A MegaAnalysis of Trust Global Trust Research Consortium
A Mega-Analysis of Trust Global Trust Research Consortium
Fundamental Questions Are we losing faith in each other? How does trust develop over the life cycle? How do generations differ in trust? Why are citizens in some countries more trusting than in other countries? • How does trust affect health, income, wellbeing? • •
Yet we work like…
Beta blockers
Benefits The benefits of harmonizing and pooling research databases are numerous. Integrating harmonized data from different populations allows achieving sample sizes that could not be obtained with individual studies, improves the generalizability of results, helps ensure the validity of comparative research, encourages more efficient secondary usage of existing data, and provides opportunities for collaborative and multi-centre research.
Comparable projects • Luxemburg Income Study [LIS] • International Stratification and Mobility File [ISMF] in Sociology • Cross-national Survey Data Harmonization [SDH] Project • Durand et al. on political trust
Ex Post Survey Data Harmonization A process: (a) in which different survey datasets that were not specifically designed to be compared are pooled and adjusted (i. e. recoded, rescaled, or transformed) to create a new integrated dataset that could be analyzed as a typical single-source dataset; and (b) that is based on clear criteria that specify which datasets are included into the new dataset and clear methods for how variables Dubrow & Tomescu-Dubrow, in the new dataset are created. 2014
Meta vs Mega-analysis • Meta-analysis also allows scholars to analyze the collective evidence on a certain phenomenon • But meta is only possible on released reports, and susceptible to publication bias • Power is limited to the #studies Meta-analysis of Individual Patient Data (IPD) = Mega-analysis
Pp. 77 -100 in: Van Lange, P. A. M. , Rockenbach, B. , & Yamagishi, T. (Eds. ). Trust in Social Dilemmas. Series in Human Cooperation, Volume 2. Oxford: Oxford University Press. https: //osf. io/umdxg/
Global Trust Research Consortium • Open Science Framework: https: //osf. io/qfv 76/ • Current members: René Bekkers, Arjen de Wit, Tom van der Meer, Eric Uslaner, Zhongsheng Wu, Bart Sandberg Please join us! You are most welcome
Surveys currently included • Multinational: ISSP, WVS, ESS, EQLS, Eurobarometer, Survey of Adult Skills (PIAAC), CID • National: German General Social Survey, BHPS / Und. Soc • Rough estimate: these surveys include about 1/3 of all trust responses ever collected • We have identified ~120 surveys since 1953 that have included variants of the trust question
Varieties of trust • Would you say… In general most people can be trusted? OR: You can’t be too careful in dealing with other people? – Forced choice format (0 – 1), the Rosenberg original (1953) – With option ‘It depends’ offered – With option ‘Don’t know’ added • These poles as Likert items (1 -5, 1 -7, 1 -10, 0 -10) • Other statements about human nature (1 -5)
Yay, we have variance! • We can leverage the pecularities of surveys as natural experiments • Use item, survey, and data quality characteristics as covariates • And add interactions with substantial correlates of trust
Predictors at 5 levels 1. 2. 3. 4. 5. Country Time Survey Item Individual • • • 88 1981 -2014 24 5 1, 237, 870
Potential Methods Effects • Question order: before / after questions that generate a ‘warm glow’ • Response category format: 0 -1, 1 -5, 1 -7, 1 -10, 0 -10 • Mode of data collection: face-to-face, paper -and-pencil, online • Data quality: response rate, #missings, interviewer ratings of ‘cooperativeness’
POWER! • We should collect as many country – year observations as possible, from as many different surveys as possible • To disentangle various methods effects • To answer questions on age, cohort and period effects on trust • To detect relationships at minuscule effect sizes
Procedure 1. Identify a survey not yet included 2. Categorize the methodology: trust measure, data collection mode 3. Provide code for harmonization 4. Add data 5. See results 6. Analyze data
Response categories 70% Trusting response 60% 64. 38% 58. 48% 58. 03% 53. 06% 52. 77% 50% 40% 30% 26. 39% 28. 49% 0 -1 31. 67% 1 -5 24. 37% 20% 10% 0% all CH UK SE others 1 -11
Survey mode 70 58. 5 60 54. 46 50 40 30 25. 47 27. 17 20 10 0 forced choice (n = 94, 844) others around scale (n = 43, 262) alone with interviewer Note: with this n, everything is significant
Age + Cohort 0. 6 0. 5 1925 1935 0. 4 1945 1955 0. 3 1965 1975 0. 2 1985 1995 0. 1 0 2005 10 20 30 40 50 60 70 80 90
And now • What would be good questions to answer? • Do you know of any surveys that we may not know of? • Would you be willing to add these surveys?
Let’s collaborate. René Bekkers @renebekkers r. bekkers@vu. n l This project is on the Open Science Framework, https: //osf. io/qfv 76/
- Slides: 32