This section offers a comprehensive review of the data, methodologies, and findings from two distinct but complementary analyses, each aimed at elucidating the pathways involved. The initial analysis employs path analysis, a subset of Structural Equation Modeling (SEM), to investigate the hypothesized links between perceptions of inequality, trust in science and society, and vaccination behavior. Path analysis is instrumental in examining both direct and indirect relationships between variables, focusing on multiple hypothesis models.
This approach provides foundational insights into the dynamics of these relationships by facilitating the exploration of underlying constructs through observed variables and offering a structured framework to assess directional relationships among these variables. Specifically, path analysis allows the study to validate an overarching pathway where perceptions of inequality may lead to decreased trust in science and society, subsequently influencing vaccine hesitancy and uptake. This methodological approach enables a nuanced examination of the proposed model, providing valuable insights into the dynamics and significance of these relationships.
The first utilizes data from the 2021 Korean Happiness Survey, administered by the National Assembly Futures Institute, which gathered responses from over 17,000 individuals representative of Korean society (National Assembly Futures Institute 2021). This survey aims to collect data on happiness, inequality, and their determinants in Korea, in accordance with OECD guidelines for measuring subjective well-being. Additionally, the survey includes variables on attitudes, beliefs, social values, and activities to elucidate the socio-psychological factors influencing happiness. The 2021 dataset incorporates variables related to perceived inequality, trust in science and society, and COVID-19 vaccines, which are critical to the analyses. The survey data were collected through door-to-door interviews conducted using tablet PCs from August 21 to October 27, 2021. The sampling frame was constructed using a nationwide multi-stage stratified cluster sampling design, combining Probability Proportionate to Size (PPS) sampling for aggregate levels and random sampling at the household level (National Assembly Futures Institute 2021).
The second analysis shifts focus to a district-level examination, employing a dataset specifically curated for this scale. This analysis comprehensively details the model and variables in a designated subsection. This methodological shift provides a broader perspective, emphasizing the use of expanded spatial scales and diverse methodologies to explore the impact of perceptions of inequality on vaccination dynamics. By connecting individual perceptions with community-level outcomes, the study clarifies how personal experiences and perceptions of inequality manifest as observable vaccine-related behaviors across various districts.
Analysis 1
Analysis 1 examines the relationship between perceptions of inequality, trust in science and society, and trust in vaccine and vaccination through path analysis. In the analysis according to the theoretical model in Fig. 1, dependent variables for path analysis are Science Trust, Social Trust, Vaccine Trust and Vaccination. Science Trust is measured by the question: “Science and technology will make human life more comfortable and convenient.” Responses range from 1 to 5, with higher values indicating greater trust in science. Social Trust is measured by the question: “In the society where I live, most people are trustworthy.” Responses range from 1 to 5, with high values indicating greater trust in society.
The constructs of vaccine trust and vaccination behavior are operationalized in this study through specific survey items. Trust in vaccine information is quantified using the statement, “It is difficult to believe the information provided about a vaccine for COVID-19.” Respondents indicate their level of agreement on a scale from 1 (Highly agreed) to 5 (Not agreed at all), with higher scores denoting increased trust in vaccine information. Conversely, vaccination behavior is assessed with the inquiry, “Have you been vaccinated against COVID-19, and if not, are you willing to get vaccinated in the near future?” This question is designed to categorize respondents based on their vaccination status or intent, assigning a value of 1 to those who have received at least one vaccine dose or intend to do so shortly, and a value of 0 to those without any intention to vaccinate or who are uncertain. These dependent variables—trust in vaccine information and vaccination status or intention—serve complementary roles. They capture both cognitive and behavioral dimensions of individuals’ responses to COVID-19 vaccination, thereby enriching the study’s analysis and interpretative scope.
