Research approach

A large-scale, cross-national survey about social cohesion and social networks has been administered in four European countries, which were selected considering the diversity in social cohesion regimes present in Europe (Dragolov et al. 2016; Green and Janmaat 2011). The selected countries are Hungary, the Netherlands, Poland, and Sweden. These countries vary in their levels of inequality, diversity, and polarisation, as well as in their norms and institutions (e.g., welfare state regimes; Esping-Andersen 1999; Ferrera and Rhodes 2000). Consequently, we can expect the countries to differ in how much they present cleavages in socialisation patterns. Also, each of the countries has excellent name statistics, which are needed for the chosen approach. For all countries, a nationally representative sample of the general population has been drawn of 1,500 people per country (so 6,000 in total). The survey center SigmaDos was contracted to administer the survey via computer-assisted personal interviews (CAPI) or computer-assisted video interviews (CAVI) of approximately an hour, and they worked with national survey centers in each country.

The questionnaire had five sections:

A. “About you – entry questions”: A few basic sociodemographic data (questions A1 to A5)  

Section A of the questionnaire kicks off with a few questions about the respondent’s gender, age category, country of birth, and nationality, which break the ice before getting into more complex questions, and ensures that later questions in the network part can be filtered by country of birth.  

B. “About society”: Ideational measures of social cohesion at the respondent level (questions B1 to B10_NE_PO) 

Section B measures ideational cohesion in an individualistic way (see Objective 4), like most national surveys have measured social cohesion so far. This section contained respondent-level questions about social and institutional trust, welfare and migration attitudes, feeling thermometers to selected religious and political groups, justifiability of norm violations, and perceptions of society. Some other measures of ideational cohesion have been included in sections C (perceived social support provision) and D (civic engagement, loneliness) for a better flow of the interview. 

C. “About your social relationships” - Social networks (questions C1 to C22) 

Section C is the core and most innovative part of the PATCHWORK survey, and also the most time‑consuming for respondents. It focuses on measuring relational cohesion—the extent to which societies are connected through broad acquaintanceship networks rather than only through close personal ties.

The section begins with a series of questions that generate aggregated relational data (ARD). These are questions of the form “How many people do you know who [have a given characteristic]?” When combined with national statistics on the prevalence of those characteristics, ARD can be used—via the Network Scale‑Up Method (NSUM)—to estimate the size and composition of respondents’ acquaintanceship networks.

The intuition behind NSUM is straightforward. If, for example, a respondent reports knowing two people with a trait that characterizes 0.3 % of the population, the respondent’s total acquaintanceship network is estimated at approximately 2 / 0.003 = 667 people. In practice, estimates become more precise as multiple traits are used and as known sources of bias are controlled.

Building on these ARD estimates, the survey then follows up on a subset of traits for which respondents reported knowing at least one person, using name interpreters. For a maximum of 20 selected alters, respondents are asked about key attributes such as gender, perceived country of birth, religion, occupational status and occupation, and political orientation. Additional questions capture relational characteristics, including the type of relationship, emotional closeness, trust, and geographical proximity, as well as selected alter–alter connections.

Section C also includes measures of core personal networks, focusing on respondents’ strongest relationships. Investigating these networks has a much longer tradition. These data serve as an important anchor for personal network research and allow comparison between close ties and broader acquaintanceship networks. Finally, a standard scale measuring social support provision (SPS‑5) is included in this section.

Due to its combination of ARD, name interpreters, and core network measures, Section C is also the most complex part of the questionnaire in terms of routing and filtering logic.

D. “About you”: Further sociodemographic and civic engagement data (questions D1 to D36_a) 

The questionnaire finishes with more traditional sociodemographic data (e.g., respondents’ region of residence, civil status, number of children, education, employment) and other questions that respondents can more easily answer even if already a bit tired, such as data on individuals’ religion, political orientation, occupation, income, social media use (e.g., Facebook network size, frequency of social media use), civic engagement, experienced discrimination, perceived health, and loneliness. In this last part, respondents who had not yet reached 55 minutes of interview were also asked if they were willing to participate in a follow-up interview. 

E. Interviewer feedback on the interview (E1 to E5 for Hungary; E2 and E4 for the other survey countries) 

Once the questionnaire was terminated, interviewers filled in a question about the perceived honesty of respondents, and in Hungary some other questions about understanding were asked. This question was asked to filter out answers that may be less accurate.

The data are currently being analyzed statistically and they are also used to simulate society-wide networks of the populations based on the survey statistics. For this aim, we construct a population that resembles the national population (but scaled down), assigning attributes to the nodes (gender, country of birth, social class, religion, and political orientation) in the proportions and with the degree of intersectionality present in the population, according to population statistics. We then specify a graph generating model (based on an Exponential Random Graph Model; see Leskovec et al. 2010; Smith and Burow 2018) to generate a network structure consistent with what we observed in the survey, i.e., based on the degree distributions, the network heterogeneity and structural parameters found in the survey. With this model, the network structure can be simulated, maintaining only the models that have a good level of fit with the parameters extracted from the survey (cf. Leskovec et al. 2010; Smith and Burow 2018). By running this simulation many times, we explore the variation in macro-level structures consistent with the parameters to evaluate macro-level patterns in more detail.

Do you participate in the survey?

Check out our page "Are you a participant" to understand better what participation entails.

We also use the data about participants' networks to examine the causal mechanisms that relate the network structures with the subjective manifestations of cohesion. In this case, we use a method called agent-based modelling and simulation, that give a dynamic modeling of these networks, using the data and theories. We will, at least initially, maintain the network topology constant (a realistic assumption for acquaintanceship networks in the medium time range). Relationship attributes can be dynamic, such as tie strength and knowledge of network members’ attributes. Based on our theoretical model, our survey estimates for associations, and potentially the results from the qualitative interviews regarding temporal processes, cognitive perceptions, and settings, we can study how the network configurations are associated with subjective manifestations of cohesion. We can validate the models with the survey data regarding tolerance, trust, and policy preferences (which have not been used for parametrisation). Note that we can formulate a more classical social influence model for political orientation. If the models reach a good fit, we can alter parameter values to explore their impact on the outcomes further. We can incorporate theoretical propositions for which we do not have data in the survey, for instance, regarding the time spent in different social settings and whom people meet there (Gershuny and Sullivan 2019), to model social influence.