Introduction

In the first months of the COVID-19 pandemic (February to December 2020), public health measures such as lockdowns were the only means societies had at their disposal to try to control the surge in mortality. Toward the end of 2020, it was becoming clear that a vaccine would be available soon and the main question became that of the time at which it would be made available. Less often discussed, however, was another phenomenon already emerging at that time, vaccine hesitancy: some segments of the population were not willing to get vaccinated and protected against a deadly virus (MacDonald et al., 2020). Griffith et al. (2021) found that only about 75% of people in Canada planned to receive one of the vaccines at the end of 2020. The anti-vaccine opinion was not new, of course, but its magnitude and the fact that it manifested after such a massive health scare prompted numerous studies to try to understand its origin.

Most of the literature on the determinants of vaccine hesitancy is concerned with belief systems about health or vaccines (e.g., Larson et al., 2014; Troiano & Nardi, 2021). We summarize the main determinants identified in the literature in Table 1.

Table 1 Main determinants of vaccine hesitancy identified in the literature

In this study, we posit that social capital could explain some of the inter-individual variation in vaccine hesitancy. This has been suggested by studies of vaccine hesitancy showing that trust in the government or healthcare workers was a determinant of the willingness to take the vaccine (Biswas et al., 2021; Mesch & Schwirian, 2015; Steinert et al., 2022). Using data collected in Ontario from September to December 2020, before any COVID-19 vaccine was made available in that province, we test whether individuals with higher levels of social capital were also those with a higher willingness to get the vaccine when such a vaccine would be ready.

Coleman (1988) defines social capital as a productive resource that is located “in the structure of relations between actors and among actors.” The measurement of social capital is multi-dimensional, encompassing cognitive (what people feel) as well as structural (what people do) elements (Xue et al., 2020). Cognitive social capital includes social trust, the perception of social support, the perception of social cohesion, perceived reciprocity, sense of belonging, and loneliness. Structural social capital includes participation in some form of social activity, networks of personal relationships, social support, social engagement, volunteering, group membership, social integration, and social relationship (Xue et al., 2020). The idea that communities with higher levels of social capital are better positioned to fight epidemics has been suggested by Pitas and Ehmer (2020), who hypothesized that communities with more social capital would do better by sharing scarce recourses and information, trusting each other, and engaging in collective action.

Social capital has already been documented to have positive effects on individual health and health-related behaviours and it has been documented that social capital has played an important role to protect individuals from the effects of the COVID-19 pandemic. In a study of seven European countries, Bartscher et al. (2021) show that a one-standard-deviation increase in social capital reduces COVID-19 cases per capita by between 14% and 34%, depending on the country, and excess deaths by between 6% and 35%. They explain that informal rules of containments are more easily adopted in areas with higher levels of social capital, which may lead to a lower number of infections. A study of 37 countries (Imbulana Arachchi & Managi, 2021) mixing high- and low-income countries found that some aspects of social capital (family bonds and security) were associated with fewer deaths during the COVID-19 pandemic but others (community attachment and social trust) were associated with more deaths. In the early phases of the COVID-19 pandemic, before the vaccines, Americans living in counties with high levels of social capital were more likely to reduce mobility linked to retail and recreational activities than people living in counties with low levels of social capital (Borgonovi & Andrieu, 2020). Closer to our question, Hu et al. (2022) show that dense social capital is positively associated with the intention to take COVID-19 vaccine booster shots among urban workers in China.

Methods

Data collection and sampling

We use data from the 2020 Canadian Community Health Survey (CCHS) accessed in the Research Data Centre (RDC) at McMaster University. CCHS is a regular (annual) survey conducted by Statistics Canada to describe the health and health-related behaviours of the Canadian population living in the community. In the fall of 2020, Statistics Canada conducted a special version of the survey, adding questions relative to intentions to get vaccinated as well as having felt symptoms of COVID-19 to the usual survey.

Data were collected between September and December 2020. The COVID-19 vaccination program started in Ontario on December 14, 2020, and vaccines became widely available in that province on April 8, 2021 (Mishra et al., 2021).