Among observed variables for path analysis, variable of interest is Inequality, which indicates perceived inequality, and it is measured by the question: “How equal or unequal do you think income and wealth are in our society?” Responses range from 1 to 5, with high values indicating greater perceived inequality. In addressing factors that might influence the diverse perceptions, this study incorporates control variables. This method aligns with the conventional approach documented in existing literature on vaccination. This analysis takes into account a range of socio-economic and demographic variables that, based on prior discussions, have the potential to influence behaviors related to vaccine uptake (e.g., Choi et al. 2023). These variables encompass demographic and socio-economic characteristics such as personal income (on a 1–12 scale), education levels (on a 1–5 scale), age (on a 1–6 scale), and political orientation (on a 1 = left-10 = right scale). Dummy variables for sex (1 = female), homeownership (1 = homeowner), professional positions (1 = manager or professional), marital status (1=living with spouse or partner), and religion (1 = having a religion) are also included (see Supplementary Appendix A for details).
Analysis 1 applies path analysis to analyze the relationship between aforementioned variables and the equations are as follows. In terms of the indirect effects, equation for Trust in Vaccine for COVID-19 is as following.
$${\rm{Vaccine}}\; {\rm{Trust}}={\beta }_{1}\cdot{\rm{Inequality}}+{\beta }_{2}\cdot {\rm{Science}}\; {\rm{Trust}}+{\beta }_{3}\cdot {\rm{Social}}\; {\rm{Trust}}+{{\rm{\varepsilon }}}_{1}$$
(1)
Where \({\beta }_{1},\,{\beta }_{2},\,{\beta }_{3}\) are the path coefficients representing the direct effects on Vaccine Trust. Next, equation for Science Trust is as following:
$${\rm{Science}}\; {\rm{Trust}}=\mathop{\sum}\limits_{{\rm{i}}=4}^{12}{\beta }_{{\rm{i}}} \cdot {\rm{X}}_{{\rm{i}}}+{\beta }_{13} \cdot {\rm{Inequality}}+{{\rm{\varepsilon }}}_{2}$$
(2)
Where \({X}_{{\rm{i}}}\) represents the nine observed variables as controls (Income, Occupation, Education, Homeownership, Sex, Age, Marriage, Political Orientation and Religious), \({\beta }_{{\rm{i}}}\) are the path coefficients. Next, equation for Social Trust is as following:
$${\rm{Social}}\; {\rm{Trust}}=\mathop{\sum}\limits_{{\rm{i}}=14}^{22}{\beta }_{{\rm{i}}}{\rm{\cdot }}{X}_{{\rm{i}}}+{\beta }_{23} \cdot {\rm{Inequality}}+{{\rm{\varepsilon }}}_{3}$$
(3)
In addition to the indirect effects through Science Trust and Social Trust, the model posits direct effects of the observed variables on Vaccine Trust. This is represented by adding terms to the Vaccine Trust equation:
$${\rm{Vaccine}}\; {\rm{Trust}}=\mathop{\sum}\limits_{{\rm{i}}=24}^{32}{\beta }_{{\rm{i}}}\cdot{\rm{X}}_{{\rm{i}}}+({\rm{Previous}}\; {\rm{terms}}\; {\rm{in}}\; {\rm{Vaccine}}\; {\rm{Trust}}\; {\rm{equation}})$$
(4)
In the presented path analysis, the \({\beta }_{{\rm{i}}}\) coefficients signify the direct influence exerted by each observed variable on Vaccine Trust. This framework incorporates two primary components: path coefficients (β) and error variances (ε), which are deduced through maximum likelihood estimation. This method utilizes the observed covariance matrix of the variables under study. It is important to note that the same equations apply when Vaccination is chosen as the endogenous variable in the model.
The analysis presented in Fig. 4 elucidates several critical relationships between Inequality, Science Trust, Social Trust, and Vaccine Trust. Initially, the data reveals a statistically significant negative effect of perceived inequality on trust in science, evidenced by an estimated coefficient of −0.037 and a p-value of 0.0. This suggests that higher perceptions of inequality associate with diminished trust in science, thereby supporting Hypothesis 1 (H1). Furthermore, Science Trust exerts a significant positive influence on Vaccine Trust, as indicated by an estimated coefficient of 0.157 and a p-value of 0.0, aligning with Hypothesis 2 (H2).
Fig. 4: Path analysis for trust in vaccine (Model 1).
