Longitudinal Pathways to Influenza Vaccination Vary With Socio-Structural Disadvantages

Abstract

Background

Although influenza vaccination can prevent influenza-related deaths, uptake remains low, particularly in disadvantaged populations.

Purpose

A theoretical model of psychological pathways to vaccination accounting for the direct and moderating role of socio-structural factors was tested. The study sought to understand the joint contributions of psychological (i.e., knowledge, attitudes, and intention) and socio-structural factors (i.e., income, education, and insurance) to influenza vaccination, prospectively.

Methods

A nationally representative empaneled sample of over 3,000 U.S. adults answered questions about vaccination knowledge, attitudes, and intentions, as well as actual vaccination across five timepoints from September 2018 to May 2019. Socio-structural factors were examined as moderators.

Results

Findings revealed strong positive associations between knowledge and attitudes, attitudes and intentions, as well as intentions and subsequent vaccination. Importantly, health insurance moderated the associations between attitudes and intentions and between intentions and vaccination, such that those without insurance had weaker associations between attitudes and intentions and between intentions and vaccination. In addition, education moderated the path from knowledge to attitude and from intentions to vaccination, such that people with lower educational attainment had weaker associations between knowledge and attitudes and between intentions and vaccination.

Conclusions

Socio-structural factors act as barriers to the influence of knowledge on attitudes, attitudes on intentions, and intentions on behavior. Future research needs to be mindful of the specific paths disrupted by social disadvantages and examine ways to intervene to decrease those effects.

Knowledge, Attitudes, Intentions, Influenza vaccine, Barriers

Introduction

Although the Centers for Disease Control and Prevention (CDC) recommends that, with rare exceptions, U.S. inhabitants over the age of six months be immunized against influenza each year [1], the uptake of the influenza vaccine in the past decade has hovered between 42% and 49% [2]. Even more troubling are the social and structural, or socio-structural, determinants that reduce the likelihood that an individual will be vaccinated against influenza. People with lower income, less educational attainment, and lack of health insurance are less likely to receive the influenza vaccine [3, 4]. Because of their higher prevalence of chronic health conditions, the groups that are less likely to vaccinate are often more likely to experience complications from influenza [5]. Understanding psychological pathways to vaccination as a function of socio-structural disparities is thus important for the health of our nation. Influenza vaccination lowers the probability of influenza-related infections and reduces influenza-related hospitalization [6]. For this reason, in the United States, death rates from influenza are inversely proportional to the uptake of the influenza vaccine. For example, the 2016–2017 season had both the highest rates of hospitalization and death as well as the lowest vaccination rates since 2011 [7]. As a result, the need to increase vaccine uptake has motivated a variety of efforts geared towards (a) improving knowledge and attitudes and (b) increasing socio-structural access to the vaccine (e.g., support from community-based pharmacists [8]).

To enhance our understanding of the interplay among psychological factors and socio-structural factors, we propose a hypothetical model of pathways to vaccination. The model assumes that first knowledge, attitudes, and intentions are necessary, but not sufficient, precursors of vaccination. Second, socio-structural barriers such as financial difficulties can directly reduce vaccination and also block the degree to which knowledge, attitudes, and intentions lead to vaccination [9]. We tested this model with a nationally representative sample of over 3,000 U.S. adults who answered prospective questions about the influenza vaccine during the 2018–2019 influenza season.

Pathways to Vaccination Behavior

We propose a hypothetical model to explain key psychological pathways to vaccination. It comprises a sequence that goes from knowledge, to attitudes, to intentions, to behaviors [10–12] and introduces socio-structural factors as moderators of these pathways [13–17]. Our model represents pathways that have been proposed in previous work [10–12, 18–20], but have not been tested in the area of vaccination behavior. Knowledge entails an individual’s understanding of influenza transmission and vaccination, attitudes entail evaluations of the vaccine as desirable, intentions involve the will to execute a behavior such as vaccinating, and behaviors are overt actions such as vaccinating. The psychological variables of knowledge/beliefs, attitudes, and intentions are important in many models of health behavior (e.g., the reasoned action approach [11, 21], the health belief model [22], the information-motivation-behavioral skills model [23]). Our model is similar to the one proposed by Albarracin [15] (see also [14, 24]) in that the information is processed in stages, which may begin from knowledge acquisition to form beliefs, and progress to form attitudes and then intentions. Key to this process understanding, progress from one stage to the next may be disrupted with disruptions in cognitive ability and motivation. Likewise, the health action process approach proposed by Schwarzer [16] states that barriers, as well as opportunities, impact the relations between the psychological factors that predict and define behavior. Furthermore, our model is not concerned with all possible factors impacting behavior (e.g., self-efficacy, perceived risk perception, or norms), as most exert their influence through intentions or are not directly implicated in the processes of transition from knowledge to attitudes or from attitudes to intentions. However, some of these factors are examined in the Appendix.