The sample was randomly selected by Statistics Canada among individuals aged 12 and over, living in private dwellings (institutional residents are excluded), in 100 regions covering all provinces, but excluding territories, Indian Reserves and Crown Lands, as well as remote areas. Because questions on social capital were not asked to respondents in all provinces, we chose to focus on the sample of respondents from Ontario. The sample size was determined by Statistics Canada so as to get enough power to be able to detect changes in health or health-related behaviours between annual cross-sections at the level of 80 health regions in the country.

Study sample

The initial sample is comprised of 7317 respondents. Five hundred eleven (511) observations were dropped due to missing values for our COVID-related variables (intention to get vaccinated and symptoms), another 143 observations due to missing values for social capital variables, and 147 observations due to missing values for other covariates. Thus, our final sample has 6516 observations.

The study sample is not different from the initial sample in age and sex distributions or in distributions by immigration status, marital status, or labour force participation status, as shown in Supplementary material (ESM Table A1).

Dependent variables

The binary dependent variable in this study is willingness to get the COVID-19 vaccine; it takes a value of 1 if the respondent intends to get vaccinated and 0 otherwise. It is based on the following question: “How likely is it that you would get a COVID-19 vaccine?”, with four response items: very likely, somewhat likely, somewhat unlikely, and very unlikely. We created a binary variable taking the value 1 for individuals responding “very likely” and 0 otherwise.Footnote 1

Independent variables

The independent variable of interest in this study is social capital.

In the literature on social capital, two types of measure exist, capturing different aspects of the concept (Kim et al., 2020). Social capital is often measured at the community level, using aggregate-level data to reflect that what matters is how well the community in which the individual lives functions. But it can also be measured at the individual level, reflecting the ability of the individual to muster resources such as trust or sense of security that are available in their community. In the current survey, we have data at the individual level only (self-reports) but we recreate community-level values by taking the mean of the individual values for all individuals living in a given community.

For each individual, we use five questions on relationships to build a numerical index, taking values between 0 (low level of social capital) and 5 (highest level). Each question is about whether the individual benefits from one aspect of social capital (Do the respondent’s relationships provide a sense of emotional security and well-being? Does the respondent feel there is someone to talk to about important decisions in life? Do the respondent’s relationships recognize their skills and competence? Do the respondent’s relationships share their attitudes and beliefs? Are there people the respondent can count on in an emergency?) and has responses from “strongly agree” to “strongly disagree”. We coded the variable as 1 if the respondent chose “strongly agree” (and 0 otherwise).Footnote 2 We then added all these variables to create an index of social capital (ISC) between 0 (lowest level of social capital) and 5. We aggregated individual values at the community level, using 76 Local Health Integration Networks (LHIN) Sub-regions provided by Statistics Canada to produce aggregate-level social capital. The main roles of 14 LHINs in Ontario are to plan, fund, and integrate healthcare services locally. LHINs were subdivided into 76 sub-LHINs to plan performance improvement and service integration at a community level (Land Information Ontario; https://www.ontario.ca/page/land-information-ontario). The sample size was increased in Ontario in order to produce estimates reliable at the sub-LHIN level, and the CCHS stratification had to be adjusted. Because values are averaged at the sub-LHIN level, the range of the aggregate-level of social capital (ALSC) is much narrower than for the ISC, between 1.6 and 3.3.

Our objective is to measure how much social capital adds to our understanding of the individual variation in willingness to get vaccinated beside usual determinants such as demographic factors (age, sex, marital status), socio-economic background (education, income, immigration, and labour-force participation status), and geography (the urban density of residence). Age is entered in quadratic form (age and age squared divided by 100), to capture non-linear effects; the reference category for sex is female (all our observations are either male or female and we study the effect of being male on willingness to get vaccinated). The reference for marital status is “not in couple”, including single, separated, divorced, and widowed, and we study the effect of being in couple versus being isolated. Education is entered in three categories, the reference being “less than secondary”, and other categories being “secondary graduation but no post-secondary education” and “post-secondary degree/diploma or above”. Income is entered as the natural log value of total household income. We enter immigration status (the reference is Canadian-born) as a proxy for race/ethnicity, and labour-force participation measured as “having worked in the past week” (the reference is those who did not work) as a proxy for labour market status. The urban density of residence is in four categories, with “rural” as the reference, and other categories being “small city”, “medium size city”, and “large city”.

