The Dimensional Structure of Interpersonal Trust: Basic Dimensions and Implications

Interpersonal trust is a key concept in social theory, but there is still ongoing debate about how to measure it and what its core dimensions are. There is no consensus on whether there are two or three core dimensions. This paper introduces a word-embedding approach to discovering the dimensional structure of interpersonal trust based on a comprehensive list of trust-related keywords. The starting point is to identify useful keywords based on a computational analysis of (1) 14 widely used traditional questionnaires of Likert scale items for measuring interpersonal trust and (2) a supplementary data resource—a unique Twitter-based dictionary of interpersonal trust (Hu et al. 2023).

Two studies were developed to identify keywords useful for measuring interpersonal trust expressed in textual data. (1) Study 1 used natural language processing (NLP) analyses to synthesize two data sources, extracting keywords from the 14 questionnaires with the Term Frequency-Inverse Document Frequency (TFIDF) algorithm and identifying relevant keywords from the Twitter-based dictionary through a clustering algorithm (Fiske et al. 2022). The results indicated that 51 most common words (16 words from the questionnaires and 35 from the dictionary) can be used for representing interpersonal trust and further classification. (2) Study 2 extracted the spatial coordinates of these 51 keywords in the 300-dimensional pre-trained Google News Word Embedding space. These spatially represented keywords were subject to a K-means clustering algorithm. The clustering analysis resulted in three distinct clusters, each corresponding to a different continuous dimension of interpersonal trust.

These clusters represent a nuanced view that captures traditional elements like competence and integrity while highlighting a continuum of trust, from caution and skepticism to full distrust. This study demonstrates how computational methods can help identify fundamental dimensions of interpersonal trust and offer new opportunities for trust research, especially using digital and open-ended data.