Knowledge can be a starting point to the formation of attitudes and intentions to vaccinate, and also to vaccinate. The association between knowledge and attitudes is small to moderate in the domain of influenza vaccination (correlation between r = .17 and r = .18) [25, 26]. Further, knowledge is a basis for attitudes toward vaccination [27]. Without knowledge, people often remain unaware of the benefits of a health behavior and thus cannot move closer to enacting it [28]. Following the path from knowledge to attitudes, our model includes a path to intentions and behaviors. In the specific area of vaccination against influenza, attitudes, and intentions have been reported to correlate between r = .32 and r = .68 [29, 30]. In a nutshell, positive attitudes toward the influenza vaccine should correlate with strong intentions to receive the vaccine. In the domain of vaccination, although the links from knowledge to attitudes to intentions to behavior are relatively noncontroversial, there is surprisingly little empirical research on the extent to which socio-structural factors moderate these links. Attitudes toward an object and intentions to engage in a behavior are likely not formed in the absence of some degree of deliberation. For example, although experimental research has shown that individuals can still learn relevant messages while being distracted, their ability to use the new knowledge to form attitudes decreases [31–33]. Likewise, forming an intention on the basis of an attitude also requires thinking about when and how to perform the behavior [34, 35]. Thus, the movements from knowledge to attitude and from attitude to intention may be less likely if people have a lower capacity to meaningfully utilize health information, as is the case when they have lower educational attainment [36], or if they are less motivated to think about the influenza vaccine because the vaccine is perceived to be out of reach due to the lack of health insurance. Thus, our model predicts that socio-structural disadvantages may decrease associations between knowledge and attitudes and between attitudes and intentions. Likewise, progress from intentions to behavior may be more likely when people have more resources. Socio-structural barriers play an important role in translating intentions into behavior [37]. By easing implementation, socio-structural factors such as income, education, and access to health insurance can increase the probability of vaccination [38]. A person may have the intention to vaccinate but lack the money or health insurance required to access the vaccine. For example, evidence suggests that socioeconomic status moderated the intention-behavior relationships across other health behaviors (smoking, breastfeeding, and physical activity) [39, 40]. In the domain of vaccination, however, previous research has not tested how these socio-structural factors moderate the psychological processes.

Past research supports our hypothesis that socio-structural factors can affect both vaccination and the pathways to it. First, people with higher income have greater access to education, which increases the capacity to process the available knowledge and derive attitudes and intentions from it [41]. Second, people with access to primary sources of information such as physicians can receive factual information relevant to health decisions, including vaccination decisions [42]. Third, people with more stable and adequate income have more access to information from both traditional and social media, which are becoming an indispensable repository of readily accessible information for the public [43]. However, this past research does not provide a complete understanding of the psychological processes by which socio-structural determinants influence vaccination.

Present Study

This study sought to understand the psychological processes by which socio-structural disadvantages (i.e., low income, low education, and lack of health insurance) affect influenza vaccination. Using a nationally representative sample, we initially assessed socio-structural factors, levels of knowledge about influenza vaccination, attitudes toward the influenza vaccine, intentions to vaccinate against influenza, and then actual influenza vaccination across the entire influenza season. Respondents answered the questions about receiving influenza vaccine at five timepoints. We tested the pathways to vaccination by including the main effects and two-way interactions with socio-structural disadvantages in a structural equation model of the longitudinal pathways to vaccination. We hypothesized that socio-structural disadvantages (i.e., low income, low education, and lack of health insurance) may decrease associations between knowledge and attitudes, between attitudes and intentions, and between intentions and behavior.

