COVID-19 vaccination rates and neighbourhoods: evidence from Sweden

ABSTRACT

This paper investigates neighbourhood characteristics related to an individual’s likelihood of getting the first COVID-19 vaccination and implementing official recommendations for the three-shot vaccination regime. We use full population-geocoded microdata for Sweden to measure important individual-level attributes and the marginalisation of their residential communities in terms of ethnicity, education and income. The findings show that the likelihood of getting vaccinated and obtaining all three recommended vaccine doses decrease for individuals residing in neighbourhoods with larger shares of marginalised residents. The effects also appear to be more pronounced if the individual themself belongs to a marginalised group.

KEYWORDS:

• vaccination
• COVID-19
• neighbourhoods
• ethnicity
• education
• poverty

JEL:

• I3
• I14
• R2

1. INTRODUCTION

The ongoing COVID-19 pandemic has reminded us that the features that endow cities with distinctive socio-economic advantages, such as their social networks and their density, also make them vulnerable to the spread of pathogens, especially emerging infectious diseases (Glaeser & Cutler, 2021; Harper, 2021; Neiderud, 2015; Stier et al., 2021). It is by now well known that the effects of urban environments on individual-level outcomes are mediated to a large extent by local residential contexts, that is, neighbourhoods (Chetty & Hendren, 2018; Chyn & Katz, 2021; Durlauf, 2004; Ioannides & Topa, 2010; Sampson, 2008; Sharkey & Faber, 2014; Suttles, 1972). In particular, health-related outcomes are among the important consequences of the neighbourhoods in which individuals are socially embedded (Diez Roux, 2001; Diez Roux & Mair, 2010; Finch et al., 2010; Li & Chuang, 2009; Ludwig et al., 2011; Meijer et al., 2012; Squires & Lathrop, 2019).

In this regard, there were concerns from the very start of the COVID-19 pandemic that the manner and pace of contagion and the impacts of the disease would be greatly magnified by neighbourhood characteristics, with neighbourhood-level inequalities expressed as concentrated poverty, segregation, and deprivation amplifying the virus’s infectivity and deadly consequences. Explanations for this include limitations on information flows (e.g., safety precautions and vaccination recommendations) due to language barriers, as well as a larger presence of overcrowded and inter-generational households in marginalised neighbourhoods.

The transmission and health consequences of an emerging infectious disease like COVID-19 were expected to be greatly modulated by the spatially embedded characteristics of neighbourhoods, creating strong place-based inequalities in outcomes. Aspects of neighbourhoods like social cohesion and connectedness to different communities extend beyond their physical boundaries and can play a role in determining access to prevention and treatment resources. With regards to transmission of pathogens and access to medical care, the early concerns on how the COVID-19 pandemic would be experienced by different local populations have indeed been borne out by numerous recent studies on urban areas and neighbourhoods throughout the globe. The evidence indicates that residents of disadvantaged neighbourhoods experienced a greater epidemic burden, with neighbourhood poverty increasing the likelihood of becoming infected, incurring more severe cases, and the probability of dying once infected (Blundell et al., 2020; Durizzo et al., 2021; Hong et al., 2021; Levy et al., 2022; Mena et al., 2021; Spotswood et al., 2021; Tandel et al., 2022).

Neighbourhood effects encompass not only outcomes but also decisions or choices made by individuals via role models, peer group influences and imitative behaviour (Durlauf, 2004; Piraveenan et al., 2021). A consequential health care-related decision – whether to get vaccinated or not – can thus be expected to be subject to neighbourhood effects through, for example, social norms and networks. Evidence confirms this expectation. While COVID-19 infection disproportionately affected disadvantaged neighbourhoods, these neighbourhoods also lagged in COVID-19 vaccinations (Barry et al., 2021; Rich et al., 2022; Sacarny & Daw, 2021; Sun & Monnat, 2022; Tolbert et al., 2021). In this paper, we focus on the propensity to get vaccinated against COVID-19 at the individual level, more specifically how this decision is influenced by the neighbourhood composition, while controlling for a rich set of individual characteristics. Individual-level factors are important to include as no matter the neighbourhood of residence, previous studies on Sweden show that individuals are more or less likely to get COVID-19 vaccination depending on, for example, his or her own gender, age, ethnic background, income, and civil status (cf. Inghammar et al., 2021; Ljung et al., 2022; Örtqvist et al., 2022; Spetz et al., 2022a; Spetz et al., 2022b). However, such previous studies focus only on either the working age population (aged 18–64 years) (Spetz et al., 2022b), the older population (60 or 70+ age groups), (Spetz et al., 2022a), specific geographical areas in Sweden (Inghammar et al., 2021), or on subgroups of the population (Ljung et al., 2022; Örtqvist et al., 2022). Most studies disregard the local context, the exception being Inghammar et al. (2021) who find that lower socio-economic status (at post code level) decreases the likelihood to get vaccinated. In contrast, our study includes the full population of Sweden (aged 16 and above) and uses a wide range of neighbourhood characteristics capturing the socio-economic status of individuals’ residential milieu. Additionally, we distinguish between the choice to get at least one dose of the vaccine and the choice to become fully vaccinated (corresponding to three doses in the Swedish context).