Controls and model fit indices are not reported. Please see Supplementary Tables B1 and B2 for full results. *p < 0.1, **p < 0.05, ***p < 0.01.
In a similar vein, Inequality negatively impacts Social Trust, as shown by an estimated coefficient of −0.102 and a p-value of 0.0, suggesting that greater perceived inequality is associated with lower Social Trust, thereby corroborating Hypothesis 3 (H3). The analysis also reveals that Social Trust positively affects Vaccine Trust, evidenced by an estimated coefficient of 0.012 and a p-value of 0.0, in support of Hypothesis 4 (H4). Additionally, a direct and significant negative relationship between Inequality and Vaccine Trust is identified, with an estimated coefficient of −0.036 and a p-value of 0.0, lending support to Hypothesis 5 (H5). Figure 5 reinforces and amplifies the insights gleaned from the data, demonstrating that the determinants affecting both trust in vaccines and the propensity to undergo vaccination are steadfast and uniform. The evidence presented in Figs. 4 and 5 encapsulates the dual aspects of individuals’ reactions to COVID-19 vaccination—cognitive beliefs about vaccine, alongside behavioral actions, such as the decision to get vaccinated. These results collectively highlight the intertwined nature of cognitive assessments and behavioral intentions within the context of vaccine acceptance, underscoring the consistent patterns observed across these dimensions.
Fig. 5: Path analysis for vaccination (Model 2).
Controls and model fit indices are not reported. Please see Supplementary Tables B3 and B4 for full results. *p < 0.1, **p < 0.05, ***p < 0.01.
In the alternative model (Model 3), trust in vaccines can be considered as a second-stage mediator affecting vaccination behavior as an ultimate outcome. In this model, the path “Vaccine Trust → Vaccination” was included to examine the influence of trust in vaccines on vaccinated or intention. However, this path was found to be statistically insignificant (see Supplementary Fig. B3 in Supplementary Appendix B). To further explain this result, it is essential to consider the possible overlap and redundancy between the constructs of Vaccine Trust and Vaccination.
If Vaccine Trust and Vaccination are conceptually and empirically similar, or if they capture overlapping aspects of the same underlying construct—encompassing both cognitive and behavioral dimensions of individuals’ responses to COVID-19 vaccination—several scenarios can arise. Firstly, when two variables are highly correlated or measure similar constructs, as indicated by the Pearson correlation coefficient of 0.7 (p-value of 0.03) between them, including both in the same model can result in multicollinearity. This multicollinearity makes it difficult to isolate the unique contribution of each variable, potentially diminishing the statistical significance of the path between them, as the model cannot clearly distinguish their individual effects.
Secondly, if the items used to measure Vaccine Trust and Vaccination are not distinct enough, respondents may answer them similarly, resulting in overlapping measurement error. This can further complicate the ability to detect a unique effect of Vaccine Trust on Vaccination, as the measurement issues obscure the true relationship between the constructs.
Both Model 1 and Model 2 show excellent fit indices across all metrics, with Model 2 having a slight edge based on the AIC and BIC values (see Supplementary Tables B2, B4, B5 and B6 in Supplementary Appendix B). Model 3, although still fitting well, does not perform as strongly as Models 1 and 2 based on the Chi2 p-value, CFI, TLI, AIC, and BIC values. In conclusion, while both Model 1 and Model 2 are robust, Model 2 is marginally superior. Model 3, despite being an alternative, does not fit the data as effectively.
Building on the initial analysis, a complementary analysis was conducted to further investigate the statistical foundations of the identified relationships, employing a series of regression models for robustness checks, as detailed in Supplementary Appendix B (pp. 18–27). The results from this complementary analysis consistently support the findings from the first analysis. These regression models assess the contributions of individual variables and their interactions, specifically quantifying the impact of each factor and examining how these impacts contribute to adverse outcomes. Together, these analyses offer a comprehensive understanding of the connections between perceptions of inequality, trust in science and society, and their implications for vaccination behavior. By integrating findings from both path analysis and regression analyses, this study not only validates the hypothesized models but also identifies potential mechanisms through which perceptions of inequality influence vaccination behaviors across different segments of society.