Beyond our proposed psychological pathways to behavior, behavioral control to overcome vaccination barriers is also positively associated with the intentions to and with the probability of influenza vaccination [44]. Several studies have also found that social norms have a significant influence on intentions to uptake immunizations as well as actual vaccination [45, 46]. However, similar to our proposed psychological pathways to behavior, the impact of behavioral control and norms on behavior may be attenuated with socio-structural disadvantages. Supplementary Appendix G extended the model to investigate whether the socio-structural disadvantages moderate paths from other psychological factors to vaccination.

Method

Participants

In this study, we analyzed four waves from a nationally representative panel study. The study included a probability-based, nationally representative sample of American adults who were randomly selected from the AmeriSpeak Panel of National Opinion Research Center (NORC) at the University of Chicago. For more details on the AmeriSpeak panel recruitment, please refer to https://amerispeak.norc.org/ [47]. All participants answered questions as part of a panel survey about vaccination administered during the 2018–2019 influenza season. Participants provided informed consent and study procedures were approved by the university’s Institutional Review Board.

Demographic information was collected during an initial assessment in September 2018 (time 1). Among other items, participants provided responses concerning their knowledge, attitudes toward the influenza vaccine, and intentions to receive the influenza vaccination during the 2018–2019 influenza season. As part of the panel survey, participants were contacted every 2 months from the initial assessment (i.e., time 2 – November 2018; time 3 – January 2019, time 4 – March 2019; time 5 – May 2019). At each timepoint, participants were asked about their attitudes toward the influenza vaccine, intentions to vaccinate, and if they had received the influenza vaccine during the current season. The vast majority of participants responded via online surveys (N = 2,725), with a small proportion responding through telephone interviews. Mean differences were assessed to compare online and phone respondents. Differences emerged wherein phone respondents had stronger intentions to vaccinate at time 1 and more knowledge about vaccines whereas Internet respondents had more positive attitudes at times 1, 2, and 3 (Supplementary Appendix A). However, the average Cohen’s d effect size of the difference was .01.

Measures

All items participants responded to were selected and crafted for this panel study based on Ajzen and Fishbein’s (1980) and Albarracin et al.’s (2018) recommended practices and principles for the design of attitudes and intention measures [10, 21, 48–49].

Knowledge

An objective 5-point scale knowledge score was calculated based on responses to a true/false 11-item battery. Participants were asked, for example, “Receiving the flu vaccine can cause someone to get the flu” and “Getting vaccinated for a specific disease guarantees that you will not get that disease.” The difficulty of the items was analyzed through Item Response Theory (IRT) and found that none of the items required a high ability to answer correctly. Further, the discrimination parameters suggest that all items were indicative of their corresponding scale (see Supplementary Appendix B for details). Knowledge was measured at time 1.

Attitudes

Attitudes were measured with three items: (a) “Based on what you know, how positive or negative do you feel about the flu vaccine?,” (b) “Just your best guess, how risky, if at all, do you think the flu vaccine is,” and (c) “Just your best guess, please indicate how effective, if at all, you think the flu vaccine will be at preventing the flu among those who get the vaccine this season?” These were rated on a 4-point scale. These three items were combined to create a composite measuring attitude (α = .94). Attitudes were measured at every timepoint.

Intentions

Intentions were assessed through a one-item question, traditionally used to measure intentions [13, 21]. “How likely, if at all, are you to get the flu vaccine before or during this flu season?” This item was rated on a 4-point scale. A similar question has been used successfully in past studies of intentions in the domain of influenza vaccination [50–52]. Intentions were measured at every timepoint as long as participants did not report receiving the vaccine. Participants who reported vaccinating were removed from analyses in future time points.

Vaccination

Participants were asked “Have you gotten the flu vaccine this season or not” during every time point (1 = Yes; 0 = No). This self-report is also used by the CDC to estimate vaccination via the Behavioral Risk Factor Surveillance System which is among the most widely used system for monitoring health behavior in the country. About 50% of the participants were vaccinated by time 5. As with intentions, vaccination was measured at every timepoint assuming participants did not report receiving the vaccine. Participants who reported vaccinating were no longer asked about their intentions and were removed from analyses in future time points. However, a second set of analyses was conducted to model the time to vaccination.