1.1. Spatial sorting

The identification of neighbourhood effects with regard to attributes of the COVID-19 pandemic, including vaccination rates, encounters two difficulties common to the investigation of the role of neighbourhoods in influencing socio-economic outcomes. First, delineating neighbourhoods remains problematic, as different definitions are often used. Unlike other geographies for which socio-economic data are typically collected by government agencies, neighbourhood-level data are usually not collected (Chaskin, 1997). Often the spatial unit of analysis is a small geographical area defined for purposes other than studying neighbourhoods assumed to have generally consistent characteristics, as is often the case with studies that use ‘Census blocks’ in the United States (Sperling, 2012). Second, the putative ubiquity and pervasiveness of neighbourhood effects also makes them difficult to parse out due to the presence of confounding factors operating at spatial and socially different scales (Klaesson et al., 2021; Lobo & Mellander, 2020; Mellander et al., 2017; Wixe & Pettersson, 2020). It is especially difficult to distinguish between the effects of individual-level characteristics and spatial sorting versus the ‘ecological’ (collective) characteristics of neighbourhoods while controlling for selection effects (Beckett et al., 2022; Macintyre et al., 2002). The problem with estimating neighbourhood effects is that people are nonrandomly allocated to neighbourhoods; people select into neighbourhoods based on their preferences (which might include who they want to live near), resources, and availability of affordable housing (Galster, 2012). This can make it difficult to separate individual characteristics from neighbourhood characteristics, particularly in their effect on the probability of getting vaccinated.

Separating individual-level characteristics from the neighbourhood’s requires the use of individual-level data and spatial units of analysis delineated with the intent of capturing neighbourhood interactions. To do this, we were able to investigate the presence of neighbourhood effects on COVID-19 vaccination rates by availing ourselves of Swedish micro-level data covering the entire resident population aged 16 years or above during the period from 27 December 2020 to 4 May 2022. With these microdata, individuals can be assigned to a specific neighbourhood based on their place of residence. Besides providing information about each individual’s socio-economic characteristics, the register data also provide information about the neighbourhood in which an individual lives. The neighbourhood spatial unit of analysis in this study is the ‘regional statistical area’ (RegSO), which has been specifically constructed to facilitate investigation of socio-economic aggregations at the neighbourhood level. While selection effects might be unavoidable, as individuals are willingly or non-willingly spatially sorted, we are able to control for individual characteristics separately from neighbourhood characteristics.

1.2. Purpose

The decision to get vaccinated is intensely personal and yet has community-wide implications. Specifically, we ask whether neighbourhood characteristics having to do with educational attainment, ethnicity, and income of the residents affected an individual’s likelihood of accepting the vaccination regime. We split our inquiry in two parts aimed at answering two questions. The first examines the willingness to get vaccinated to start with. The second examines the willingness to follow official Swedish health authorities’ recommendations to get three shots in total. The aim of this paper is, in other words, to analyse both the propensity to get the COVID-19 vaccine shot to start with, and the propensity to take all three shots. Data for Sweden indicate that approximately 86% of the population got vaccinated at least once, but only 60% received three shots or more. In Sweden’s urban areas, there was a concern that spatially segregated marginalised groups (i.e., migrants, poor) might have exhibited greater hesitation to get vaccinated on account of information deficits regarding vaccines. As a result, vaccination programmes were implemented specifically to reach marginalised neighbourhoods.

1.3. Results

We find that individuals who reside in marginalised neighbourhoods have a decreased likelihood of both getting the first vaccination as well as getting all three vaccine shots. While there is no formal definition of a marginalised neighbourhood, we define it based on the share of socio-economic marginalised individuals (based on ethnicity (foreign-born), low education level or low income) who live in the neighbourhood. The higher the share, the more socio-economically vulnerable we assume the neighbourhood to be. Overall, we find that the neighbourhood coefficients increase in magnitude when we look only at these socio-economically vulnerable groups (foreign-born, low educated, or low income) in marginalised neighbourhoods. This suggests that the likelihood of reaching marginalised groups with vaccination programmes decreases if they live in neighbourhoods where a large share belong to the same marginalised group.

