What holds us together? Measuring dimensions of social cohesion in Canada

1.Introduction

Social cohesion generally refers to social connectedness, solidarity and trust amongst individuals, within and across communities and organizations, and within society at large. While there is no single definition of the concept, it is generally understood to measure the strength of the bonds between societal members – or, alternatively, the weakness of these bonds.

Social cohesion acts as a valuable barometer of the fabric or ‘glue’ that connects members of society. Identifying low cohesiveness or social fragmentation, which carry negative societal consequences, has garnered substantial attention in the last two decades. The erosion of social ties [1, 2], lower trust in institutions [3, 4], higher inequality and reduced social mobility [3, 5], among other occurrences, all have the potential to challenge individuals’ unity and solidarity.

Given social cohesion’s role in holding societal members together, the concept can provide a useful analytical lens to study emerging social trends. In contrast, if not concisely defined, the concept of social cohesion can serve as an all-encompassing term that obscures crucial differences in cohesiveness across dimensions and subpopulations. Delimiting the concept’s scope using survey data and its implications across different population groups are key contributions of this article.

This study, expanding upon prior work within the context of an in-depth review on social cohesion for the United Nations Economic Commission for Europe [6], empirically identifies distinct dimensions of social cohesion. In this context, a dimension refers to a constituent part of social cohesion. The article addresses two main research questions: (1) What dimensions of social cohesion can be empirically identified using Canadian social surveys? (2) What household characteristics correlate with higher or lower social cohesion, and how does this vary across identified dimensions of social cohesion?

This study is divided into six additional sections. The first provides an overview of seminal scholarship on social cohesion. The next section explores existing frameworks and the characteristics of contemporary definitions of social cohesion. The third additional section covers details about the social survey data used in this study. The subsequent section identifies key dimensions of social cohesion using factor analysis. The following section uses latent class modelling (LCM) to study groupings of individuals with varying levels of social cohesion and the disaggregation of constituent member characteristics. Finally, the last section provides concluding remarks and discussion points for future research on social cohesion.

2.Broadly defining social cohesion

Social cohesion broadly refers to the strength of bonds or social distance between societal members. The concept of social cohesion has evolved since early sociological and psychological scholarship.

The concept can be traced back to Émile Durkheim, one of the architects of modern sociology [7]. Durkheim defined social cohesion as the interdependence or solidarity between individuals with strong social bonds and without social conflict [8]. Similarly, the seminal work of Tönnies contrasting “community” (as defined by shared physical space, beliefs and norms) and “society” (based on self-interest for membership) lays foundational principles for studying cohesion within communities [9]. Building on these concepts in the 1950s, Talcott Parsons “argued for a functionalist approach, which could treat society as a system, composed of interdependent subsystems, held together by shared values reproduced by socialisation” [10].

In psychoanalysis and social psychology, Sigmund Freud studied emotional ties, or emotional commitment, as the phenomenon whereby individuals that share common characteristics form emotional ties [8]. Building on similar notions, Lott and Lott defined social cohesion as reciprocal positive attitudes among individuals of a group [11].

In contrast, Allport theorized that common characteristics were not only unnecessary, but that intergroup contact could foster cohesion [12]. His contact hypothesis posited that intergroup contact reduces prejudice and fosters co-operation, provided conditions such as equal status and shared goals. The notion that social cohesion is not dependent on homogeneity across individuals is echoed in the current understanding of cohesiveness, which emphasizes homogeneity in values, trust, sense of belonging, social participation, socio-economic opportunities, social mobility, and social capital [3, 8, 10, 13, 14, 15, 16, 17, 18, 19, 20, 21, 22] over the homogeneity of individual characteristics.

