results to, for example, marketing strategies. A high cut-off (large frequency of reported linkages) will create a simplified map with few linkages but may miss useful information; conversely a low cut-off (low frequency of reported linkages) will create a complicated map that may be difficult to interpret.

Previous research (Pieters, Baumgartner, & Allen, 1995) has suggested four heuristics for choosing cut-offs. The first is pragmatic and determined by whatever “leads to the most informative and interpretable solution” (p. 239); an approach that has also been recommended by others (Audenaert & Steenkamp, 1997). The second is a type of “goodness of fit” in which the cut-off is chosen on the basis of the sample size and the number of ladders that can account for two-thirds of all relations (Reynolds & Gutman, 1988). The third approach suggests creating a type of scree plot based on the number (or percentage) of connections and various cut-off levels and looking for some kind of “elbow” in the scree. The fourth approach, chosen by Pieters et al. (1995) is similar to heuristic 2, in that within the data matrix a comparison is made between the proportion of “active cells” to the proportion of all connections at a given cut-off. A judgement is then made on what reflects a large percentage of the total number of connections accounted for by a relatively small number of distinct relations. The problem is that for some data these heuristics or “rules of thumb” are not always successful. Work in our laboratory (Flight, Russell, Blossfeld, & Cox, 2003) found that in trying to account for two-thirds of relations (for example using heuristics 2 and 4) very low cut-offs were required; this created an HMV that was too complex to graphically display or easily interpret. Alternative methods to these heuristics for determining cut-offs are required in laddering research. (Grunert, Grunert, & Sorensen, 1995) also echo this sentiment with their suggestion that the use of arbitrary cut-off levels is an important issue to be considered in laddering research.

It can also be common in laddering studies that consumer panels with typically 50–60 individuals provide a very large number of (partial) ladders. In such a case the number of linkages generated could be of the order of 1000 or more, and the procedure suggested above would produce a small cut-off value (around four or five linkages) that would subsequently lead to many connected links in a complicated HVM.

The matter has become more complicated with the emergence of different forms of eliciting means-end-chains such as the Association Pattern Technique (Feunekes & den Hoed, 2001; ter Hofstede, Audenaert, Steenkamp, & Wedel, 1998) and various forms of hard laddering (Botschen & Thelen, 1998; Russell et al., in press; Walker & Olson, 1991). These studies have shown that the number of ladders elicited, and therefore the make-up of the HVM, is determined by the initial elicitation method (hard vs soft, computerised vs pencil-and-paper, APT) as well as the analysis methods leading to the production of the HVM. Botschen and Thelen (1998), for example, compared hard (pencil-and-paper) and soft (interview) laddering. In this study, the soft laddering technique produced higher frequencies of “elements”, especially at the attribute and consequence level, and hence more ladders were elicited than by the hard laddering technique. Despite this, both methods were reported with a cut-off level of 2. Comparisons of HVMs can be misleading by using a cut-off level that does not reflect the difference in the sizes of datasets. Laddering studies such as this would be more likely to produce comparable and consistent results if an improved means for determining the cut-off level was adopted which was responsive to the variations in frequencies of elements elicited for each group.

Another complicating factor is that a group of participants in a laddering study will not necessarily make the same total number of linkages between any two levels of abstraction. Participants subjected to soft laddering, for example, are likely to make more linkages at lower levels of abstraction (e.g. from attribute to functional consequence) than at higher levels (e.g. from psychosocial consequence to instrumental value) as it becomes increasingly more difficult for them to suggest responses as the level of abstraction increases. This suggests that using the same value as the cut-off level may not be appropriate where the number of links between various levels of abstraction is (inevitably) variable. That is, by choosing the same cut-off level between attributes and consequences, and between consequences to values, implies that the HVM will display relatively more of the attribute–consequence linkages when there is a greater number of total linkages at the attribute to consequence level. In this paper we search for a method of analysis that will take into account the number of linkages selected between the various levels of abstraction.