2. METHODS AND DATA

The official COVID-19 vaccination programme in Sweden started on 27 December 2020. Due to an initial vaccine supply shortage, priorities were scheduled into three phases. The first phase prioritised the elderly living in elderly homes and the caretakers who worked closely with these individuals. Phase 2 covered everyone aged 65 or older. Phase 3 included individuals aged 60–64 as well as individuals considered to be part of risk groups aged 18 or above. In the last and fourth phase, the vaccine was provided to everyone aged 18 or above. From October 2021, vaccinations were also offered to individuals aged 12–17 (Folkhälsomyndigheten, 2021). The Swedish government’s recommendation was for everyone 18 years or older to get three vaccine shots and two doses for children aged 12–17. Since October 2022, the Swedish government no longer recommends that individuals between 12 and 17 years get vaccinated. Unlike many other European countries (Krueger et al., 2022), vaccine passports were never introduced in Sweden. This means that individuals who decided not to get vaccinated did not experience restrictions affecting their daily routines. Swedish employers, however, had the right to demand that their employees get vaccinated. The only monitoring of the vaccinations was via the National Vaccination Register (the source of some of the data used in the analysis reported here). On a voluntary basis, vaccinated individuals could download vaccine passports to facilitate travelling to other countries.

In the context of Sweden’s vaccination regime, we examine the likelihood of being vaccinated against COVID-19 as well as the likelihood of getting all three recommended vaccination shots. We employ a probit modelling framework to test these two phenomena.¹ The estimation is thus done in two separate regressions: (1) the probability of getting vaccinated; and (2) the probability of following the recommendations to get all three vaccine shots. In both cases, the main interest is to estimate to what extent three neighbourhood characteristics – ethnicity, income level or education level – are significantly related to the likelihood to get vaccinated. In both estimations, we control for several individual and regional characteristics potentially influencing the willingness to get vaccinated.

2.1. Delineating neighbourhoods

To analyse how the neighbourhood of residence affected the likelihood of getting vaccinated and following the three-shot recommendation, we take advantage of the geographical information in the data from Statistics Sweden to construct neighbourhood characteristics based on ethnic background, education level, and income. There is no unique neighbourhood definition in Sweden, as several definitions have been developed during different time periods. The SAMS definition (small areas for market statistics) was introduced in 1994 and has often been used in Swedish studies of neighbourhood effects (Statistics Sweden, 2020). However, this definition has been criticised because these neighbourhood areas are not as homogeneous as it is sometimes assumed. In 2018, SAMS was replaced by a new neighbourhood definition known as DeSO (demographic statistical areas) that accounts for changes in buildings and infrastructure since SAMS was developed, as the SAMS definition was no longer considered to meet existing needs.

The spatial units defined as DeSO are, however, relatively small, especially in urban areas, where they may sometimes capture just a housing block rather than a neighbourhood. We thus define neighbourhoods according to Statistics Sweden’s (2020) recent classification of regional statistical areas (RegSO), which builds on an aggregation of the DeSO areas to delineate 3363 units of analysis covering the entire nation. This geographical specification is especially advantageous for studying neighbourhood effects since the purpose of RegSO is to allow for statistical studies that assess socio-economic segregation (Statistics Sweden, 2020). The final RegSO classification was introduced in 2020 and is intended to remain unchanged over time. A benefit of this classification of neighbourhoods is that location-specific demographic conditions are defined with spatial (neighbourhood) borders following, for example, streets, waterways and railways. RegSO areas thus represent actual neighbourhoods rather than arbitrary administrative constructs. Each neighbourhood population ranges from 569 to 21,222 individuals. The most populated neighbourhoods are found in the Stockholm region. Many of the least populated neighbourhoods are located in the northern parts of the country that are more rural or even remote. All variables are described in detail in Appendix A, Sections A2–A4, in the supplemental data online.

Based on the residential geo-coded information, we can construct the socio-economic characteristics of the neighbourhood in which an individual resides. Although there is no general definition of what a marginalised neighbourhood is, we define it based on the share of the population that lives in a neighbourhood belonging to a socio-economically vulnerable group in terms of ethnicity, education or income. The higher the share of socio-economically vulnerable individuals in the neighbourhood, the more marginalised we assume the neighbourhood to be.

2.2. Data

To be able to disentangle the relation between individual and neighbourhood characteristics for the likelihood of getting vaccinated against COVID-19, one needs fine-grained individual-level data combined with detailed geographical information. The vaccination data were retrieved from the Swedish National Vaccination Register covering the period from 27 December 2020 to 4 May 2022. The register is maintained by the Public Health Agency of Sweden (Folkhälsomyndigheten). It was introduced in January 2013 after a decision by parliament that healthcare providers be required by law to report to this register all vaccinations covered by Swedish vaccination programmes.

We thus have access to information about all vaccinated individuals in Sweden aged 16 years and above, including who got vaccinated against COVID-19, the dose number, as well as the provider of the vaccine. We were only unable to match approximately 40,000 individuals (equivalent to 0.5% of the data) to the microdata. The most likely reason for this lack of data may be that they were not Swedish residents in December 2020. Note that while we have access to all COVID-19 vaccinations, we also do not have data on whether these individuals had received any other types of vaccinations before the age of 16, since this is the starting year of our population. This means that, given that mandatory reporting to the National Vaccination Register which started in 2013, most individuals in our data would not have had their childhood vaccinations registered.