We also control for possible confounding factors, individual characteristics susceptible to varying with social capital and the willingness to get vaccinated identified in the literature, such as self-reported health status (the reference is poor, with four possible categories: fair, good, very good, and excellent) and self-perceived symptoms of COVID-19 since the beginning of the pandemic (the reference is no reported symptoms).

Statistical analysis

The design of the study is observational cross-sectional. As the dependent variable is binary, we estimate logistic regressions (one with social capital measured at the individual level and one with social capital measured at the aggregate level).

One issue worth noting is the response rate: whereas Statistics Canada usually achieves a high response rate on its surveys, this one, being conducted in the middle of a pandemic, was accepted by only one selected individual out of five, yielding a sample of 7317 respondents in Ontario. This is an issue, but we address it as follows: (a) We use post-sampling weights provided by Statistics Canada that help guarantee the weighted sample reproduces the distribution of the Ontarian population by age and sex in each province; (b) We do not produce any univariate statistics (e.g., proportion willing to get tested), but only estimate multivariate relationships: if it is highly likely that non-response biases the mean or other statistics of some variables, it is much less likely that a relationship between several variables is significantly affected by non-responses (Gelman, 2007).

Results

Table 2 shows that the study sample is close to the Ontarian population aged 12 and older on basic demographic characteristics: 49% are males and the average age is 45, close to the average age of the population 12 and older in Ontario (at 46, authors’ calculations based on the life table for 2020), meaning the non-response was not concentrated on some age groups. Immigrants were more likely to respond than Canadian-born, yielding a proportion of immigrants in the sample of 34% versus 29% in Ontario (according to census 2016). Table 2 also shows that 50% of respondents were very likely to get the COVID-19 vaccine (and 76% very or somewhat likely).

Table 2 Descriptive statistics

Tables 3 and 4 present the results (odds ratios and their confidence intervals, and marginal effect calculated at the sample mean on all other variables with its standard error and p-value) of the models, Table 3 for social capital measured at the individual level and Table 4 for social capital measured at the aggregated level.

Table 3 Association between individual-level social capital and willingness to get vaccinated
Table 4 Association between aggregate-level social capital and willingness to get vaccinated

At the sample mean for all other independent variables, each level of individual-level social capital is associated with an increase in the intention to get vaccinated by 2 percentage points. Therefore, from the lowest level of social capital to the highest, the intention to get vaccinated increases by 10 percentage points. To check that our results did not reflect our linear assumption (treating ISC as a continuous variable), we ran a regression using all levels of ISC as dummy variables (level 0 being the reference); we confirm a positive association, with levels 0 to 2 being associated with the same willingness to get vaccinated and levels 3 to 5 being associated with a higher willingness.

The association between aggregate-level social capital and willingness to get vaccinated is not significantly different from 0 at the 5% level (the confidence interval for the odds ratio includes 1) but the marginal effect is quite large (suggesting a 9.5-percentage point difference across the range of values for aggregate-level social capital (1.6–3.3)). The association remains not significant when we define ALSC as a binary variable (low: lower than first quintile of the distribution, versus other values).

Other determinants: once social capital is controlled, we find no effect of self-assessed health on the willingness to get vaccinated for COVID-19; controlling for social capital at the individual level neutralizes the effect of marital status, but those living in couple are more likely to get vaccinated when social capital is controlled at the aggregate level. Other determinants remain statistically significant when social capital is controlled whether at the individual or the aggregate level: the willingness decreases with age until approximately 30 and then increases rapidly (see Fig. 1). Men are more likely to get the vaccine, as are those who are more educated, live in medium or large cities, are born in Canada, have a higher income, or have experienced symptoms of COVID-19.

Fig. 1
figure 1

Probability (Pr) of being very likely to get the vaccine by age (marginal effect)

We check whether these associations between social capital and willingness to get vaccinated were homogeneous across age categories by re-running the analyses separately for three age groups: younger than 20 (teenage), 20 to 60 (working-aged people), and greater than 60 (older people). The upper panel of Table 5 shows that individual-level social capital is significantly positively associated with the intention to get vaccinated after 20, but not for teenagers. When social capital is measured at the aggregate level, the association is not significant among active age adults (20–60) but significant and strong (an effect of 28.6-percentage point over the range of ALSC) among older adults.

