Measuring and Modelling Trust

1 The growing literature on trust, social capital and well-being relies almost exclusively on a single survey measure of interpersonal trust: " Generally speaking, do you think that most people can be trusted, or that you can't be too careful in dealing with people? " Efforts at modelling response to this balanced question are becoming increasingly sophisticated. Notable recent examples include Alesina et al. (2000) on the individual and contextual determinants of trust and Glaeser et al. (1999) on the link between response to the question and actual behaviour in laboratory trust games. The measure underpins a cross-national growth industry, mainly through the World Values Survey (WVS). And international evidence from the WVS has been brought back home, as it were, to explain ethnic and regional differences in trust in the US (Rice and Feldman 1997). Only rarely, however, is any attention is paid to what response to the question actually means. Most critically, does the question elicit response-based perhaps on past experiences-that indicates persons' real expectations of others' trustworthiness? Or does it register a moral predisposition, a statement about how one should react to others? This paper holds up a mirror to the standard question by comparing it with other renderings. Some of the alternatives are, like the original wording, about trust as a general proposition. Our most pointed demonstration, however, involves comparison with questions about a specific trust situation, a lost wallet. These questions mimic a widely publicized field experiment (Knack 2001). Some of the comparison involves simple response distributions, joined to observations on items' face validity. The most telling comparisons deploy multivariate techniques to plumb the sources of response. Where response to the traditional, highly general indicator is powerfully shaped by cultural norms, response to the specific, wallet question is sensitive to context and life experience. That said, the traditional indicator still gauges how much respondents believe others to be trustworthy. In getting to this point, we also draw a lesson about Canada. Our sample and our questionnaire are rich in representation of Canada's ethnoreligious diversity, of its high rate of immigration, and of its group life. In line with hypotheses that seemed plausible but were hitherto unsupported empirically, " bridging " groups appear to be more highly correlated than " bonding " groups with generalized trust. We also reproduce Rice and Feldman's (1997) finding that civic attitudes of immigrants are highly correlated with attitudes in their …