We combine the vaccination data with micro-level data from Statistics Sweden on all individuals in the age group 16 years or older providing us with information about their age, gender, income, education level, ethnicity, civil status, occupation and housing type. We also have geo-coded information about the individuals’ residential location, which makes it possible for us to identify who lives in the same neighbourhoods.

3. RESULTS

The regression results are organised based on neighbourhood characteristics, starting with the share of the neighbourhood residents who were:

• foreign born from a country outside of the Nordics/EU (column 1);

• had elementary school as their highest educational attainment (column 2); or

• were at risk of poverty, defined as having an equivalised disposable income below 60% of the corresponding national median (Eurostat, 2021) (column 3).

We employ categorical variables of the neighbourhood shares to capture non-linear effects. Categorical variables also make it easier to interpret the estimated marginal effects. The baseline categories, with the lowest shares of marginalised individuals (ethnicity, education, or income) cover approximately 25% of the individuals in this study. The category with the highest shares of the respective group starts above the 99th percentile. Descriptive statistics for each type of marginalised neighbourhood can be found in Appendix A, Section A5, in the supplemental data online.

Table 1 displays the marginal effects from the first probit estimation, the probability of getting vaccinated to start with, given the neighbourhood characteristics. Table 2 presents results for the second probit estimation: what are the neighbourhood characteristics related to the likelihood of getting all three recommended shots. All results are presented as average marginal effects since the probability for each individual will differ in a probit estimation. The coefficients should therefore be interpreted as the average change in the probability of either getting vaccinated or getting all three recommended shots when the explanatory variable changes by one unit. For categorical variables, the average marginal effect is interpreted as the change in probability from belonging to a specific category as compared with the base category (see Table A2 in Appendix A in the supplemental data online for all baseline categories).

Table 1. Marginal effects from probit regression for the probability of getting vaccinated explained by neighbourhood characteristics with individual and municipal characteristics control variables.

Neighbourhood characteristics	Ethnic		Low education		Poverty
	> 7% to < 15%	0.0084***	> 7% to < 10%	−0.0187***	> 10% to < 17.5%	−0.0091***
	15% to < 25%	0.0035***	10% to < 15%	−0.0299***	17.5% to < 25%	−0.0108***
	25% to < 35%	−0.0071***	15% to < 20%	−0.0336***	25% to < 32.5%	−0.0101***
	35% to < 45%	−0.0119***	20% to < 25%	−0.0350***	32.5% to < 40%	−0.0114***
	45% to < 55%	−0.0267***	25% to < 30%	−0.0349***	40% to < 47.5%	−0.0070***
	55% to < 65%	−0.0267***	30% to < 35%	−0.0342***	47.5% to < 55%	0.0001
	65% to < 75%	−0.0270***	≥ 35%	−0.0275***	≥ 55%	0.0198***
	≥ 75%	−0.0194***
Vaccinated share of marginalised socio-economic group in neighbourhood	0.1879***		0.1512***		0.2881***
No-go zone	−0.0151***		−0.0203***		−0.0296***
Regional characteristics
Population density	−0.00002		−0.0019***		0.0023***
Regional type
Urban	0.0093***		0.0075***		0.0089***
Metro	−0.0028***		−0.0068***		0.0078***
Individual characteristics
Ethnic background
Nordic	−0.0465***		−0.0460***		−0.0452***
EU15	−0.0863***		−0.0888***		−0.0859***
West Balkan	−0.1444***		−0.1417***		−0.1450***
East Europe	−0.2376***		−0.2395***		−0.2375***
Middle East	−0.0742***		−0.0719***		−0.0756***
East Africa	−0.0434***		−0.0423***		−0.0457***
North–South-West Africa	−0.0845***		−0.0839***		−0.0862***
South–Central Asia	0.0103***		0.0110***		0.0102***
Southeast–East Asia	0.0114***		0.0109***		0.0121***
South–Central America	−0.0374***		−0.0370***		−0.0370***
North America–Oceania	−0.0790***		−0.0806***		−0.0821***
Education
High school	0.0155***		0.0148***		0.0154***
Shorter higher education	0.0452***		0.0432***		0.0446***
Longer higher education	0.0640***		0.0619***		0.0641***
Income	0.0310***		0.0312***		0.0318***
Age (years)
30–39	−0.0074***		−0.0069***		−0.0071***
40–49	0.0288***		0.0295***		0.0296***
50–59	0.0557***		0.0562***		0.0562***
60–69	0.0815***		0.0815***		0.0811***
70–79	0.1052***		0.1062***		0.1056***
80–89	0.1073***		0.1086***		0.1078***
90+	0.0855***		0.0870***		0.0859***
Female	0.0138***		0.0141***		0.0140***
Civil status
Single with child	0.0075***		0.0084***		0.0089***
Married	0.0267***		0.0264***		0.0271***
Married with child	0.0378***		0.0377***		0.0390***
Frontline worker	0.0126***		0.0137***		0.0124***
Elderly home	0.0250***		0.0276***		0.0273***
House type
Tenant-owned (apartment)	0.0239***		0.0235***		0.0261***
Owner-occupied (house)	0.0194***		0.0131***		0.0191***
Inter-generational household	−0.0467***		−0.0486***		−0.0471***
Crowded household	−0.0379***		−0.0381***		−0.0388***
Observations	8,665,674		8,665,746		8,665,745
Pseudo-R^2	0.167		0.167		0.167

Note: Dependent variable = 1 if a person has received one shot or more, 0 otherwise.