Analysis 2
Analysis 2 builds on the groundwork laid by Analysis 1 by investigating the interplay between inequality and vaccination rates at the district level, shifting the focus from COVID-19 vaccinations to influenza vaccination data from 2015 to 2021. This strategic choice allows for a broader examination of vaccine behavior beyond the immediate context of the COVID-19 pandemic, offering a more generalizable understanding of public health responses.
Analysis 1 focuses on individual perceptions of inequality as the primary independent variable, investigating how these perceptions influence trust in vaccine-related information and vaccination decisions. Analysis 2 extends this investigation to the district level, where economic inequality within administrative divisions (si-gun-gu) serves as the main independent variable. This extension is premised on the idea that individual perceptions and behaviors are significantly shaped by the broader socio-economic environment of their residential districts.
Individuals assess economic inequality by evaluating their position within the income or wealth distribution and comparing their economic status to that of their reference group (Fraile and Pardos-Prado 2014; Gimpelson and Treisman 2017; Hauser and Norton 2017). Research by Bartle et al. (2017), Han and Kwon (2023), Newman et al. (2018), and Szewczyk and Crowder-Meyer (2022) supports the view that geographical location profoundly impacts daily experiences, shaping individual attitudes and awareness. In addition, according to the 2021 Social Integration Survey conducted by the Korea Institute of Public Administration (2022), 71.8% of respondents reported a high sense of belonging to their residential si-gun-gu. This data underscores the role of these administrative units in fostering community and identity, thereby influencing individual perceptions and behaviors. These administrative divisions provide a framework for regional-level analysis, illustrating the direct and indirect effects of local inequality on health behaviors.
By examining influenza vaccination data over an extended period, Analysis 2 offers valuable insights into public vaccine behavior in non-pandemic conditions. This analysis explores how economic inequalities at the district level impact influenza vaccine uptake, reflecting the social context in which residents experience and perceive inequality. Understanding these dynamics within administrative units helps elucidate the relationship between local inequalities and health behaviors, specifically vaccine uptake.
In this district-level analysis, the primary outcome variable is the influenza vaccination turnout rate across 252 administrative districts in Korea, encompassing cities (si), counties (gun), and districts (gu), over the period from 2015 to 2021. The selection of this specific timeframe was primarily driven by the availability of relevant data for these years.
The main explanatory variable is economic inequality at the district level. Due to the absence of direct economic inequality data at the administrative district level, this study employs a measurement methodology (e.g., Han and Kwon 2023) using data from Korea’s public health insurance system. By leveraging national health insurance premiums, which are mandatory for all Korean citizens, the study constructs a proxy for economic inequality. According to Heo et al. (2021), over 51 million Koreans, accounting for at least 97% of the population, were covered by this insurance scheme as of 2021. These premiums, calculated based on individual income and wealth, provide a reliable metric for assessing economic status, thereby offering an approach to examining economic disparities at the district level.
To capture the notion of social context in this study, which refers to how individuals perceive their economic status relative to a reference group, the Palma ratio is employed. This ratio is a measure of income and wealth distribution that compares the income and wealth of the richest segment of a population to those of the poorest. It is particularly pertinent in this study as it mirrors the inequality perceived through a subjective recognition process in the living environment. The economic inequality measurement method of the Palma ratio is thus applied to the national health insurance premium data, disaggregated by household and administrative district, to assess regional economic inequality.
$${p}^{d}=\frac{\sum _{h\in d({R}_{10})}\,{I}_{R10}}{\sum _{h\in d({P}_{40})}\,{I}_{P40}}$$
(5)
In Eq. (5), \({p}^{d}\) refers to Palma ratio, with high values indicating higher economic inequality. \({I}_{R10}\) represents the average insurance premium of the top 10 per cent, and \({I}_{P40}\) represents the average insurance premium of the bottom 40 per cent. The ratio that adds the insurance premiums of households (h) in each income bracket within district (\(d\)) represents the economic inequality level of the region. High \({p}^{d}\) indicates high economic inequality levels, and low \({p}^{d}\) indicates low economic inequality levels within a district.