The past two decades of research on social cohesion mark a departure from earlier writings in terms of its increasingly pluralistic approach, instead focusing on defining cohesion through measurable facets, or dimensions. In particular, policy research on social cohesion has been highly influential in its attempts to explicitly define social cohesion through select dimensions that are empirically measurable [15], specifically focusing on socio-economic disparities and social exclusion [3, 13, 14, 21]. For example, the Council of Europe’s approach to social cohesion emphasizes a society’s ability to ensure individuals’ well-being, minimize disparities and avoid marginalization [23].

Overall, a plurality of definitions and applications of the concept of social cohesion has been advanced by international organizations, national governments and academic researchers. The next section focuses on more contemporary definitions of social cohesion through a list of constituent parts or measurable dimensions [15, 18].

3.Contemporary analysis of social cohesion

The concept of social cohesion has evolved since early sociological and social psychology scholarship on the topic. As discussed in the previous section, the contemporary literature is more pluralistic than earlier scholarship, focusing on the various facets of social cohesion being defined across multiple dimensions that are easier to empirically measure. In this context, a dimension refers to a constituent part of social cohesion. Moreover, social cohesion is a multi-level concept, meaning it varies across individuals, but also various aggregation of individuals (e.g., groups, communities, institutions, regions). Finally, beyond differences in academic studies of social cohesion, various national approaches to social cohesion warrant mention to explore lessons learnt across regional and temporal contexts.

4.Data

Statistics Canada’s 2020 GSS on the theme of Social Identity (Cycle 35) covers data pertaining to several dimensions outlined in the literature on social cohesion. The 2020 GSS included an oversample of selected population groups identified in Canada’s Employment Equity Act.22 Interviews were conducted using computer-assisted telephone interviews and electronic questionnaires, with an overall response rate of 40.3%.

The context in which the survey was administered is vital to interpret results appropriately. The 2020 GSS data – collected between August 2020 and February 2021 – should be interpreted within the context of the COVID-19 pandemic. Given that several social cohesion dimensions were likely impacted by social distancing, isolation, and other pandemic-related health measures, the results may mirror those of other social and well-being studies whereby vulnerable individuals (e.g., younger individuals) were disproportionately affected during this period [41, 42, 43].

This paper’s focus is on the social dimensions of cohesiveness rather than socio-economic dimensions such as economic inequality, poverty, and social mobility [3]. Of particular interest are those 2020 GSS variables that fall under the broad umbrella of social cohesion as they pertain to the dimensions of legitimacy, belonging, social capital, social participation, and shared values. While including socio-economic dimensions would be valuable, no other survey or combination of data sources provided both socio-economic variables and the detailed social dimensions captured within the 2020 GSS.

At the outset, 60 2020 GSS variables that fell under the umbrella of social cohesion were considered in factor analysis and latent class modelling. Using factor analysis, this number reduces to 43 variables with high response rates and load strongly onto the exploratory factor model. The following section describes this process in greater detail.

5.Factor analysis of social cohesion dimensions

A multi-dimensional approach is key to capturing the many facets of social cohesion. While there are benefits to the analytical depth permissible when narrowly studying a single dimension of social cohesion, this first section focuses on the broader identification of dimensions and their interactions. Therefore, the emphasis is on concurrently analyzing dimensions to better understand the relationships between these social cohesion dimensions.

This however poses a challenge. As the number of dependent social cohesion variables increases, the number of correlations within the set of social cohesion variables and their relationships to other variables of interest (e.g., socio-demographic variables) increases at a faster rate. Analytical methods employing dimension reduction are therefore necessary to balance the parsimony and interpretability of results with the intrinsically multi-dimensional aspect of the research.

Factor analysis is a statistical method that is well suited for this task. Factor analysis models observed and correlated variables in a dataset in terms of a lower number of unobserved variables called factors. Each variable in the data is modelled as a linear combination of unobserved factors plus an error term, with coefficients referred to as factor loadings. Factor loadings describe the magnitude of the relationship between each factor to the observed variable.