***Significance at 1% level, **significance at 5% level and *significance at 10% level.

Table 2. Marginal effects from probit regression for the probability of getting the three recommended shots explained by neighbourhood characteristics with individual and municipal characteristics control variables.

Neighbourhood characteristics	Ethnic		Low education		Poverty
	> 7% to < 15%	0.0072***	> 7% to < 10%	−0.0200***	> 10% to < 17.5%	−0.0019***
	15% to < 25%	0.0049***	10% to < 15%	−0.0319***	17.5% to < 25%	0.0010**
	25% to < 35%	0.0019***	15% to < 20%	−0.0350***	25% to < 32.5%	0.0031***
	35% to < 45%	−0.0034***	20% to < 25%	−0.0413***	32.5% to < 40%	−0.0033***
	45% to < 55%	−0.0102***	25% to < 30%	−0.0395***	40% to < 47.5%	0.0084***
	55% to < 65%	−0.0175***	30% to < 35%	−0.0544***	47.5% to < 55%	0.0201***
	65% to < 75%	−0.0127***	≥ 35%	−0.0692***	≥ 55%	0.0533***
	≥ 75%	−0.0494***
Vaccinated share of marginalised socio-economic group in neighbourhood	0.1924***		0.1100***		0.2707***
No-go zone	−0.0292***		−0.0264***		−0.0462***
Regional characteristics
Population density	−0.0128***		−0.0147***		−0.0084***
Stockholm	−0.0148***		−0.0166***		−0.0048***
Individual characteristics
Ethnic background
Nordic	−0.0292***		−0.0279***		−0.0282***
EU15	−0.0101***		−0.0108***		−0.0098***
West Balkan	−0.1917***		−0.1860***		−0.1901***
East Europe	−0.1580***		−0.1563***		−0.1571***
Middle East	−0.2389***		−0.2322***		−0.2399***
East Africa	−0.2819***		−0.2746***		−0.2846***
North–South-West Africa	−0.2087***		−0.2045***		−0.2099***
South–Central Asia	−0.0807***		−0.0767***		−0.0795***
Southeast–East Asia	−0.0077***		−0.0055***		−0.0065***
South–Central America	−0.0576***		−0.0552***		−0.0571***
North America–Oceania	−0.0197***		−0.0205***		−0.0192***
Education
High school	0.0256***		0.0250***		0.0257***
Shorter higher education	0.0619***		0.0595***		0.0614***
Longer higher education	0.1021***		0.0991***		0.1022***
Income	0.0149***		0.0150***		0.0162***
Age (years)
30–39	0.1220***		0.1229***		0.1226***
40–49	0.2486***		0.2498***		0.2498***
50–59	0.3257***		0.3268***		0.3267***
60–69	0.3679***		0.3691***		0.3685***
70–79	0.3692***		0.3709***		0.3702***
80–89	0.3886***		0.3906***		0.3898***
90 +	0.3137***		0.3158***		0.3148***
Female	0.0315***		0.0317***		0.0316***
Civil status
Single with child	−0.0817***		−0.0806***		−0.0791***
Married	0.0319***		0.0317***		0.0328***
Married with child	−0.0324***		−0.0325***		−0.0299***
Frontline worker	0.0233***		0.0240***		0.0230***
Elderly home	0.0106***		0.0101***		0.0124***
House type
Tenant-owned (apartment)	0. 0250***		0. 0238***		0.0270***
Owner-occupied (house)	0.0157***		0.0086***		0.0170***
Inter-generational household	−0.0245***		−0.0266***		−0.0254***
Crowded household	−0.0441***		−0.0438***		−0.0452***
Dose 1 type
Comirnaty	0.0717***		0.0720***		0.0723***
Vaxzevria	0.1390***		0.1390***		0.1394***
Observations	7,573,906		7,573,971		7,573,968
Pseudo-R^2	0.182		0.182		0.182

Note: ***Significance at 1% level, **significance at 5% level and *significance at 10% level.

Robust standard errors are shown in parentheses. Dependent variable = 1 if the individual has got all three recommended shots.

Starting with the probability of getting vaccinated at all (Table 1), we find that it decreases as the share of marginalised groups increases in the neighbourhood. For example, the likelihood that an individual is vaccinated is more than 2.67 percentage points lower in neighbourhoods where foreign-born individuals from countries outside of the EU makes up more than 45% of the population, compared with the baseline neighbourhood where this group makes up less than 7%.

