Proffitt, J.B., Coley, J.D., & Medin, D.L. (2000). Expertise and category-based induction. Journal of Experimental Psychology: Learning, Memory, and Cognition, 26(4), 811-828.

 Purpose

 This article examined whether different models of category-based induction could account for the patterns of reasoning used by experts in a particular domain. In inductive tasks, participants use information from one set of categories to make inferences about another category. People may, for example, be told that one or more categories have a certain property and they then are asked to infer whether the entire superordinate category contains the same property. Researchers use novel or “blank” properties to prevent subjects relying only on knowledge retrieval.

 Typicality effects have been found in experiments in which participants are required to judge the strength of arguments. That is, people make stronger inferences from typical category members (such as robin) than they do from atypical category members (such as turkey) to the entire category (bird). The diversity phenomenon occurs when people make inferences based on how differently they regard two or more category members. Thus, for example, robins and turkeys are seen as more diverse than robins and sparrows. People are more likely to generalize a characteristic from the more diverse members to the whole group, than from the less diverse members. Sparrows add little coverage to robin whereas turkey adds a lot more coverage to cardinal.

 Two models often used to explain typicality and diversity were discussed. The similarity-coverage model (SCM) describes these effects in terms of coverage. According to the SCM, typicality effects occur because typical members are more similar to other category members. Diversity occurs because members of the category that are seen as very different to one another cover more of the category than members that are seen to be more similar to each other. The feature-based induction model (FBIM) uses a connectionist mechanism and was not specifically tested in this article.

 Both typicality and diversity effects have not been found universally. For example, in a study conducted in central Guatemala, the Itzaj Maya showed evidence of using typicality, but not of using diversity in inference reasoning tasks. This does not support the SCM which is based on a unified framework to encompass both diversity and typicality. The reason suggested for this dissociation was that Itzaj had a more extensive knowledge base in the studied domain, which enabled them to use a variety of strategies to solve reasoning tasks.

 These experiments examined typicality and diversity reasoning among different types of US tree experts. In order to assess the degree to which these experts evaluated argument strength on the basis of coverage. These experts included landscapers, taxonomists, and parks maintenance personnel who had a similar culture to US undergraduates, but had the extensive domain knowledge of experts. The expertise enabled them to use a variety of reasoning strategies.

 When reading this paper, it is important to keep in mind that taxonomists classify living things by: Kingdom/Phylum/Class/Order/Family/Genus/Species/Sub-species. Folk generic classification is not as structured. It is based on commonly held beliefs such as common names, location found, functions, and other loose categories.

 For example:

Experimental work

Experiment 1

Subjects first performed a sorting task to classify the experts. There were three categories of experts: landscapers, taxonomists and parks maintenance personnel. Global coverage of the whole category was assessed. This global score was based on the individual’s sorting behavior and was designed to assess the psychological distance between the trees. These scores were averaged between each group. The subjects were instructed to: “Put together the trees that go together in nature.”

Subjects were presented with two hypothetical diseases that affected different species of trees. They were then asked which disease would affect more of the other kinds of trees in the area. Single-premise (each disease affecting one type of tree) and dual-premise (each disease affecting two types of trees) items were presented. Subjects were asked which of two arguments provided better support for a conclusion and to explain their reasoning.

Results showed that tree experts do not use global coverage to guide induction for single-premise items. Family size (that is, the disease that infects the tree with the largest family)  was a significant justification for all responses. This indicated that they used local coverage. Reasons were collapsed into similarity-based and causal ecological. Overall, experts justified choices on the basis of causal and ecological mechanisms of disease spread.

For dual-premise items, taxonomists used global coverage (supporting diversity) and family size. There was little agreement on the justifications used by the experts. When the categories were collapsed again, this disagreement continued. Landscapers and maintenance workers used more causal-ecological justifications and taxonomists used more similarity based justifications.

Experiment 2

Subjects were 17 of the 23 tree experts who had participated in the first experiment. The second experiment occurred 25 months after the first one. Experiment 2 was the same as Experiment1 except that subjects were asked if they believed that a novel disease would affect all trees.

Results for single premise items indicated that although global-based responses increased from the level of experiment 1 and the change in wording resulted in an increase in diversity-based justifications for all experts the findings were basically the same from Experiment 1. Again, justifications varied according to the type of expert. For dual-premise items there was an increase in diversity-based predictions, but causal-ecological justification predominated.

Experiment 3

In this experiment, subjects were recruited from an arborist meeting. They were given the names a pair of trees that share a hypothetical disease and asked to name other trees that might be susceptible and why. The reasons for this open-ended questioning was to determine how far the properties could be extended and also to determine whether the premise pairs were plausible.

 For trees linked at the genus level, bridging patterns indicated local coverage. When the trees were linked at a higher order, there was more linking to other members of the same folk-generic category.

Conclusions

Overall, the experts’ patterns of reasoning were more similar to those of the Itzaj Maya than to that of undergraduate students. Also, experts use more local than global coverage in reasoning. Experts in different areas of specialization show different patterns of reasoning and changes in instructions lead to subtle differences in patterns of reasoning.

Although novices and experts reason differently, it is not a case of them using different strategies. Both groups can use many strategies, but the experiment design may inhibit some strategies. Typicality and diversity are two frequently used strategies, but using them does depend on domain knowledge. Use of local coverage takes more expertise than using global coverage, and the causal–ecological strategies depend on an extensive knowledge in the domain.

Discussion questions

The article claims that undergraduates do in fact have more reasoning strategies than they use, and that these strategies are constrained by the experiment requirements. How would you test this?

Was the sorting method an accurate method of obtaining a global score given that taxonomic distance (not psychological distance) is a good predictor of whether trees will get the same disease? Landscapers justified decisions of diversity based on a taxonomy that was more similar to that of maintenance workers.

Does undergraduate reasoning make sense based on lack of explicit knowledge? Think about the spreading of colds. Our intuition on how we catch a cold often has no basis in science. Is that faulty reasoning or lack of knowledge? For example, how would reasoning change if students were given the classification of trees, or given the information that gingko trees were highly resistant to leaf diseases.

We are not given details of how the experts classified the trees, and so we don’t know if all/some experts classify trees according to folk biology. We are given no criteria for inclusion into the category of expert other than work experience in the area. These experts had varying amount of formal education, and not all had training in the area of trees.  Does working with trees necessarily mean they are experts? How much does this affect reasoning ability and their ability to categorize trees and give plausible justifications?

How does experience (practical knowledge) versus theoretical knowledge (book learning) affect decisions and justifications?

Do the concepts of local and global coverage represent fundamentally different strategies or are they versions of the same idea?

Is expertise a function of having more strategies?

Would strategies change if proper Latin names were used?

Does the small n affect results

This page was last updated:
07/18/2006 00:36