More formally, factor analysis aims to model the overall variability in a dataset by identifying groups of variables whereby the correlation of variables in that group is higher than with out-group variables, and postulates that each of these groups can be described with a single latent variable, the corresponding factor, without substantial loss of information. In this way, factor models can substantially reduce the number of variables in a dataset while preserving the variability across observations and has the potential to produce meaningful factors that can be understood in terms of the observed variables.

For each variable xs in the data with mean μs, the factor model assumes that xs can be represented by:

Where xs-μs is the mean-adjusted observed vector; li,s is a scalar coefficient (called the factor loading of variable s onto factor i); fi is the ith factor; and εs the corresponding error term. The number of factors k is determined exogenously, and can be identified with any number of criteria, often utilizing the variance-covariance matrix of the data itself. The derivation of the matrix of factor loadings li,s is usually of interest, as they represent the strength of the relationship between each observed variable and each unobserved factor. Factor loadings with magnitude less than 0.3 are typically treated as extraneous to that factor, and variables with a factor loading above this threshold for a given factor are said to load onto that factor.

In practice, conducting a factor analysis typically involves two steps. First, an exploratory factor analysis (EFA) is conducted on the data and aims to generate an initial factor model, identifying the appropriate set of variables to include, the correct number of factors, and identifying which variables correspond to which factors (via factor loadings). Second, a theoretical model is constructed based on the results of the EFA (taking into consideration extraneous factor loadings) and subjected to confirmatory factor analysis (CFA), which measures the goodness of fit of the hypothesized model to the data.

Starting with the complete set of 60 variables identified as relevant for this analysis, an iterative EFA was carried out in three stages. First, the appropriate number of factors was determined using the Kaiser-Guttman criterion [44, 45] on the polychoric correlation matrix of the data. Second, the correlation matrix was used to calculate a factor model, yielding both the matrix of factor loadings and a vector of uniqueness measures (corresponding to the amount of variance unexplained by the factor model). Third, variables were dropped from the model if they exhibited a high proportion of non-responses, if their uniqueness was higher than 0.8 (i.e., the model explained less than 20% of the variation in the variable), or if no factor loadings exceeded the 0.3 threshold. This iterative process was considered complete when no remaining variables were eliminated based on these criteria.33

Table 1 presents the factor loadings for the resulting exploratory factor model. The iterative process described above results in 17 of the 60 variables being removed from the analysis, and yields a nine-factor model that explains 64.3% of the total variance across the remaining 43 variables. Factor loadings presented in Table 1 quantify the magnitude of the relationship between each variable and the corresponding factor, and range from -1 to 1. Loadings close to 1 or -1 for a variable-factor pair indicate strong positive/negative loading onto that factor, and loadings close to 0 (between 0.3 and -0.3) are suppressed and considered extraneous. Factors are presented in order of decreasing proportional variance, with Factor 1 explaining the greatest proportion across all variables.

The resulting factors, though referred to as dimensions for simplicity and ease of presentation, could instead constitute antecedents or other correlates of social cohesion as several studies have pointed out [15, 27, 28]. As such, the factors are not intended to be a prescriptive list of social cohesion dimensions.

Confidence in institutions (Factor 1), explaining the greatest share of variance overall, suggests that it is a key dimension of social cohesion. Confidence in Canada’s federal parliament, the justice system and courts, schools, banks and major corporations ranged from 0.70 to 0.83. This aligns with prior research outlining confidence in institutions as a dimension of social cohesion [10, 13, 16, 17, 18, 30], although prior research occasionally regroups common values (Factor 7) and common civic culture (Factor 1) [19, 20]. Factor loadings for confidence in the Canadian media and police were slightly lower, at around 0.67, while the factor loading on confidence in local merchants and business people was relatively low (0.47). Pride in diversity and in the way democracy works in Canada also loaded onto confidence in institutions (Factor 1), albeit at weak levels (0.41 and 0.47, respectively), with pride in democracy weakly loading onto geographic sense of belonging as well (Factor 5).