Laddering studies have been utilised previously to understand international markets (for example, Bredahl, 1998; Grunert, 1997; Nielsen et al., 1998). However, variations in datasets between cultural (national) groups are not uncommon in laddering studies, and previously have been analysed using the same or similar cut-off value even when comparisons are desired. For example Bredahl (1998) sought to compare the cognitive structures associated with genetically modified yoghurt and genetically modified beer in Denmark, Germany, Italy and the United Kingdom. Cross-national differences were of interest and were to be determined by the links on the HVM. The same number of interviews (50) was conducted in each country for each product. The total number of ladders elicited for the yoghurt ranged from 530 in Italy to 643 in Denmark and from 405 ladders for beer in the UK to 543 for beer in Denmark. The corresponding averages for Denmark, Germany, UK and Italy were 3.5, 2.8, 2.4 and 2.6 for yoghurt and 3.1, 2.6, 2.4 and 2.2 for beer, respectively. Despite these differences in the number of ladders elicited, the HVMs were reported with various cut-off levels such as a cut-off of 4 for Denmark and 5 for Germany for yoghurt even though the average number of ladders was lower for the German respondents for this product. There is a clear need to be able to deal with such differences in datasets to facilitate meaningful comparisons.

In summary, the matter of how to determine the size of the cut-off value to obtain a balanced and interpretable HVM is unresolved. It is complicated by the fact that the magnitude of the entries in the SIM varies according to number of participants as well as the total number of links chosen by the group of participants and the particular level of abstraction. Researchers using the laddering technique face the dilemma that they may generate large datasets if they are using a topic where many linkages are elicited such as in situations of high consumer involvement or complexity of the topic, or when using a hard laddering techniques (Russell et al., in press). In these instances, in particular, we have found it inadvisable to use previous suggestions of methods (Pieters et al., 1995) for determining the cut-off level as those methods result in the generation of too many “real” connections to be portrayed or interpreted successfully on a HVM. Furthermore, the value of laddering as an approach to understanding markets needs to be enhanced by accommodating differences in datasets according to group, such as gender or culture.

A strategy is presented in this paper that determines the cut-off level in a way that addresses the issues outlined above and hence produces a better and clearer interpretation of laddering data.

Section snippets

Methods

To illustrate our proposed strategy to determine an appropriate cut-off level, we use data obtained from a study investigating mothers' perceptions of the perceived role for food in influencing a child's cognitive performance. Aspects of the study are described in more detail in a previous paper (Russell et al., in press).

Results and discussion

The amount of detail to be included on a HVM is dependent upon the aims and the purpose of the study. The aim of the current study was to understand the perceptions of Australian and Malaysian mothers regarding the characteristics of breakfasts that may improve the cognitive performance of their child. Therefore, it was of interest to examine what were the strongest linkages (most widely held or agreed upon by the group) between the attributes of breakfast cereals, their physical consequences

Conclusions

In conclusion, currently HVM construction would appear to be determined by considerable subjectivity, pragmatism and investigator bias, which makes validation through repetition problematic and comparison across studies and groups difficult. A need for an improved method for determining at what level a link is allowed to appear on the HVM was recognised. Because the number of links and number of ladders will vary according to the study, the number of participants, the levels of abstraction and

Acknowledgements

We thank Unilever Research Vlaardingen, particularly Mr. Jeroen A. van Lawick van Pabst for data collection support and permission to report the data.

References (20)

There are more references available in the full text version of this article.

Cited by (82)

  • Preschool children's preferences for library activities: Laddering interviews in Chinese public libraries

    2019, Library and Information Science Research
    Citation Excerpt :

    The cut-off level depended on the number of participants who constructed ladders (de Costa et al., 2007) and was set at three for the itsy-bitsy spider activity and two for the other two activities. The researcher set these cut-off levels for goodness of fit based on the sample size and the number of ladders that could account for two-thirds of all relations (de Costa et al., 2007; Leppard, Russell, & Cox, 2004; Reynolds & Gutman, 1988). In this step, codes that had related meanings and similar paths to other elements were merged or positioned adjacently to strengthen the links and then retained in the network (Reynolds & Gutman, 1988).

View all citing articles on Scopus
View full text