How hierarchical structures may influence the way that we think†
This is the space reserved for copyright notices.
This paper presents experimental results showing that a user's interaction with hierarchical classification may affect the user's conceptual structures. After performing a directed browsing task using a hierarchical classificatory structure, subjects rated as more important the concepts that were organized in the top tier of the hierarchy. The results of the study suggest that, although subjects adapted their conceptual structures to the hierarchical structure in some conditions, these effects depend not only on the structure, but on the specific concepts that appear in the structure.
Much empirical research has been devoted to a user's interaction with hierarchical classificatory structures. Larson and Czerwinski (1998) found that users were more successful with two-tiered structures than with three-tiered structures. Miller and Remington (2004) found that deeper structures were more effective provided that the labels used in the structure were unambiguous. However, there has been comparatively little empirical research that explores how interaction with a hierarchical structure affects the manner in which a person thinks about the concepts represented in the structure. If consistent patterns are observed in how a classificatory structure affects a user's conceptual structures, it will help us understand what causes a structure to be more or less effective.
This study is an investigation into whether users adjust their internal conceptual structures to fit the hierarchical structure that they are browsing and, if so, which tiers in the structure are associated with largest adjustments. It is not certain whether the highest tier of the structure or the lowest tier will tend to be most salient to users. This is because, when a person browses a hierarchy from the top down, his first step is to select from the groups that occur at the first tier of the hierarchy. At that point, the principles used to divide the top tier are likely to be particularly salient to the user. Once a group is selected, then the user proceeds to select an item from within that group. At that point in the process, he is likely to attend to principles by which the items can be differentiated from each other. Theoretically, the top tier was the first to claim the user's attention, but the user may focus his attention away from that tier when making the final selection of items.
112 participants were recruited via Amazon's Mechanical Turk website <http://www.mturk.com>, an online recruitment tool for completing micro-tasks. Each of the participants was at least 18 years of age and participated via an IP address from the United States. As Mechanical Turk workers, they were proficient in navigating web pages. Each participant was paid $10 upon successful completion of the study.
The study used nine vacation descriptions of approximately 150 words in length. They were adapted from descriptions that appear in websites devoted to booking vacations.1 Each of the nine vacations was given a value for both Affordability and Activities (e.g., “High Budget,” “Sightseeing”). A pilot study found that there was no significant correlation between Affordability and Activities as important factors for subjects when planning a vacation.
The set of vacations was organized into two hierarchies. One hierarchy was organized at the top tier by Affordability, while Activities was used to distinguish the items within each group. The other hierarchy was organized in the opposite manner, with Activities used at the top tier and Affordability used to distinguish the items within each group. The study used two versions of each hierarchy, one with expressive labels (e.g., “High Budget”) and one with non-expressive labels (e.g., “Section One”). In each hierarchy, the order in which the groupings and items occurred was counterbalanced.
In the pre-test, subjects were given a list of ten vacation characteristics. Among the characteristics were Affordability and Activities. The subjects were asked to rank the characteristics in terms of how important they were when planning a vacation. After one week, each subject was given one of the hierarchies and was asked to browse until they selected a vacation that they would most like to take. Immediately after the browsing task, they were given the post-test, in which they again ranked the ten vacation characteristics.
Subjects who were given the hierarchy in which Activities was the top tier, and in which the labels were expressive, rated Affordability as significantly less important in the post-test than in the pre-test (p<.01) (see Table 1). No other changes were significant in any conditions. Previous research in categorization has shown that subjects are more likely to change their typicality judgments for items that are considered to be “moderately typical” of a category (Barsalou, 1987, p. 112). Therefore, the results were filtered to only the subjects who rated a characteristic as moderately important in the pre-test (i.e., subjects who gave the characteristic a lower ranking than 3 but a higher ranking than 8). In the filtered results, subjects who had been given the hierarchy in which Activities was the top tier, and in which the labels were non-expressive, rated Activities as significantly more important in the post-test than in the pre-test (p<.01).
DISCUSSION AND CONCLUSION
Taken together, the two significant effects found in the study support the conclusion that the top tier of the hierarchies was more salient to the subjects than the bottom tier. This was true even when the top-tier groupings were given non-expressive labels. This result suggests that the subjects responded not only to the labels, but to the groupings themselves. However, these results were found only when the top tier was Activities. When the top tier was Affordability, the results were not significant. Further research is required to determine why effects were found in only one hierarchy. One possibility is that the subjects were not accustomed to seeing vacations organized by Affordability and resisted adapting their conceptual structure to that particular framework.
The results support theories that people adjust their conceptual structures to fit the hierarchy that they are browsing (e.g., Jacob, 2001), with particular attention being paid to the hierarchy's top tier. However, people's willingness to adjust their conceptual structures may depend on the specific concepts that are used in the hierarchy. Further research into the effect of a classificatory structure on a person's conceptual structures will help us to identify approaches to classification that are more likely to result in effective information retrieval.
Table 1. Changes to rankings, from 1 (most important) to 10 (least important)
I would like to thank Dr. Elin K. Jacob and Dr. Robert L. Goldstone for their very helpful comments regarding this study's design.
I am grateful to Jodine Perkins, a former travel agent, for her assistance in adapting descriptions for use in this study.