Income, education, and health insurance

Participants were asked about their household income in the past year. Participants could choose their income category starting from 1 = Less than $5000 to 8 = $35,000 to $39,999, by increments of $5,000. Response options then were 9 = $40,000 to $49,999 to 10 = $50,000 to $59,999. Increments changed to $15,000 starting from 11 = $60,000 to $74,999 to 13 = $85,000 to $99,999. The final increments were $25,000 from 14 = $100,000 to $124,999 to 18 = $200,000 or more. To take into account household size, our measure of income in our regression models is the ratio of the median income level in the category selected by the survey respondents divided by square root of the household size. Additionally, participants were asked about their educational attainment. They could choose from four categories: 1 = No HS diploma, 2 = HS graduate or equivalent, 3 = Some college, and 4 = BA or above. Participants were also asked if they currently had health insurance. Income, education, and health insurance was measured at time 1.

Statistical Analysis

We used multilevel structural equation modeling (MSEM) to examine the effects of individual and socio-structural factors and mediation paths from knowledge, to attitudes, to intentions, to vaccination. The first equation in the model uses a linear model to predict attitudes, the second linear model predicts intentions, and a logit model that predicts vaccination decision. Timepoints (level 1) were nested within individuals (level 2). Variables at level 1 were standardized within timepoints and variables at level 2 were standardized between timepoints. Vaccination was modeled prospectively with variables from the previous timepoint predicting vaccination at the subsequent timepoint. There were four timepoints nested within level 2 because vaccination was predicted in the following timepoint and only five timepoints were available. Thus, vaccination could be predicted only for four of these timepoints.

Our goal was also to examine the moderation of these paths from knowledge, to attitudes, to intentions, to behavior. Thus, we introduced income, education, and health insurance, as well as their interactions with each of the focal predictors (i.e., knowledge, attitudes, and intentions) into our MSEM model [53–55]. For simplicity, we treated these variables as continuous in all analyses. Supplementary analysis replicated our models treating the predictors as categorical factors. All models controlled for race/ethnicity (i.e., dummy coded with White as the reference group). Time was also accounted for in the analyses as a main effect and through interactions with knowledge, attitudes, and intentions because effects might vary over the influenza season.

Given the study design, where the endpoint of participation was a vaccination event after which assessments ceased, in a second set of analyses, time-to-event analyses with a right-censored data (respondents not vaccinating during the survey period) were used to model the overall rate of vaccination and its predictors. In particular, we modified the MSEM to investigate whether the socio-structural and psychological factors affected time to vaccination. This model involved linear models to consider associations from knowledge to attitudes and from attitudes to intentions, but a Weibull time-to-event model to predict time to vaccination.

We considered an alpha correction given the large number of significance tests conducted, but concluded that it would be inappropriate because of the comparisonwise, compared to familywise, nature of our hypothesis tests—each test as independent and not contingent on results from other paths within the model [56–61]. Analyses were conducted with corrections for weighting to yield nationally representative estimates.

Results

Sample Characteristics

Table 1 presents the breakdown of demographics, income, education, health insurance, and each of the focal variables (i.e., knowledge, attitudes, intention, and vaccination) within the sample for each of the timepoints of the study. The sample was primarily White, with Whites comprising 63%, Latinx 16%, Black respondents 12%, Asian 4%, and all other racial/ethnic groups comprising 5%. In terms of socio-structural factors, the sample had a middle-class income of around $46,421. The sample included 33% participants with a college education or more, 28% with some college, 29% with a high school diploma, and 11% without a high school degree. About 89% of the participants had health insurance.

Table 1.

Summary characteristics

	Mean	SD	Min	Max
Demographics
Asian	0.04	0.20	0	1
Latinx	0.16	0.37	0	1
Black	0.12	0.32	0	1
White	0.63	0.48	0	1
Female	0.51	0.50	0	1
Age	47.62	17.46	21	80
Socio-economic factors
Education	2.83	1.00	1	4
Income	46421.65	36288.52	2041.24	230000
Insurance	0.89	0.31	0	1
Vaccination and individual factors
Previously vaccinated	0.41	0.49	0	1
Knowledge	3.13	1.32	1	5
Attitude time 1	2.96	0.94	1	4
Attitude time 2	2.95	0.96	1	4
Attitude time 3	3.00	0.96	1	4
Attitude time 4	3.04	0.98	1	4
Intention time 1	2.63	1.27	1	4
Intention time 2	1.98	1.08	1	4
Intention time 3	1.62	0.84	1	4
Intention time 4	1.51	0.78	1	4
Vaccinated time 2	0.10	0.36	0	1
Vaccinated time 3	0.20	0.49	0	1
Vaccinated time 4	0.05	0.35	0	1
Vaccinated time 5	0.14	0.22	0	1

