https://orcid.org/0000-0003-2931-935X
Abstract
Amidst the global surge in healthcare expenditures, there is a growing political and academic debate about individual versus collective responsibility for health and healthcare costs. This study explores the causal effects of health risks and income cues on citizens’ attribution of responsibility for health outcomes and healthcare costs. An original vignette experiment was conducted among the German population. Those in need of medical treatment who have been exposed to environmental, institutional, or biological health risks are held less responsible for their health outcomes than those exhibiting behavioural health risks. However, the impact of health risks appears to be somewhat weaker in determining who should bear the costs for the medical treatment. Furthermore, higher-income groups are more likely to be blamed for their health outcomes and are considered less deserving of society’s help in paying for medical treatment. These findings enhance our understanding of how the criteria of control and economic need shape public attribution of responsibility in the health(care) domain.
Introduction
The expansion of healthcare systems in the last decades has contributed to rising life expectancy (Korda et al., 2007; Leung and Wang, 2010; Zarulli et al., 2021). Hence, it is not surprising that research finds strong public support for government responsibility for health (Mikler, 1999; Lundell, Niederdeppe, and Clarke, 2013; Traina, Martinussen, and Feiring, 2019) and healthcare (Gevers et al., 2000; Wendt et al., 2010; Missinne, Meuleman, and Bracke, 2013; Naumann, 2014; Jensen and Naumann, 2016; Azar et al., 2018; Burlacu and Roescu, 2021; Busemeyer, 2023) across countries. Meanwhile, healthcare systems provide healthcare for those in need and establish institutionalized solidarity between people with low and high health risk profiles (Maarse and Paulus, 2003; Jensen, 2011; Wendt, 2019; Prainsack and Van Hoyweghen, 2020). The pooling of risks and resources in healthcare systems distributes the costs of disease and medical care of individuals across members of society. However, along with a global surge in healthcare expenditures, the political and academic debate has shifted emphasis from collective to individual responsibility for health and healthcare costs (Trappenburg, 2000; Ter Meulen and Maarse, 2008; Brown, 2013; Laverty and Harris, 2018; Davies and Savulescu, 2019; Verweij and Dawson, 2019; De Marco, Douglas, and Savulescu, 2021).
Within a policy context that is challenged by a wide diversity of health risk factors (Rettenmaier and Wang, 2013; Bartley, 2016) and the persistence of social health disparities (Mackenbach, 2012; Kröger, Pakpahan, and Hoffmann, 2015; Lago et al., 2018; Barnes, Hall, and Taylor, 2023), it is important to understand how citizens attribute responsibility for health outcomes and healthcare costs. However, previous research on public responsibility attribution is constrained by three major shortcomings. First, while the deservingness literature has established that individuals who engage in risky health behaviour are perceived less deserving of collectively financed healthcare (Ubel et al., 2001; Wittenberg et al., 2003; Gollust and Lynch, 2011; Gollust, Lantz, and Ubel, 2013; Stegeman et al., 2014; Jensen and Petersen, 2017; Knotz et al., 2021; Reeskens, Roosma, and Wanders, 2021; Gandenberger et al., 2022; Schaeffer and Haderup Larsen, 2023), these studies fail to account for the diversity of health risks individuals may face over their life course. Most notably, the impact of non-behavioural risks such as environmental, institutional, and biological health risks has been largely overlooked or studied for one particular health problem at a time. Second, previous literature often assumes that lay attributions of health can be applied homogenously to different population groups (Robert et al., 2008; Lundell, Niederdeppe, and Clarke, 2013; Kahissay, Fenta, and Boon, 2017). Thereby, it fails to recognize that the exposure and vulnerability to and controllability of many health risks are socially stratified (Barosh et al., 2014; Diekmann et al., 2023). Given these social inequalities, citizens may attribute responsibility for health and healthcare costs differently for different socio-economic groups in society. Third, the current literature lacks a multi-dimensional perspective on public attribution for health outcomes and healthcare costs. Rather than studying these two components in isolation, an integrative framework would allow us to investigate whether they are driven by the same factors and thereby provide a more comprehensive understanding of healthcare deservingness.