Moving on to marginalised neighbourhoods in terms of education level, we find a similar pattern of a decreased likelihood of getting vaccinated in neighbourhoods with higher shares of low-educated individuals. In neighbourhoods where more than 10% are low-educated individuals, the probability of getting vaccinated is, on average, 3.0–3.5 percentage points lower compared with the baseline neighbourhood where less than 7% are low-educated.

The last estimation captures neighbourhoods based on the share of individuals who live at risk of poverty. Here our results may be perceived as counterintuitive. The likelihood of getting vaccinated decreases as the share of poor increases, but it becomes insignificant and even positive in the neighbourhoods with the highest shares of poor individuals. However, this last result can be explained by the fact that many of the poorest neighbourhoods are student areas in college towns, and therefore cannot be considered truly marginalised based on low income. In these types of neighbourhoods, the probability of getting vaccinated increases by 1.98 percentage points. At the individual level, income is significantly related to the likelihood of getting vaccinated. We also find a positive relation between the share of poor in the neighbourhood that has gotten vaccinated and the likelihood of any individual in that neighbourhood to also get the vaccine.

Next, looking at the individual characteristics, we see that the likelihood of getting vaccinated increases with the level of education. Those with a higher education background were more than 6 percentage points more likely to get vaccinated than those with a lower education level than high school. At the neighbourhood level, we find, however, that if a larger share of the low educated have gotten vaccinated, it increases the likelihood of getting vaccinated in general. This might point to a peer effect that compensates for the general negative effect of residing in a low-educated neighbourhood.

No matter the neighbourhood of residence, individuals residing in overcrowded and/or inter-generational households were less likely to get vaccinated, which may have further increased the spread of infection. Interestingly, our results for the individual level characteristics regarding age, gender, education, ethnic background, and civil status are well aligned with the results of Spetz et al. (2022a; 2022b).

Of the 8.6 million people included in the data, approximately 86% are vaccinated, but only about 60% of the population received all three shots. So, what was the probability that an individual got the three recommended shots? Table 2 illustrates the results from these estimations.

Again, we show the average marginal effects from each neighbourhood type (differentiated by ethnic composition, educational attainment, and poverty levels). We find that individuals living in marginalised neighbourhoods were less likely to get all three shots after taking the first shot. For the ethnicity effect, we find it to be the biggest in neighbourhoods where more than 75% of the residents are foreign born. Here, individuals were almost 5 percentage points less likely to get all three shots compared with the baseline group (neighbourhood with less than 7% foreign-born). In marginalised neighbourhoods based on education, there is also a decreasing likelihood of getting fully vaccinated the larger the share of low educated. In neighbourhoods where more than 35% of residents had a low-education level, individuals were close to 7 percentage points less likely to get all three shots compared with the baseline. For marginalised neighbourhoods based on the risk of poverty, we find approximately the same pattern as above. In places where more than 55% of the residents were defined as at risk of being poor, individuals were more than 5 percentage points more likely to get vaccinated, again probably due to the fact that many of these neighbourhoods are located in college towns, home to university students.

At the individual level, we find that people who belong to these marginalised groups were less likely to get all three shots. Starting with ethnicity, we find that individuals from East Africa were more than 27 percentage points less likely to get all three shots as compared with native Swedes, followed by individuals from the Middle East (approximately 23 percentage points), North–South-West Africa (more than 20 percentage points), and West Balkans (around 19 percentage points). We also find that individuals with lower education levels were less likely to get all three shots, and that low-income individuals were less likely to get vaccinated.

4. ROBUSTNESS CHECKS

To further investigate the role of marginalised neighbourhoods, we focused our procedure on these subgroups more specifically. While the first estimations (Tables 1 and 2) included all individuals regardless of ethnicity, education level, or income, we now select each of these groups specifically (keeping in mind that many times they can be overlapping). We thus re-ran the analysis again and only include (1) foreign born individuals, (2) low-educated individuals or (3) poor individuals. Our objective with this disaggregated analysis is to see if neighbourhoods matter differently for individuals belonging to these marginalised groups. (In the re-estimation including only marginalised groups, as per Tables 3 and 4, the variable ‘Vaccinated share of marginalised socio-economic group in neighbourhood’ is excluded, since it becomes too closely related with the neighbourhood shares of this group.)

Table 3. Marginal effects for the probability of getting vaccinated explained by neighbourhood characteristics with individual and municipal characteristics control variables – only for individuals in marginalised groups.