Table 5 Analyses of heterogeneity of associations by age

The association between social capital and willingness to get vaccinated is the same for both genders (Table 6).

Table 6 Analyses of heterogeneity of associations by gender

Discussion

Using data from 2020 CCHS September to December special interviews, Ontario sample, we find that individuals with higher levels of trust in their community as well as larger social networks were more likely to plan to get the COVID-19 vaccine when available than individuals with lower levels of trust and smaller social networks, controlling for other determinants of the willingness to be vaccinated. On the other hand, living in a community with higher level of social capital is not meaningfully associated with the willingness to be vaccinated, except for older adults: after age 60, our model predicts a variation of 29 percentage points in the willingness to be vaccinated across the range of aggregate-level social capital (the variation in actual vaccination rates observed across neighbourhoods in Ottawa is 15 percentage points, https://www.neighbourhoodstudy.ca/covid-19-vaccination-coverage-in-ottawa-neighbourhoods/).

Socio-economic determinants of vaccine hesitancy are highly context dependent. Our main contribution is to show that social capital, measured as the amount of trust in the community and the perception one has to have a social network, is associated with the willingness to get vaccinated: the association of individual-level social capital with willingness to get vaccinated is comparable to that of education (12 percentage points between lowest and highest levels of education and 10.5 percentage points between lowest and highest levels of social capital). The association between aggregate-level social capital, which represents the fact of living in a community where individuals have more trust or a better perception of their own social network, and willingness to get vaccinated, is large among older adults, suggesting that, for them, what matters is to live in a sharing community rather than the amount of social capital they can muster for themselves. We are not aware of any previous study of the association between social capital and willingness to get vaccinated, but our results confirm findings that level of trust in institutions or the healthcare system is associated with vaccine hesitancy or acceptability.

Our study is cross-sectional and, as a result, we cannot infer causality and recommend that investing on social capital would improve vaccine acceptance. We can, however, point toward the fact that, in this sample, communities with lower levels of social capital (where lower income and lower education individuals tend to live) can be less receptive to traditional public health messages based on the rational cost–benefit analysis of the outcomes of the vaccine. Also, the results from our sample confirm that receptiveness to messages on vaccines are linked to levels of trust in society in general. If we could understand why some individuals are less integrated in society and feel marginalized, we could certainly find ways to convince them to get vaccinated and protect their own community. Or, in a more pessimistic way, we can conclude that vaccine hesitancy is not something that public health can really cure, as it reflects a deeper social ill, that sociologists call anomy (Durkheim, 1897) and that a recent study of mortality among Americans without a bachelor’s degree (BA) described as a deprivation of meaning and structure for working-class communities (Case & Deaton, 2021).

Strengths and limitations

A limitation of our analysis is that the response rate is lower than that in previous years of CCHS. Using post-stratification weights, and focusing on co-variations rather than univariate statistics, we do not think this limitation prevents us from concluding there is a protective effect of social capital, at least in Ontario. We were reassured to see that intention to get the vaccine (very or somewhat likely) as estimated in our sample is very close to the figure found by Griffith et al. (2021), using Twitter profiles in Canada. Also, all associations with variables such as age or having had symptoms of COVID-19 have the expected signs.

Another limitation is that the questions on social capital have not been posed to respondents from other provinces and we cannot therefore test for our hypotheses outside of Ontario.

The main strength of this study is to prove the association of vaccine hesitancy with a community-driven determinant, something that had been suggested but never established in previous literature.

Conclusion

Using a representative sample of the population living in Ontario, we show that social capital, measured as the perception to benefit from social support, was associated at the end of 2020 with intentions to get vaccinated against COVID-19. The effect is strong at all ages except teenagers when social capital is measured at the individual level, and strong among seniors when measured at the community level. We cannot claim any causal link but our results are suggestive of a plausible explanation for vaccine hesitancy among socially disadvantaged Canadians.

Contributions to knowledge

What does this study add to existing knowledge?

  • We find that social capital has a positive effect on the willingness to get the vaccine against the COVID-19 pandemic after controlling for demographic and socio-economic factors in Ontario.

What are the key implications for public health interventions, practice, or policy?

  • Social capital plays an important role against the pandemic. Thus, our study encourages policymakers to make efforts to strengthen and expand social capital through social transfer, or local community programs during the pandemic.