Notes: Means are weighted to be nationally representative. Education is measured on a 4-point scale with (1 = no high school diploma, 4 = BA degree or more). Income is the ratio of income level by the square root of the household size. Knowledge is measured on a 4-point scale with (1= low knowledge to 5 = high knowledge). Attitude is measured on a 4-point scale with (1= very negative to 4 = very positive). Intention is measured on a 4-point scale with (1= not likely at all to 4 = very likely).

As shown, the sample was fairly knowledgeable about vaccines. Further, the sample had somewhat positive attitudes toward the influenza vaccine as well as moderately positive intentions to receive the influenza vaccine—both attitudes and intentions were rated above the midpoint of the scale. In line with reports from the CDC, only around half of the survey respondents received the influenza vaccine during the 2018–2019 influenza season—45% among adults over the age of 18 [2]. All in all, our large sample had adequate variance to consider the effects of income, education, and being insured.

All psychological variables (i.e., knowledge, attitudes, and intentions) were positively related to the socio-structural variables. Additionally, all individual and socio-structural variables were related to vaccination behavior (see Supplementary Appendix C for correlations).

Analyses were conducted to understand how attrition and missing data may have affected the results. First, using pairwise t-tests, we investigated whether covariates varied significantly during our study period. If the characteristics of the participants at the later timepoints were systematically different from the sample at the earlier timepoints, then our estimates might be confounded with these changes. However, no variable at later timepoints differed from the same variables measured at the first timepoint. We also investigated whether the proportions of vaccinated to unvaccinated people for those who completed all timepoints were significantly different from the vaccination rate of people who did not participate in all timepoints. These comparisons were reassuring, suggesting that participants who stayed in the study did not differ from those who dropped out. In addition, analyses for missing data led us to believe that data were missing completely at random (see Supplementary Appendix D for details).

Given that repeatedly measuring a variable can induce measurement sensitization, this study collected an additional sample of 1,005 participants who completed the questionnaire in March 2019 without participating in the panel study. Participants were recruited through the same panel and were comparable to those in the main study in terms of demographics. This control sample reported a 52% influenza vaccination rate, which can be compared to the 49% vaccination rate by March in the main study. These results suggest that the reported results are unlikely to be due to measurement sensitization.

Main Effects

Columns (1), (3), and (5) in Table 2 present the coefficients for hypothesized relations examining the mediational paths from knowledge, to attitudes, to intentions, to vaccination. Each set of analyses appear in the first, second, and third vertical panels of Table 2, respectively. All the models controlled for race/ethnicity and time trends. Including covariates did not influence the estimate of the main effects of interest but did increase the precision of the estimates.

Table 2.

Associations between knowledge and attitudes, attitudes and intentions, and intentions and vaccination

Panel A
	Knowledge → attitude		Attitude → intention		Intention → vaccination
	β	β	β	β	OR	OR
	(1)	(2)	(3)	(4)	(5)	(6)
	Base	Moderation	Base	Moderation	Base	Moderation
Focal	0.06***	0.07***	0.25***	0.25***	1.16*	0.87
Education	0.01	0.02	0.01	0.01	0.98	0.86
Income	0.20***	0.18***	0.25***	0.24***	4.23***	3.61**
Insurance	0.25***	0.24***	0.44***	0.44***	6.05***	4.89***
Focal × education		0.02**		0.04		1.22*
Focal × income		0.03		0.08		1.14
Focal × insurance		0.00		0.13*		1.32*
Panel B
	Knowledge → attitude		Attitude → intention		Intention → time to vaccination
	β	β	β	β	HR	HR
	(1)	(2)	(3)	(4)	(5)	(6)
	Base	Moderation	Base	Moderation	Base	Moderation
Focal	0.09***	0.09***	0.80***	0.59***	4.58***	3.60***
Education	0.01	0.02	0.01	0.01	0.96	0.89
Income	0.15***	0.12**	0.04	0.04	1.56***	1.26
Insurance	0.20***	0.21***	0.21***	0.28***	1.30**	1.02
Focal × education		0.04***		0.00		1.06
Focal × income		0.06*		0.09*		1.16
Focal × insurance		0.02*		0.16***		1.23*

