Commentary - (2021) Volume 8, Issue 2

The Individual-level sources of eroding social trust in the United States were investigated in a longitudinal research

Ioana R*
*Correspondence: Ioana R, Banat University of Agricultural Sciences and Veterinary Medicine, Romania, Email:

Author info »




social science, shipper dispatching


In recent decades, the United States has seen a significant fall in social trust. Surprisingly few research have looked into whether individual-level reasons can account for this drop. We examine the effects of four possible individual-level sources of changes in social trust: job loss, social links, income, and faith in political institutions, using three-wave panel data from the General Social Survey (2006–2014). Fixed-effects linear regression models imply that all factors except social relationships are important. Then, using GSS data from 1973 to 2018, we estimate trust based on observed unemployment, confidence in institutions, and income satisfaction, versus a counterfactual scenario in which those three predictors' values are maintained constant at their mean levels in the early 1970s. The predicted values in these two scenarios differ significantly, implying that a decrease in institutional trust and an increase in unemployment scarring may account for nearly half of the observed reduction in social trust in the United States.


The United States has seen a worrying fall in social trust over the previous several decades (Clark, 2015; Clark and Eisenstein, 2013; Paxton, 1999; Putnam, 1995a, 1995b; Rahn and Transue, 1998; Uslaner, 2002; Wilkes, 2011). When questioned in 1973 whether “most people can be trusted or you can't be too careful when dealing with others,” 46% of respondents said they could trust most people, according to data from the US General Social Survey (GSS). By 2018, the percentage had dropped to 32%. While other countries, namely those characterised as liberal welfare states, have experienced occasional declines in social trust (Hall, 1999; Larsen, 2007), the United States' prolonged downward trend stands out.

In uncertain conditions, social trust encourages collaboration between strangers, lowering transaction costs (Rothstein and Stolle, 2008). As a result, it is a valuable, if not indispensable, moral resource for resolving collective action issues and social dilemmas. Only the latter viewpoint would demand social initiatives targeted at reducing the existing level of distrust in the United States. Our research is guided by the latter, so-called "experiential" perspective on social trust (Dinesen and Bekkers, 2017). However, all of these studies I use macro- or meso-level reasons for the loss in social trust in America rather than individual-level explanations, and ii) depend solely on cross-sectional data, that is, data from respondents who were only questioned at one moment in time. This raises several unsolved questions. Unemployed Americans, for example, exhibit lower levels of social trust than employed Americans, according to Brehm and Rahn (1997).


Several Individual data from two independent US General Social Survey surveys were used (Smith et al., 2020). Both data files are open to the public and are available for download. An independent, complete probability sample of English- and Spanish-speaking people living in non-institutional arrangements or homes in the United States was recruited for each panel study. Each of the three GSS panel studies randomly chose 2000 cases for two re-interviews over the period of four years from the initial GSS surveys in 2006, 2008, and 2010. In 2006–2010, each GSS began a new four-year/three-wave panel. In the third and final waves, only respondents who took part in the corresponding second waves were re-interviewed. There were no geographical restrictions on re-interviews (NORC, 2017). GSS response rates (at the baseline of each rotating panel study) ranged from 70% in 2010 to 71% in 2006. Our pooled 2006–2014 sample contains n = 2340 respondents due to the focus on within-change, which necessitates at least two repeated measures from each respondent, and the GSS's split-ballot design.

In the second part of the analysis, we looked at whether there were any trends in those predictors of trust that corresponded to the long-term decline in US social trust, as well as how social trust would have developed if GSS respondents had reported 1973 mean values for unemployment, satisfaction with income, and confidence in political institutions over time. We used data from the GSS 1973–2018 cumulative file for this (Smith et al., 2020). Since 1972, the GSS has been conducted annually or biennially, and it provides the most comprehensive, nationally representative picture of social trust changes in the United States. Between 2014 and 1993, response rates ranged from 60% to 82 percent. We excluded the years 1974, 1977, 1982, and 1985 since the standard trust measure was not included in those polls. Because the 1972 sample lacked information on unemployment experiences and trust in political institutions, it was eliminated. We additionally deleted cases from the black oversamples to estimate nationally representative year-specific averages for trust and its potential predictors, as recommended on the GSS homepage. Our GSS 1973–2018 dataset had n = 36,533 cases after listwise deletion of missing instances.


To see how social trust responds to changes in individual-level factors across time, we used fixed-effects linear regression models (Allison, 2009; Vaisey and Miles, 2017). These models look at relationships between people over time, thus they can't be skewed by omitted variable bias caused by unobserved time-invariant variables.Race (Wilkes, 2011), upbringing (Uslaner, 2008), and genetic characteristics are examples of such variables (Sturgis et al., 2010). Within-individual variance is limited since each of the three GSS panel studies conducted between 2006 and 2014 had a maximum of three panel waves. As a result, we followed Hout and Hastings (2016) and combined the GSS panel data from 2006 to 2014 into a single datafile. We added dummies for the survey years 2008, 2010 and 2012 to account for the possibility of period effects between 2006 and 2014 (reference year: 2006; the 2014 dummy is removed because this year is specified by the combination of panel wave = 3 and year = 0).

Author Info

Ioana R*
1Banat University of Agricultural Sciences and Veterinary Medicine, Cluj-Napoca str. Calea Manastur no. 3-5, Cluj-Napoca, Romania

Published Date: Jun 30, 2021

Copyright: This is an open access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.

Get the App