The factor explaining the second highest share of variance is trust, specifically trust in so-called “out group” members such as strangers and people with different linguistic, religious or ethnic backgrounds. This dimension resembles dimensions of openness [18] and interpersonal trust [15, 18, 30, 47] in other studies of social cohesion. However, it focuses more specifically on affective sentiments towards individuals outside one’s immediate entourage. Questions pertaining to trust of people with the same or different ethnicity, religion and language loaded strongly onto Factor 2, each with a factor loading of 0.9 or higher. The trust of strangers and trust of neighbours loaded onto this factor as well, but with values of 0.55 and 0.61, respectively

The factor model splits experiences of discrimination into two distinct groups each. The division of experiences of discrimination between Discrimination A (Factor 3 pertaining to elements of age, sex, physical appearance, disability, sexual identity and gender identity) and Discrimination B (Factor 6 pertaining to ethnicity, skin colour, language and religion) is noteworthy, and suggests that the model identifies experiences in these two groups as being distinct. Experiences of discrimination or unfair treatment on the basis of age, sex, physical appearance, disability, sexual identity and gender identity loaded onto Factor 3, with factor loadings generally ranging from 0.61 to 0.77. Both discrimination groups are comparable dimensions to recognition, equality and inclusion in the social cohesion literature [10, 13, 16].

Electoral participation (Factor 4) serves as a narrower measure of social participation [14, 17, 30], which several researchers deem an important dimension of social cohesion and a source of social interconnectedness [16, 18, 46]. However, in contrast with certain authors, electoral participation does not load onto civic culture (i.e., confidence in institutions) and shared values [19, 20]. While group membership variables were excluded based on the exclusion criteria of this study, questions pertaining to participation in the prior federal, provincial and municipal elections all have factor loadings above 0.9. Respondents’ intentions to vote in the next federal election also loaded on Factor 4, with a factor loading of 0.7.

Sense of belonging, like discrimination, is similarly split into two distinct factors: geographic or local belonging (Factor 5) and identity-based belonging (Factor 8). Although often grouped under a broader sense of sense of belonging, this split aligns with prior research [19, 20]. Factor loadings were stronger for variables pertaining to the local level, that is, sense of belonging to the municipality (0.88) and one’s local community (0.78), than for variables pertaining to broader entities, specifically, their province (0.72) and to Canada (0.58). In contrast, sense of belonging to identity groups (Factor 8), especially along ethno-cultural, religious or linguistic lines, have loadings ranging from 0.74 to 0.83. While past research has typically not split the sense of belonging into two groups, sense of belonging is a well-documented dimension of social cohesion [10, 13, 14, 15].

Shared values (Factor 7) is another identified social cohesion dimension [14, 19, 20]. Based on 2020 GSS responses about agreement with values such as gender equality, respect for Indigenous culture, ethnic and cultural diversity, and human rights, ‘value distance’ is measured as the difference between their personal views and their perceived societal view. This captures the degree to which individuals felt their views were congruent or incongruent with the broader Canadian population. These four ‘relative issue importance,’ or ‘shared values,’ variables load onto Factor 7, with factor loadings on relative importance attached to respect for Indigenous culture and ethnic and cultural diversity of 0.8 or more.

The ninth factor identified pertains to peoples’ social ties and assessments of their neighbours and neighbourhoods. Questions regarding social contacts, helpfulness and reciprocity with neighbours load onto this factor, with factor loading ranging from 0.63 to 0.8. These variables align with previous research tying social cohesion to social capital and social ties [3, 18].

Building on the EFA’s intuitive results, a CFA validates the exploratory results by assigning each group of variables to a dedicated factor. Appendix Table 1A presents the factor scores produced by the confirmatory factor model. It shows the final structure assigned to the model. The CFA indicates that this model is a good fit to the data44 and produces factor loadings similar to those produced in the exploratory factor model, except for pride variables.55 This final set of factor loadings and the polychoric correlation matrix are used to calculate factor scores for each respondent across the nine factors. Although some variables load onto more than one factor with loadings exceeding 0.3, these loadings are relatively small (i.e., under 0.35) and are not identified in the CFA model for brevity. Empty cells indicate that a variable is not assigned to that particular factor.