To address these lacunae, this study investigates the causal effects of diverse health risks and income cues on public responsibility attribution for health outcomes and healthcare costs. To this end, we theoretically and empirically distinguish behavioural, environmental, institutional, and biological health risks1 as well as low, medium, and high-income levels. We, thereby, differentiate in much greater detail between diverse health risks and income groups compared with previous literature. In addition, by studying their causal effects on responsibility attribution for health outcomes as well as healthcare costs, the study speaks to major ongoing debates surrounding health inequality and healthcare financing (Mackenbach, 2012; Marmot, 2015; Wendt, 2019; Barnes, Hall and Taylor, 2023). To empirically test our hypotheses, we designed a novel vignette survey experiment that was fielded among a large sample of the German population. The vignette survey experiment allows us to make causal claims about the effects of health risks and income on responsibility attribution while reducing social desirability bias. Our results reveal that responsibility attribution is conditional on the nature of health risks to which those in need of medical treatment have been exposed, as well as on their income position. However, the causal effects of health risk cues vary across different dimensions of responsibility attribution. Attitudes towards healthcare costs are less strongly affected by the nature of health risks, signalling a certain degree of risk solidarity among the population. By providing a comprehensive framework that considers the diversity of health risks as well as the social gradient in health, this study advances our understanding of public attitudes towards collective versus individual responsibility for health and who deserves society’s help in paying the costs of medical treatment.
Health attribution: behavioural, environmental, institutional, and biological risks
Previous research that investigated lay attributions of health and illness has identified a variety of beliefs of ill-health causation (Hilbert, Rief, and Braehler, 2007; Robert et al., 2008; Leikas et al., 2009; Gollust and Lynch, 2011; Lundell, Niederdeppe, and Clarke, 2013; Ramírez et al., 2013; Kahissay, Fenta, and Boon, 2017). Building on this literature, we differentiate four distinct health risks, corresponding to behavioural, environmental, institutional, and biological explanatory models for health disparities. Behavioural attribution narrows down to the idea that health problems result from damaging health behaviours or lifestyle factors. Medical research shows that healthy lifestyle factors are associated with substantial gains in life expectancy (Sun et al., 2022). The assumption behind behavioural health attribution is that health largely lies within one’s control. Hence, it corresponds with the belief that the sick could have obtained better health outcomes if they had made the right behavioural health choices. For example, behaviour related to dietary choices, alcohol consumption, smoking, physical exercise, and sun protective measures are all considered individual actions that are within one’s control and are found to determine one’s health outcomes and life expectancy (Armstrong and Kricker, 2001; Memon et al., 2021; Sun et al., 2022).
In contrast, environmental attribution explains health outcomes by exposure to the physical environmental risks where people live or work.2 Workplaces and residential areas can present hazards in the form of environmental pollutants (e.g. toxic and microbial agents), chemical contaminants of the food and water supply, or structural hazards (e.g. roadway design and worksite conditions). For example, exposure to air pollution is associated with an increased risk of respiratory and cardiovascular diseases (Pražnikar and Pražnikar, 2012); air pollution was responsible for 2.4 million lost years of life and 237,810 premature deaths in the EU27 in 2020 (Eurostat, 2023). Furthermore, physical environment factors (such as accessibility, opportunities, and aesthetic attributes) are associated with physical activity behaviour (Humpel, Owen, and Leslie, 2002), while residence in green areas is associated with decreased rates for all-cause and circulatory mortality (Mitchell and Popham, 2008; Rigolon et al., 2021).
Institutional attribution assumes that health is shaped by individuals’ medical care experiences related to the institutional structures and provisions of the healthcare system (Wendt, 2009; Beckfield et al., 2015). Medical care does not often determine whether a person is exposed to a particular risk factor nor affect one’s vulnerability or resilience in the face of that risk (except for vaccination). Nevertheless, medical prevention and treatment significantly predict population health (McGinnis, Williams-Russo, and Knickman,, 2002; House, 2016). A wide body of research has discussed the unequal access to (and quality of) preventive and curative healthcare (Fiscella et al., 2000; Schwierz et al., 2011; Dickman, Himmelstein, and Woolhandler, 2017; Lee et al., 2019). Access and quality of medical care determine whether and when a disease is detected as well as a patient’s treatment and prognosis. Hence, healthcare system deficiencies can do harm. Timely, error-free treatment and early diagnosis are therefore important to monitor health risks and slow down the progress of a disease.
