The spatial arrangement method of measuring similarity can capture high-dimensional semantic structures

Area of Science:

• Cognitive Psychology
• Psychometrics
• Computational Linguistics

Background:

• Psychologists gather similarity data to understand categorization, generalization, and mental representations.
• Collecting pairwise similarity judgments is time-consuming, necessitating efficient methods.
• The Spatial Arrangement Method (SpAM) was developed to expedite similarity measurement by having participants arrange items in a 2D space.

Purpose of the Study:

• To evaluate the effectiveness of SpAM for measuring similarity in higher-dimensional, conceptual stimuli.
• To assess SpAM's ability to capture semantic structures of word categories and personality traits.
• To determine the reliability and accuracy of SpAM for high-dimensional data.

Main Methods:

• Study 1: Participants used SpAM to arrange words from eight categories; distances were compared to pairwise similarity judgments and analyzed using multidimensional scaling.
• Study 2: SpAM was used to map the Big Five personality trait adjectives, with solutions validated across dimensions.
• Reliability was assessed using split-half correlations and inter-participant agreement.

Main Results:

• SpAM distances showed strong correlations (r = .71) with traditional similarity judgments.
• Aggregated SpAM data favored higher-dimensional solutions (more than 2 dimensions) for most word categories.
• SpAM demonstrated high split-half reliability (r = .90) for dissimilarities, despite low inter-participant agreement (r = .15).
• SpAM successfully recovered the Big Five personality factor space, favoring 4- or 5-dimensional solutions.

Conclusions:

• SpAM is a reliable and accurate method for measuring similarity among high-dimensional conceptual stimuli, such as words.
• The method effectively captures underlying semantic and psychological structures, even when individual participants vary in their focus.
• The study provides valuable, publicly available data for future research in cognitive and quantitative psychology.