6.Latent class analysis

Factor models are a useful tool for understanding the underlying structure of variability in a dataset. In the previous section, factor analysis identified nine factors of social cohesion from the 2020 GSS where each factor can be thought of as condensing a set of highly correlated variables into a single representative factor. This is useful for the purposes of understanding social cohesion in a multi-dimensional framework but is restricted to identifying patterns across variables and not across observations.

As this study is interested in socio-demographic and household characteristics that may be coincident to varying levels of social cohesion, LCM is also employed to complement the factor analysis conducted above. Latent class models are a subset of structural equation models that relate observations in a dataset to a set of underlying latent classes. Whereas factor analysis models each variable as a linear combination of the underlying factors, latent class models identify patterns between variables and observations via maximum likelihood estimation (MLE) of a non-parametrized model.

In the resulting classification the manifest variables are assumed to be statistically independent conditional on class membership so that observable patterns and correlations within a class are understood as resulting from their class membership. As the model is estimated using MLE, classification of observations in the data is probabilistic so that each observation is assigned a vector of probabilities of class membership to each of the g classes. LCM treats patterns in manifest variables as being “symptomatic” of membership to one of g latent classes with the number of classes g being determined exogenously.

More formally, the probability that an observation i with responses yi belongs to class g is modelled by:

where k∈K indexes the manifest variables; o∈Ok indexes the possible values for manifest variable k;po,k,g is the probability that a randomly selected observation has the value o in variable k in class g; and δi,k,o is the Dirac delta function equal to 1 when observation i has value o in variable k and 0 otherwise. Taking the sum of this product over the classes g∈G yields:

with the parameters of interest p,γ being derived via likelihood maximization.

Latent class modelling has several advantages over data clustering methods, which although similar to LCM differ in several important ways. LCM is non-parametric and estimates membership probabilistically using MLE, and for small number of classes (g< 9) were found to yield consistent results and achieve the same maximum likelihood for the estimator. Cluster algorithms in contrast are initialized with a random cluster specification and produce results that are dependent on initial conditions, requiring a larger number of trials to identify the best fit overall.

Discrete data cluster models such as the k-modes algorithm are also limited in that defining distance between observations and group centroids can be ambiguous for categorical data, although can be meaningful on binary data [48]. As latent class modelling does not identify group membership based on a distance metric, it can handle any number of discrete variables with many responses with relative ease, and is hence well suited for survey data

Latent class models however are limited in that they assign a discrete probability distribution of class membership to each observation rather than a single class, making the resulting classes “fuzzy”.66 The non-parametric nature of the model also means that the class determination process is a so-called “black box” in terms of retracing the role of each variable in determining class membership.

Using the same 43 observed variables that loaded strongly in the factor model in the previous section combined with a selection of socio-demographic and household variables of interest, a variety of different model specifications with different numbers of classes were produced. In keeping with best practices [49], the number of classes was identified on a model without the inclusion of socio-demographic covariates to avoid model misspecification. Once the appropriate number of classes was identified, covariates were re-incorporated so that latent classes could be identified along those features as well.

A 3-class solution was identified as yielding latent classes that easy to interpret and concise. Akaike and Bayesian information criteria both yielded lower values for a higher number of latent classes indicating a better fit of the model. However, it was observed that observations of a given class in the 3-class model tended to be sorted into similar groups of classes as a cohort when more classes were added to the model. These groupings could be identified for the 4 and 5-class models and did not necessarily yield a more intuitive explanation despite better information criteria. The 3-class model hence strikes a balance between objective goodness of fit measures and subjective interpretability.