Finally, biological attribution ascribes health outcomes to genetic predisposition and inborn characteristics (Brand, Brand, and Schulte in den Bäumen, 2008). Genetic predisposition implies that one’s predisposition to health or disease is embedded in our genetic blueprint that takes form at conception. Meanwhile, inborn characteristics refer to the occurrence of inborn variants of genetic codes that confer disadvantage (McGinnis, Williams-Russo, and Knickman, 2002: p. 80). In short, the biological attribution builds on the idea that human health is largely determined at birth due to either genes or inborn characteristics that determine the likelihood of developing certain diseases over the life course and the biological limits of our life expectancy.
Health risks, income, and responsibility attribution
One key question is how behavioural, environmental, institutional, and biological health risk factors shape responsibility attribution for health outcomes and healthcare costs. Our theoretical expectations draw on what we define as risk controllability, namely the extent to which health risks can be reduced or avoided by individual agency. Behavioural, environmental, institutional, and biological health risks significantly differ in risk controllability. We acknowledge that health behaviour is often not exclusively voluntary, especially among low-income or other disadvantaged groups that face major societal barriers to pursuing a healthy lifestyle. However, relative to environmental, institutional, and biological health risks, behavioural health risks generally have a higher level of controllability. As such, behavioural health risks can be mitigated through low-risk behaviour, although external factors such as socialization and group norms may impact health behaviour. At the other end of the spectrum, one could position biological risk such as hereditary predisposition for specific diseases because these are defined at birth and cannot be modified over the life course. Environmental and institutional health risks could be considered in between, as they are characterized by relatively low-risk controllability. For example, escaping exposure to harmful environmental conditions (such as air and water pollution where people live and work) requires substantial efforts such as relocation or seeking employment in a different industry sector. Finally, institutional risks in the healthcare system (such as medical errors in screening and treatments) are often hard if not impossible to prevent by the patient’s actions.
On the basis of this concept of risk controllability, one can expect that how citizens attribute responsibility for health and healthcare costs depends on the nature of the health risks to which a person has been exposed. Scholars have argued that healthcare systems involve both rights and obligations (Davies and Savulescu, 2019). While individuals operating in a solidarity system have claims on others and the system in which they operate, they also have obligations to others since solidarity is reciprocal. In this regard, Davies and Savulescu (2019) argue that those who avoidably incur health burdens violate the obligations of solidarity. Their individual choices may affect their health and consequently place additional costs on society. Treating an individual requires resources that could be spent elsewhere and may delay or limit treatment options for others who are unavoidably ill (Davies and Savulescu, 2019: p. 133).
Indeed, citizens tend to punish others for behavioural health risks. Surveys show that those who exhibit risky health behaviours are given lower priority for vaccinations, transplantable organs, and intensive care unit treatment (Ubel et al., 2001; Wittenberg et al., 2003; Knotz et al., 2021; Reeskens, Roosma, and Wanders, 2021; Gandenberger et al., 2022; Schaeffer and Haderup Larsen, 2023), are expected to pay higher contributions for their insurance and treatment costs (Gollust and Lynch, 2011; Stegeman et al., 2014), and are considered less deserving of state support such as disability benefits (Geiger, 2021). Furthermore, health services to prevent lifestyle-related diseases receive lower priority than those to prevent genetic disorders (Singh et al., 2012), suggesting that individual responsibility affects the public’s valuation of healthcare interventions. Additionally, believing that obesity results from sinful behaviour is associated with lower support for redistributive and compensatory policies to reduce obesity rates and stronger support for charging overweight or inactive people higher premiums (Barry, 2009). Overall, these empirical studies suggest that individual behaviour is an important consideration in citizens’ allocation preferences for medical treatment, and those who engage in unhealthy behaviour are perceived as less deserving of society’s help in paying medical costs.
