What this research found
Half a century of the General Social Survey's best-known question — whether most people can be trusted — was analysed with the weight NORC currently recommends for pooled analysis of the cumulative file. Weighted trust fell from 44.80% in the 1970s to 28.70% in the 2020s, a slide of about 0.31 percentage points every year for fifty years. In the most recent survey year, education split Americans further apart than five decades of decline had: 9.59% of adults without a high-school credential expressed trust against 62.12% of those with a graduate degree.
- Weighted trust declined in every decade: 44.80% in the 1970s, 43.23% in the 1980s, 36.88% in the 1990s, 35.75% in the 2000s, 33.38% in the 2010s, and 28.70% in the 2020s. That is a loss of roughly 16 percentage points, about 36% of the original level, with the sharpest single-decade drop falling between the 1980s and 1990s.
- Regressing the 29 annual weighted percentages on survey year gives a slope of −0.314 percentage points per year (standard error 0.039, t = −7.98, p = 1.4 × 10⁻⁸, R² = 0.70). An independent individual-level weighted regression over all 40,456 valid respondents, with heteroskedasticity-robust standard errors, lands at −0.318 points per year (p = 1.5 × 10⁻⁵³).
- The trend is not an artefact of how the volunteered "depends" answer is coded. Counting it as non-trusting shifts the series down by 1.5 to 2.5 points but yields a nearly identical slope of −0.320 points per year; "depends" is only 4.7% of substantive responses.
- In 2022, trust ranged from 9.59% among adults with less than a high-school education to 62.12% among graduate-degree holders — a spread of more than 52 percentage points, larger than the entire fifty-year secular decline. Only the two highest degree categories sat above the 28.70% national average.
- The gradient is not strictly monotone: adults whose highest credential is an associate or junior-college degree reported 19.36% trust, below the 28.16% of high-school graduates. That category has the smallest sample (89 valid responses) and so the widest sampling uncertainty.
How it was done
The GSS 1972–2022 cumulative datafile (Release 3, 75,699 records) was pulled from NORC, checksum-verified, and restricted to survey years 1972–2022, leaving 72,390 records across 34 survey years. Every weight-like variable was audited before settling on the person-level post-stratification weight WTSSPS: the composite weight assumed in the original plan is absent from Release 3, and the legacy alternative is missing for 7,576 records inside the analysis window. Two trust indicators were built in parallel, one excluding the volunteered "depends" response from the denominator and one folding it in with the non-trusting, so no single coding choice could drive the result. The linear time trend was then estimated two independent ways — ordinary least squares on the 29 annual weighted percentages, and individual-level weighted least squares with robust standard errors — and the 2022 education breakdown used the same weighted-percentage formula within each of the five degree categories.
Data sources
- General Social Survey 1972–2022 cumulative datafile, Release 3, NORC at the University of Chicago — 72,390 records, 34 survey years, 40,456 valid trust responses
- Twenge, Campbell & Carter, Psychological Science 25:1914–1923 (2014) — trust and institutional confidence through 2012
- Mewes et al., Social Science Research 95:102537 (2021) — individual-level sources of declining US trust
- Putnam, Journal of Democracy 6:65–78 (1995) and Bowling Alone (2000)
Limitations
The work is descriptive and identifies no causal effects; the education gradient may partly reflect more trusting people pursuing more schooling rather than schooling building trust. The standard errors use robust rather than design-based estimation and so ignore the GSS clustering and stratification, and the 2020s reading rests on a single post-pandemic survey year.
How this research was produced
K-Dense Web planned and ran this sociology investigation end to end — gathering the sources, carrying out the analysis, producing the figures, and drafting the report. The full session transcript, including every intermediate step, is available to view.