The 3-class model yields classes which roughly correspond to a class exhibiting signs of lower cohesion (Class 1); a class with high confidence in institutions and high sense of belonging (Class 2); and a class with high trust in others and good neighbourhood connections (Class 3). The addition or omission of socio-demographic variables has little effect on the makeup and composition of the resulting classes. The labels “Low”, “Confidence-Belonging”, and “Trust-Participation” are given to these classes for descriptive purposes only and are not intended to be prescriptive.

Results from the model with socio-demographic variables included are presented here, with the “Low”, “Confidence-Belonging” and “Trust-Participation” cohesion classes, making ups 36.7%, 28.5% and 34.8% of survey respondents, respectively. Table 2 presents average factor scores by latent class, with negative values indicating low adherence to the corresponding concept (e.g., lower trust, fewer reported experiences of discrimination) and positive values conversely indicating higher adherence.

The “Low” cohesion class has respondents who report lower than average confidence in institutions, trust in others, electoral participation, geographic sense of belonging, and neighbourhood ties. This group also sees close to average experiences of Discrimination B, value differences and identity sense of belonging, and sees fewer experiences on average with Discrimination A than both other groups.77

The “Confidence-Belonging” class consists of individuals who report the highest confidence in institutions and sense of belonging both for geographic belonging and identity belonging. Individuals in this group report close to average trust in others and neighbourhood ties, and the lowest scores for both Discrimination A and B (indicating more experiences with both), election participation, and value differences.

The “Trust-Participation” cohesion class consists of individuals who report higher than average values for trust in others, electoral participation, Discrimination B (indicating fewer experiences), value differences and neighbourhood ties. This third group has confidence in institutions and Discrimination A values that are in between those of the “Low” and “Confidence-Belonging” cohesion groups, and lower values for identity-based sense of belonging.

Turning now to the demographic distribution between groups in Table 3, the “Low” cohesion group has an overrepresentation of younger Canadians, with more than half of Canadians aged 15–35 being allotted to this group. Conversely, the “Confidence-Belonging” group has wide range of ages and the “Trust-Participation” group skews older, representing approximately 46% of respondents aged 55 or older.

The “Trust-Participation” group has a slightly higher representation of women than men (55% and 45% respectively), with these proportions roughly reversed for the “Low” cohesion and “Confidence-Belonging” groups.

The “Trust-Participation” consists of a larger proportion of respondents reporting annual incomes of $100,000 or more compared to the other two groups. In contrast, the educational compositions of each group were fairly similar, with the main exception of a larger proportion of college diplomas or University certificates within the “Trust-Participation” group. Another smaller exception is the larger proportion of respondents reporting having completed less than high school within the “Low” cohesion and “Confidence-Belonging” groups. These patterns are reflected in homeownership across the three groups, with the “Trust-Participation” group having the highest rates of homeownership at 87%, compared to 64% to 67% for the “Confidence-Belonging” and “Low” cohesion groups.

The “Confidence-Belonging” group is mostly comprised of first-generation immigrants to Canada, who account for slightly more than 8 in 10 individuals in that group. Comparatively, 57% of the “Low” cohesion group, and 21% of the “Trust-Participation” group, reported being a first-generation immigrant to Canada. As for other immigration status groups, both individuals reporting being second generation or third generation or higher are relatively underrepresented within the “Confidence-Belonging” group.

The “Low” cohesion group contains an underrepresentation of married respondents relative other groups and an overrepresentation of single person households, with most single respondents appearing in the “Low” cohesion category. Married households make up just shy of 60% of the other two groups, and 41% of the “Low” cohesion group.

The “Trust-Participation” group has an overrepresentation of respondents living either in smaller cities/towns or in rural areas. Respectively, 89% and 90% of the “Low” cohesion and “Confidence-Belonging” groups live in a Census Metropolitan Area (CMA), while a smaller proportion (63%) of the “Trust-Participation” group lives in a CMA. The majority of rural respondents and those living in a Census Agglomeration (CA) are categorized into this latter “Trust-Participation” group. Additionally, the “Trust-Participation” group has a noticeable overrepresentation of respondents in Atlantic provinces, with a majority of survey respondents from this region appearing in this group.