Notes: Panel A presents the coefficients from the MSEM model that investigates the psychological pathways to vaccination decision and Panel B presents the coefficients from the model that investigates the psychological pathways to time to vaccination. The focal predictor is the predictor stated in the panel header. The first panel predicts attitude from knowledge. The second panel predicts intention from attitude. The third panel predicts vaccination from intention at the previous timepoint. β are standardized estimates. OR are odds ratios and HR are hazard ratios. Race/ethnicity and time trends were used as control variables.

^***p < .001,

^**p < .01,

*p < .05.

Knowledge, attitudes, and intentions were positively related to attitudes, intentions, and vaccination. Among socio-structural variables, insurance positively correlated with attitudes, intentions, and vaccination. Finally, the likelihood of vaccination and the path from intentions to vaccination decreased as the influenza season progressed, which is expected because influenza vaccination is recommended for the fall. However, attitudes became more positive over time. This model explained 26% of total variance in attitude, 67% of the variance in intentions, and 65% of the variance in vaccination (⁠$R_{attitude}^2=0.26,R_{intention}^2=0.67,R_{vaccine}^2=0.65$⁠).

Panel B of Table 2 presents the coefficients from the MSEM model that investigates the psychological pathways and moderation by socio-structural disadvantages affected when during the season people vaccinated. These results indicated that having health insurance increased hazard and thus reduced time to vaccination (Panel B in Table 2; Column (5)).

Socio-Structural Moderators

We next examined the moderation of paths from knowledge, to attitudes, to intentions, to vaccination, by socio-structural variables by including the interaction terms. Columns (2), (4), and (6) in Table 2 present these analyses. The goodness of fit indexes in Supplementary Appendix E indicate that the model with interaction terms fit well whereas the model without interaction terms fit poorly.

Knowledge to attitude path

Column (2) in Table 2 contains the MSEM results for the knowledge to attitude path. Among socioeconomic variables, education moderated the association between knowledge and attitudes, such that low education appeared to affect attitude formation (β =0.02, 95% CI [0.01,0.09], p < .01). To explicate the moderation effects of socio-structural factors, Figs. 1(a)–(c) present the relations between knowledge and attitude for different levels of socio-structural factors. Figure 1(b) shows that higher knowledge was less strongly associated with more positive attitudes at 1 standard deviation below the mean level of education than at 1 standard deviation above the mean level of education. An additional 2% of variance in attitude was explained by the interaction terms in the model (⁠$R_{attitude}^2=0.28$⁠).

Fig. 1.

Socioeconomic moderators of associations between knowledge and attitudes, attitudes and intentions, and intentions and vaccination.

Attitude to intention path

Column (4) in Table 2 contains the estimated coefficients for the attitudes to intention path. In contrast to the association between knowledge and attitude, not having insurance coverage moderated the relation between attitudes and intentions (β = 0.13, 95% CI [0.02,0.26], p < .05), suggesting that lack of insurance limits intention formation. Figures 1(d)–(f) graphically illustrate the moderation effect of the association between attitudes and intentions by socio-structural disadvantages. Similar to predictions in Table 2, Figure 1(d) shows that the lack of insurance coverage weakened the influence of attitudes on intentions. An additional 2% of variance in intention was explained by the inclusion of interaction terms (⁠$R_{intention}^2=0.69$⁠).

Intention to vaccination path

The last path was from intentions to actual vaccination (Column (6) of Table 2). The association between intentions and vaccination was moderated by health insurance (OR = 1.32, 95% CI [1.02,1.71], p < .05). The association between intentions and vaccination was also moderated by education (OR = 1.22, 95% CI [1.02,1.46], p < .05). Figure 1(g)–(i) graphically depict the moderation effects identified in the relation between intentions and vaccination. Specifically, having more or less positive intentions was weakly associated with vaccination among those who did not have health insurance, relative to those who had health insurance (Fig. 1(g)). In addition, having more or less positive intentions was weakly associated with vaccination at 1 standard deviation below the mean level of education than at 1 standard deviation above the mean level of education (Fig. 1(h)). The interaction terms explained 3% of the total variance in vaccination $(R_{vaccine}^2=0.68)$⁠.