Neighbourhood characteristics	Only ethnic		Only low education		Only poverty
	> 7% to < 15%	0.0138***	> 7% to < 10%	−0.0251***	> 10% to < 17.5%	−0.0024
	15% to < 25%	0.0056***	10% to < 15%	−0.0412***	17.5% to < 25%	−0.0050***
	25% to < 35%	−0.0045**	15% to < 20%	−0.0509***	25% to < 32.5%	−0.0044***
	35% to < 45%	−0.0105***	20% to < 25%	−0.0501***	32.5% to < 40%	−0.0141***
	45% to < 55%	−0.0255***	25% to < 30%	−0.0528***	40% to < 47.5%	−0.0002
	55% to < 65%	−0.0296***	30% to < 35%	−0.0523***	47.5% to < 55%	0.0123***
	65% to < 75%	−0.0344***	≥ 35%	−0.0463***	≥ 55%	0.0393***
	≥ 75%	−0.0213***
Control variables	Yes		Yes		Yes
Observations	1,512,712		638,214		1,282,938
Pseudo-R^2	0.134		0.114		0.147

Note: ***Significance at 1% level, **significance at 5% level and *significance at 10% level.

Dependent variable =1 if the individual has had at least one vaccine shot. All control variables are included.

Table 4. Marginal effects for the probability of getting the three recommended shots explained by neighbourhood characteristics with individual and municipal characteristics control variables – only for individuals in marginalised groups.

Neighbourhood characteristics	Only ethnic		Only low education		Only poverty
	> 7% to < 15%	−0.0102***	> 7% to < 10%	−0.0203***	> 10% to < 17.5%	−0.0060***
	15% to < 25%	−0.0275***	10% to < 15%	−0.0311***	17.5% to < 25%	−0.0047**
	25% to < 35%	−0.0414***	15% to < 20%	−0.0354***	25% to < 32.5%	−0.0055**
	35% to < 45%	−0.0496***	20% to < 25%	−0.0451***	32.5% to < 40%	−0.0207***
	45% to < 55%	−0.0636**	25% to < 30%	−0.0466***	40% to < 47.5%	−0.0044*
	55% to < 65%	−0.0742***	30% to < 35%	−0.0647***	47.5% to < 55%	0.0124***
	65% to < 75%	−0.0757***	≥ 35%	−0.0625***	≥ 55%	0.0359***
	≥ 75%	−0.1007***
Control variables	Yes		Yes		Yes
Observations	1,081,971		483,721		929,442
Pseudo-R^2	0.156		0.201		0.184

Note: ***Significance at 1% level, **significance at 5% level and *significance at 10% level.

Dependent variable =1 if the individual has had all three recommended shots. All control variables are included.

The research question now becomes whether the likelihood for foreign born/low-educated/low-income individuals to get vaccinated and get all three recommended vaccinations is affected by whether they live in a neighbourhood where their own group comprises a larger share of the population. For example, if you are born in a country outside of the Nordics/EU, will the probability of getting vaccinated decrease by the share your own group makes up in neighbourhood? Furthermore, will it decrease more than in what we found for the overall population as per Tables 1 and 2. Tables 3 and 4 illustrate the neighbourhood average marginal effects for each of the marginalised groups (full tables are available from the authors upon request):

Starting with the first estimation (Table 3), we find that someone born outside the Nordics/EU becomes less likely to get vaccinated as the larger the share of their own group that lives in the neighbourhood. We also get similar results for those with a lower education or low income. For example, in Table 1, we saw that the likelihood for someone who lives in a neighbourhood where foreign-born individuals make up 65–75% of the population, the likelihood of getting vaccinated is approximately 2.7 percentage points lower for people on average compared with the result if that person lives in a neighbourhood where foreign-born make up less than 7%. But here we show that foreign-born individuals become 3.4 percentage points less likely (Table 3) if they live in the same type of neighbourhood. In other words, the magnitude of the coefficient increases for individuals who belong to a marginalised group and who live in a neighbourhood where the share of this group is larger. This points to relatively strong peer effects regarding COVID-19 vaccinations within marginalised groups.

We find a similar pattern for the likelihood of getting all three recommended vaccine shots (Table 4). Using the same example again, someone (regardless of place of birth) who lives in a neighbourhood with 65–75% foreign-born is 1.27 percentage points less likely to get all three vaccine shots compared with an individual who lives in a neighbourhood where foreign-born make up less than 7% of the population. But for foreign-born individuals who live in a neighbourhood with 65–75% other foreign born, the likelihood is now approximately 7.6 percentage points less than for a foreign-born individual who lives in a neighbourhood where their own group makes up less than 7% of the population. The same pattern holds for low-educated and low-income individuals.

5. CONCLUSIONS

By now it is a foundational assumption of urban studies and urban sociology that neighbourhoods are a persistent feature of cities that exert causal effects on a wide variety of individual-level outcomes and decisions, and that neighbourhoods mediate and are mediated by society-wide processes and structures (e.g., political, economic, legal) and individual-level processes and choices (Sampson, 2019). There has been an extensive literature on quantifying the relationship between various aspects of the residential environment and numerous outcomes for individuals, but the unequivocal identification of the causal mechanisms that yield these relationships remains challenging (Galster, 2012). The most severe problem in the identification of causal neighbourhood effects is selection bias as a result of selective sorting into neighbourhoods. People sort themselves into and out of neighbourhoods, and selection bias occurs when the selection mechanism into neighbourhoods is not independent from the outcome studied. As a result, it is difficult to establish whether correlations between neighbourhood characteristics and individual outcomes reflect causal effects or are the result of neighbourhood selection.