Respondents identifying as LGBTQ2+ are overrepresented in the low social cohesion group, making up around 7% of this group relative to roughly 3% for the other two groups. Among respondents identifying as LGBTQ2+, slightly over half (58%) appear in the “Low” cohesion group.

The “Trust-Participation” group has an underrepresentation of people not identifying as White, with 84% of respondents identifying as White compared to 43% for the “Low” cohesion group and 27% for the “Confidence-Belonging” group. Between the “Low” cohesion and “Confidence-Belonging” groups, more South Asian and West Asian (including Middle Eastern and Arab) respondents appear in the latter group. Individuals identifying as Chinese, East or Southeast Asian, Black, Latin American or Mixed represent relatively similar proportions of both groups.

Many of the patterns in responses to the social cohesion variables reflect their demographic compositions. The “Confidence-Belonging” group for instance is mostly comprised of first-generation immigrants to Canada, and exhibits low electoral participation, possibly due to a lack of registration or residency status. Higher reported experiences with discrimination falling under the category of Discrimination B in the “Low” and “Confidence-Belonging” cohesion groups also correspond to higher representation of population groups other than individuals identifying as White in these classes, with a similar pattern for Discrimination A and the overrepresentation of LGBTQ2+ individuals in the “Low” cohesion group.

Overall, LCM with three classes offers an intuitive way of understanding social cohesion in Canada. Moreover, it paints a picture that relates concepts of cohesion from the factor analysis to demographic groups in Canada, thereby illustrating how various cohesion concepts align along socio-demographic lines.

7.Concluding remarks and discussion

Social cohesion is a multi-dimensional concept referring to social connectedness and solidarity amongst individuals, their communities, and their representative institutions. Measuring social cohesion ensures Canadians are well-informed on trends regarding their interactions with fellow Canadians and organizations, or conversely their relative sense of societal detachment. Better understanding the constituent parts that contribute to or detract from social cohesion, as well as those identifying groups of Canadians lagging in their perceptions of connectedness to Canadian society, sheds light on social priorities moving forward.

Building on prior work using factor analysis and the 2020 GSS [6], nine dimensions of cohesiveness are identified. Subsequently, employing LCM, the study identifies three groups (“Low”, high “Confidence-Belonging” and high “Trust-Participation” cohesion groups) of individuals that share common traits and prioritize certain dimensions of social cohesion. Individual characteristics of each group provide insights into common social sorting mechanisms and the characteristics of those categorized in each group.

This article contributes to the literature on social cohesion as a stepping stone for future work. Two complementary paths forward can be identified: (1) enhancing data collection to capture additional variables and dimensions, and (2) disaggregating the analysis of specific dimensions of social cohesion and their variation over time.

Several limitations pertaining to the data and survey collection methods warrant further consideration. For instance, certain commonly cited social inclusion (e.g., socio-economic (in)equality, poverty) and social mobility [3, 5, 13, 16] dimensions are excluded from this study because the necessary variables are unavailable using the 2020 GSS. Measures of affective distance between individuals would also complement existing measures of trust in so-called “out group” members and a growing literature on polarization (and “social distance”) within societies [20, 50, 51, 52, 53]. Methodologically, low social survey response rates (and potential self-selection biases), survey mode effects (e.g., data collection by telephone versus online surveys impacting responses), survey design and alternative data sources are all priorities in the study and measure of social cohesion.

Future research should seek to delve deeper into the specific dimensions of social cohesion and variation over time. While analyzing overall social cohesion (i.e., as an aggregate) yields interesting results, it can also obscure variation across dimensions of cohesiveness and population groups. Moreover, while this research provides a high-level static portrait of social cohesion within Canada, ongoing research is needed to measure how social cohesion fluctuates over time. Such longitudinal approaches would ensure Canadians remain informed of variations and emerging trends in social cohesion moving forward.