Additional plots are found in Supplementary Appendix F which test the moderation of paths from knowledge, to attitudes, to intentions, to vaccination, by socio-structural variables as categorical factors. The results are replicated across analyses treating variables as continuous or categorical.

Similar to our findings on the odds ratio of vaccination, we hypothesized that socio-structural disadvantages may impede early movement from intentions to vaccination. These results are reported in Panel B of Table 2 and are graphically presented in Fig. 1(j)–(l). Consistent with our hypothesis, our results indicated that the lack of health insurance coverage was associated with a longer time to vaccination (HR = 1.23, 95% CI [1.02,1.48], p < .05).

Discussion

This study used a nationally-representative panel survey to test a theoretical model to understand factors leading to vaccination behavior during the 2018–2019 influenza season. A prospective approach was adopted to predict vaccination behavior from psychological and socio-structural factors throughout the influenza season. This study aimed to test a hypothetical model to investigate whether socio-structural factors moderated the associations between psychosocial predictors and vaccination. As predicted by theory and the extant literature, knowledge predicted attitudes, attitudes predicted intentions, and intentions predicted vaccination. Furthermore, emerging research on structural and social factors, such as socio-economic status and education, has highlighted their importance in predicting vaccination [62–64]. This work demonstrates how socio-structural factors influence vaccination via mediating pathways through knowledge, attitudes, and intention. In particular, structural barriers and lacking health insurance moderated the paths from knowledge to influenza vaccination, with lack of health insurance and lower educational attainment leading to less movement along the paths.A noteworthy issue is that knowledge alone likely does not elicit vaccination but instead must yield vaccination attitudes and intentions to influence actual vaccination. Accordingly, information that yields attitude change, rather than any factual information, is probably necessary to ultimately produce vaccination. Moreover, in the current media climate, stories of purported negative effects of vaccination are sensationalized, appealing to emotions rather than facts may be an appropriate route to change knowledge. Importantly, the knowledge-attitude link was moderated by education, which is consistent with a role for deliberation about vaccinations [65]. Specifically, knowledge influenced attitudes less when there is lower educational attainment because education presumably facilitates the acquisition and utilization of health information. Our results suggest that offering vaccine recommendations alone may be less effective among people with lower educational attainment. However, health care providers may increase vaccine acceptance among families with a low socioeconomic status by both recommending and offering a vaccine to ensure that financial barriers are effectively sorted out [66, 67].

Next, and not surprisingly, attitudes predicted intentions. Multiple meta-analyses and reviews have established both the correlational and causal effects of attitudes on intentions [13, 68, 69]. Attitudes likely act as a guide for forming intentions, such that if attitudes are positive, then subsequent deliberation will yield corresponding actions [69, 70]. Our work replicates this finding and extends it by testing theory with a nationally-representative sample, and increasing the generalizability of the important attitude-intention link within the influenza vaccination context. Although attitudes were strongly correlated with intentions, lacking health insurance moderated this association. Even when attitudes were positive, the lack of health insurance appears to reduce attitudinal influences on the intention to vaccinate. Importantly, this effect was seen after controlling for more general structural factors such as income and education, speaking to the practical importance of ensuring insurance coverage. One such solution is to continue to work to increase access to health insurance (such as the Affordable Care Act (ACA)) and preventive medicine to individuals across the United States. Although offering insurance coverage has a significant influence on intentions to uptake vaccines, it might further accelerate the formation of positive attitudes towards vaccines when combined with interventions such as reminders and other opportunities to evaluate and make decisions about vaccines.

Finally, in regard to the main pathways and in line with previous work, intentions predicted vaccination [71]. Our finding is prospective, specifically indicating that intentions at an earlier timepoint predicted vaccination in the following timepoint. Although other studies have shown this link, to the best of our knowledge, ours is the first to do so prospectively and with a representative U.S. sample and while accounting for socio-structural factors. However, as was the case with the other pathways to vaccination, the intention-vaccination link varied as a function of health insurance. Although intentions influenced vaccination, this effect was smaller when respondents lacked health insurance. As a preventive service under the ACA, the vaccine is available at no cost for individuals with insurance. However, uninsured individuals can face out-of-pocket costs for the vaccine. Evidence suggests that the federally-funded vaccine program, which provides vaccines at no cost to those who might not otherwise be vaccinated because of inability to pay, has contributed to higher vaccination rates [72]. Thus, providing vaccines free-of-charge is a potentially successful intervention to increase adult vaccination rates among adults with lower socioeconomic status. Together these findings suggest that accomplishing a high vaccination rate will require addressing potential disruption of the psychological process produced by social health disparities while also offsetting those disparities. Importantly, disruptions may accumulate and exert additive effects over time. For example, people who lack health insurance may have less positive attitudes, weaker intentions, and a lower probability of making the decision to vaccinate consequently. In this way, the effects of disparities may be difficult to reverse by simply providing access, particularly once people have formed attitudes toward vaccination.