The effects of neighbourhood characteristics on the health of its residents have received a lot of attention among researchers aiming to understand neighbourhood effects. These investigations have tended to focus on environmental hazards and access to resources (Darling & Steinberg, 1997). The study of neighbourhood effects on health suffers from this selection bias, compounded by using evidence generated by studies designed for purposes other than the examination of the role of neighbourhoods in determining health (Pickett & Pearl, 2001; Kawachi and Berkman, 2003). The novelty of our investigation thus stems from studying neighbourhood effects through the lens of an individual-level decision affecting health (Larsson et al., 2022; Pingali et al., 2021; Stoeckel et al., 2021).

Is the likelihood of getting vaccinated – an individual decision with significant group health effects – affected by the characteristics of one’s residential community? The results presented here strongly indicate that in the case of Sweden, the likelihood of getting vaccinated, as well as getting all three recommended vaccine shots, decreases for individuals residing in neighbourhoods with larger shares of marginalised residents. Further, the magnitude of the regression coefficients increases if an individual belongs to this marginalised group himself or herself. There thus seems to exist strong peer effects within marginalised groups.

We can also conclude that there is a are strong peer effect of vaccination rates at the neighbourhood level (as captured by the variable measuring the vaccinated share of marginalised socio-economic groups in neighbourhood). At the same time, individual characteristics and ethnic background, which potentially also might be a proxy for cultural networks, play a large role in the choice to get vaccinated. Not surprisingly, the likelihood of getting vaccinated and receiving all three recommended dosages increases with age, education, and income.

The data used in our investigation are individual-level data: we are able both to detect the decision made by an individual as to whether to vaccinate or not, and to locate each individual in a specific residential community. Nevertheless, despite the use of individual-level data and individual-level controls, we still cannot firmly conclude whether the results obtained signal the presence of neighbourhood effects or selection effects. We cannot say that there is a causal effect of living in a more marginalised neighbourhood, on the tendency of individuals to resist vaccination. It may be that individuals who are marginalised (immigrant, low education, low income) tend to hesitate to get vaccinated and so move to places where other vaccination-hesitant individuals already live. Or it may be that when individuals move to neighbourhoods with a large share of residents who are vaccine hesitant (and marginalised), they are influenced by their neighbours and become vaccine hesitant themselves. Nevertheless, the important result of either process is that vaccine-hesitant individuals cluster spatially, which is what the analysis presented here reveals. But we also report that the individual-level decision to vaccinate may be affected by salient features of their residential communities. For policy makers who aspire to less inequality among neighbourhoods with respect to health outcomes, this is an unfortunate outcome as it can lead to more infections and deaths. Understanding how peer groups and neighbourhood effects inform an individual’s decision to vaccinate against new pathogens will become more urgent as the warming effects of climate change aggravate existing pathogenic transmission pathways and increase cross-species viral transmission risk (Carlson et al., 2022; Mora et al., 2022).

Besides spatial concentration effects (a more precise and accurate term than ‘neighbourhood effects’), we also find strong variation across ethnic backgrounds. Why do individuals coming from East African and the Middle East behave much differently from native Swedes when it comes to taking three vaccination doses rather than one or two? Is it the result of neighbourhood effects? The results show that even controlling for neighbourhood effects these nationalities have a lower vaccination propensity, so the answer is no. Only some part of their lower vaccination propensity is due to neighbourhood effects. Or how do current local neighbourhood effects interact with the information flow from migrants’ native home communities with regards to choices about health care? These questions are deserving of further study.

This study is also limited in the sense that it does not take any actual networks of contacts into account; we cannot know for sure who is networking with whom. Neither can we account for the social media networks that very likely played a significant role in the likelihood of getting vaccinated. It is also plausible that some individuals in these marginalised neighbourhoods (e.g., religious leaders, anti-vaccine believers, etc.) played significant roles in the spread of information (or misinformation) regarding the COVID-19 vaccinations. Our findings should therefore be considered as interim results in light of these caveats.

DISCLOSURE STATEMENT

No potential conflict of interest was reported by the authors.

Additional information

Funding

For this work, approval was obtained from the Swedish Ethics Review Authority (Etikprövningsmyndigheten) [reference number 022-00345-02].

Notes

1. For robustness reasons we also ran the regressions as a Heckman model where we formulated the regression equations in two steps: first, what is the likelihood of getting the vaccine; and second, given that an individual got the first vaccine shot, what is then the likelihood of getting all three recommended shots. We also ran the regressions using a multilevel model framework based on a 10% sample of our population. All neighbourhood results are presented in Appendix A, Section A1, in the supplemental data online.