Definitions and Measurement of Trust

In the last chapter, we reviewed three major sociological paradigms. The unit of analysis of these paradigms range from micro (studying individual decisions and dyadic interactions) to macro (studying institutions and structures as a whole). The spectrum is useful for organizing literature on the definition and measurement of trust, as paradigms influence the way researchers frame their questions. For example, a economist often study trust in a dyadic trust game, which is on the micro end of the spectrum; and a political scientist often study trust in institutions, which falls on the macro end of the spectrum.

Trust is an incredibly broad term and loosely used in many contexts. Here we do not attempt to come up with one single definition of trust, but rather provide an overview and organization of different definitions and measurement of trust found in studies in many disciplines, including sociology, economics, organizational behavior, and psychology.

Trust can either be viewed as a macro, structural phenomenon, or a micro individual process with very specific targets (A trusts B to do X).

The macro view of trust originates from Luhmann's work in the late 1970s, who describes trust as "a mechanism for the reduction of social complexity" [1], and focuses on the overall level of trust in society. Scholars like Arrow and Fukuyama believe that the level of trust in a society strongly predicts its economic success. This belief follows the functional view of Parsons -- trust can make the social systems function better. Trust in this context is usually discussed very theoretically and philosophically, making it difficult to generate empirical evidence, especially before large scale data collection became a reality as in modern day web systems. Trust in this context is usually measured by the trust question in General Social Survey (GSS) or the World Values Survey (WVS), "Generally speaking, would you say that most people can be trusted or that you can't be too careful in dealing with people?" Whenever we read articles with titles like The Decline of Trust in the United States, chances are they are using data from GSS. This measure and its variations have been widely used particularly in political science. For example, questions like "How much of the time do you think you can trust the government in Washington to do what is right?" by Miller [2]. Although these measures have also been criticized for having little predictive power when it comes to specific behaviors of individuals [3], they are useful (and have the conceptual advantage) in reasoning about the baseline level trust for a society, tracking changes long periods of time (decades) and as an additional control factor for micro-oriented studies.

The micro view of trust focuses on individual decisions and traits. This view is primarily found in economics and psychology. The strength of this view is its ability to gather empirical evidence through experiments with small groups of participants. For example, the "trust game" by Berg [4] is an investment game that uses the amount of investment to denote the level of trust in the participant's partner. One participant is given a certain amount of money and can choose an amount between zero and the total amount to give to the partner in the experiment (which measures trust). The experimenter will triple the amount that is sent to the other participant, and the other participant can decide how much between zero and the total endowed amount to send back to the first participant (which measures trustworthiness). The conceptual setup here is that the more one trusts the partner, the more amount in the beginning should be sent (total amount), as if the partner returns half of the amount endowed, both would end up with better outcomes than not cooperating. But the risk is definitely that the second partner will not return any amount for self-interest. The results show that trust indeed exists. In one of the conditions, the first participant sent on average $5.36 out of $10, and the other participant returned on average $6.46 out of ~$15 endowed.

In addition to the economic approach, micro view of trust also involves the study of individual psychology. The psychologists Rotter developed the Rotter Interpersonal Trust Scale (ITS) consisting 25 items, asking whether experts, parents, students, repairmen, salesmen and organizations could be trusted [5]. The outcome of this measure is considered as an individual trait similar to personality trait. But if one examines the items in ITS closely, some items are actually quite broad and bearing a lot of similarity with the GSS, such as "Most people can be counted on to do what they say they will do.", which we already pointed out having little predictive power in specific behaviors such as in the trust game [3]. In addition, some items on scale are also quite domain specific -- item 23 on the scale is "Most repairmen will not overcharge, even if they think you are ignorant of their specialty" and item 24 is "A large share of accident claims filed against insurance companies are phony" is equally predictive. Trust in one domain may not transfer to another. Maybe item 23 can predict whether one is more likely to use freelancing marketplace TaskRabbit, yet it is hard to imagine that item 24 can predict something outside of the domain of insurance -- e.g. the adoption of Airbnb.

To summarize, there are many different ways to measure trust, such as general attitude question, investment game, scale with specifically designed items. With our increasing ability to collect data, such as outcome (e.g. whether a crowdfunding project is funded or not [6]), new ways of measuring trust are being developed (e.g. using natural language processing and machine learning). For example, in Ma et al. 2017, an natural language model was developed to be able to automatically predict whether an Airbnb host profile will be perceived as more or less trustworthy compared to profiles of similar lengths [7]. These new scalable approaches together with data recorded by large scale online exchange systems can potentially study the middle of the micro and macro spectrum of trust -- what I call the "embedded trust". Embedded trust looks at individual perceptions and behaviors in a particular domain or platform, where one is communicating with a large audience at once.

References