Limitations and Conclusions

This research extends our knowledge of health behavior change in multiple ways. First, the use of a nationally representative U.S. sample based on probability sampling strengthens prior work on the determinants of vaccination. Second, the panel design over the entire 2018–2019 influenza season allowed us to see both within- and between-person factors that influence vaccination patterns. Third, our work identifies important socio-structural factors that affect each specific path to vaccination. Although our research has clear strengths and provides suggestions for future research, it has limitations. Whereas the current work focused on individual psychological factors and socio-structural factors that directly influence the individual, health is also determined by interpersonal- and societal-level forces. Among them, the relations between patient and provider can be important. If a patient does not trust the provider or the provider does not actively suggest vaccination, rates of vaccination are often suboptimal [73]. The current work did not focus on provider recommendations, but future work should incorporate the effectiveness of medical team efforts and behavioral nudges in tandem with the factors analyzed in the current work. Another concern to consider is the methods of measurement for vaccination. In line with organizations such as the CDC, all constructs in the current analysis were measured via self-report and are susceptible to biases such as selective memory and presentation concerns [74, 75]. Although our analyses do not suggest a measurement artifact, future research may consider objective measures for vaccination behavior. Another limitation is that, because insurance was assessed as a binary variable, differences in insurance type (e.g., private vs. Medicaid) might not have been captured in this study. Finally, the sample had limited variation in insurance coverage and about 89% had insurance, which could reduce the generalizability of our findings.

In conclusion, because a single path alone cannot produce vaccine uptake, each path should be the object of attention when crafting interventions. Intentions are important for behavior, but depend on attitudes, which in turn depend on knowledge. Interventions are likely to be more effective if they influence all paths and are structured to meet the social and economic needs of the individual. To design interventions, socio-structural factors should be considered, if not changed outright. To this end, new policy implementation to improve the psychological paths to healthy behaviors is important even though targeting knowledge, attitudes, and intentions must go on [76]. However, in the immediate future, targeting psychological paths (e.g., knowledge through intentions) will likely increase the probability of vaccination. All in all, the current model represents a succinct path to influenza vaccination and the socio-structural factors that should not be ignored in determining optimal interventions to increase the health of this nation. Understanding the psychological pathways to influenza vaccination provides some insight into the potential barriers to be addressed as part of COVID-19 vaccination efforts. Under federal law, the COVID-19 vaccine is available at no cost, but questions remain about whether providing access is sufficient to close the gap between uninsured and insured populations. Our model can be utilized to understand reasons for hesitancy should it arise nationally and paths that improve the opportunity to create herd immunity at this crucial time for the United States and the world.

Acknowledgments

The authors acknowledge 2018–2019 ASK Group at the Annenberg Public Policy Center for contributions to the survey design and implementation. The 2018–2019 ASK Group included Ozan Kuru, PhD, Dominik Stecula, PhD, Hang Lu, PhD, Yotam Ophir, PhD, Sally Chan, PhD, Ken Winneg, PhD, Kathleen Hall Jamieson, PhD, and Dolores Albarracín, PhD. Annenberg Public Policy Center, University of Pennsylvania. We thank Ken Winneg for coordinating the survey implementation.

Funding

Research reported in this publication was supported by the Science of Science Communication Endowment of the Annenberg Public Policy Center (APPC) at the University of Pennsylvania, the National Institute of Mental Health under Award Number R01MH114847, the National Institute on Drug Abuse under Award Number DP1 DA048570, and the National Institute of Allergy and Infectious Diseases under Award Number R01AI147487. The content is solely the responsibility of the authors and does not necessarily represent the official views of the National Institutes of Health.