Due to the unavailability of direct average district-level income and wealth data, this study employs an approach to gauge income and wealth by measuring the average health insurance premium in each administrative district unit. In this context, a higher average premium is indicative of a larger presence of higher income and wealthy individuals in that district, thereby serving as a proxy for the area’s economic level. To accommodate its aggregate nature, the natural logarithm transformation is applied to this variable.
To enrich the analysis and account for potential confounders, several district-level control variables are incorporated. These include factors such as the level of education, which is hypothesized to be associated with vaccine uptake or social behavior (Choi et al. 2023; Han and Kwon 2023). Additionally, demographic variables like the average age and the male-to-female ratio within each district are considered. The degree of urbanization of each administrative district is also factored into the analysis, recognizing that urban and rural areas might differ significantly in terms of healthcare access and attitudes towards vaccination (Wu et al. 2023). Another critical control variable included is hospital availability, quantified as the population per hospital in each district (Wang et al. 2024). This metric provides insight into the healthcare infrastructure and accessibility in each area, which can influence vaccination rates. The methodology and detailed data regarding these control variables are comprehensively outlined in Supplementary Appendix A of the study.
In the principal phase of the analysis, five distinct estimation techniques are executed, as detailed in Table 1. These include: (1) Ordinary Least Squares (OLS), (2) Fractional Probit, (3) Random Effects (RE), and (4) Fixed Effects (FE) models, each incorporating a comprehensive range of control variables. Subsequently, in model (5), a two-way fixed effects model is implemented, which offers more conservative estimates by controlling for unobserved heterogeneity that is specific to both district and time. This approach effectively accounts for regional characteristics and external shocks, such as those brought about by COVID-19. This methodological diversity allows for a robust analysis of the data, ensuring that the results are comprehensive and account for various potential influences and biases inherent in the study’s design.
Table 1 Inequality and vaccinations.
The first model employed is the OLS model, a common econometric method for estimating linear relationships between variables. The OLS results indicate that the coefficient for inequality is −0.011 with a standard error of 0.001, statistically significant at the 1% level. This significance suggests a robust negative relationship between economic inequality and the dependent variable. Following the OLS model, a Fractional Probit model is applied, which is particularly well-suited for fractional dependent variables ranging from 0 to 1. In this model, the coefficient for inequality is estimated at −0.032 with a standard error of 0.003, also statistically significant at the 1% level. The larger magnitude of this coefficient compared to the OLS model suggests that the impact of inequality is more pronounced when using this nonlinear specification, which accounts for the bounded nature of the dependent variable.
The analysis further includes a RE model, appropriate for panel data where unobserved heterogeneity varies across entities but is assumed to be uncorrelated with the independent variables. The RE model yields a coefficient for inequality of −0.014 with a standard error of 0.001, maintaining statistical significance at the 1% level. This result emphasizes the consistent negative effect of inequality on the outcome variable, even when accounting for both within- and between-group variations.
Additionally, two distinct FE models are utilized. The first FE model controls for district-specific fixed effects, yielding a coefficient for inequality of −0.014 with a standard error of 0.006, significant at the 5% level. The second FE model incorporates both district and year fixed effects, offering a more conservative estimate. This model produces a coefficient for inequality of −0.011 with a standard error of 0.005, also significant at the 5% level. These FE models are crucial for isolating the impact of time-varying factors within districts, providing a deeper understanding of the relationship between economic inequality and the dependent variable.
The consistency of results across the various econometric models—OLS, Fractional Probit, FE, and RE—strengthens the credibility of the findings. This multi-model approach ensures robustness by confirming that the observed relationships are consistent across different econometric specifications and not artifacts of a particular model’s assumptions. By comparing results from models with and without fixed effects, we can gain a clearer understanding of how much the observed relationships are driven by within-district changes versus between-district differences.
The key takeaway from this analysis is the consistent negative relationship observed between economic inequality and vaccination rates at the district level. These findings suggest that as economic inequality rises within districts, where residents perceive and experience inequality in their social context, vaccination rates tend to decline.