In contrast, non-behavioural health risk cues are likely to evoke stronger solidarity since there is little to no individual responsibility in the exposure to such risks. Gollust (2013) found that survey participants who viewed an article on the genetic or environmental (social composition of the neighbourhood, food, and activity environment) determinants of diabetes were more likely to support increased government spending on research about diabetes prevention and treatment than those viewing an article with no causal language. Furthermore, people who attribute health differences to system factors such as prejudice, health system failure, or economic system failure are more likely to prefer stronger government involvement in healthcare (Gollust and Lynch, 2011). Other studies show that substantial shares of society perceive inequality in healthcare access and quality as unfair (Lynch and Gollust, 2010; Knesebeck, Vonneilich, and Kim, 2016; Immergut and Schneider, 2020) and that such perceptions of unfairness are associated with stronger support for government provision of health insurance (Lynch and Gollust, 2010). Similarly, empirical research finds that public trust in healthcare strongly matches the views of users about the quality of healthcare, including perceptions of whether doctors always make the right diagnosis, do enough tests and prescribe medicine on time, refer patients on time, and provide the right dose of medicine or best treatment (Calnan and Sanford, 2004). These studies further highlight the potential role of institutional factors in shaping solidarity in healthcare systems.
Since environmental, institutional, and biological health risks are characterized by a low level of risk controllability, we expect them to reduce the attribution of responsibility towards the individual for health outcomes as well as healthcare costs. What concerns accidents, previous research confirms that the extent to which a health risk is perceived as controllable determines the causal attribution of responsibility to the individual (Rickard, 2014). Applying this logic to the broader realm of health, we hypothesize:
H1: Environmental (H1a), institutional (H1b), and biological (H1c) health risk cues reduce the attribution of responsibility for health outcomes to the individual compared to behavioural health risk cues.
H2: Environmental (H2a), institutional (H2b), and biological (H2c) health risk cues reduce the attribution of responsibility for healthcare costs to the individual compared to behavioural health risk cues.
A second key question is how and to what extent economic need shapes responsibility attribution for health outcomes and healthcare costs. Davies and Savulescu (2019) argue that violating solidary obligations requires more than avoidability of health risks; violation requires that the risk taken is unreasonable and made under conditions conducive to sufficiently autonomous choice (i.e. conditions of decision-making that enable a well-considered, uncoerced choice). The latter aspect is critical because the presence of autonomous choice is not equally distributed across members of society (Ter Meulen and Maarse, 2008; Davies and Savulescu, 2019). For example, individuals with relatively higher incomes can more easily escape environmental health risks because they have resources and exit options to avoid polluted living and working environments or exchange them for healthier ones. Higher-income groups can also afford better health insurance, which fosters prevention and early diagnosis, and minimizes institutional health risks. Furthermore, they have less exposure to psychosocial stressors positively related to behavioural risk factors such as smoking, drinking, and unhealthy diets (Pampel, Krueger, and Denney, 2010; Moor et al., 2014). Hence, economic resources are an important factor in one’s ability to maintain a healthy lifestyle, not least regarding dietary choices and exposure to the natural environment (Barosh et al., 2014; Nesbitt et al., 2019). Conversely, lower-income groups are more likely to receive lower healthcare priority and service quality (Fiscella et al., 2000; Schwierz et al., 2011; Lee et al., 2019), increasing their exposure to institutional health risks. In summary, lower-income groups often do not possess a sufficiently autonomous choice to maintain good health. In this regard, previous research found that a person is blamed less for his health when portrayed as working class instead of middle class (Gollust and Lynch, 2011). Hence, one can expect that, ceteris paribus, higher-income groups are considered more to blame for their health problems.
Furthermore, we expect that economic need determines allocation preferences regarding collective and individual responsibility for healthcare costs. Healthcare systems entail redistribution between individuals with low and high health risk profiles but may also institutionalize redistribution between the rich and poor. Previous research shows that low-income earners are typically considered more deserving of welfare support (Heuer and Zimmermann, 2020; Reeskens and Van der Meer, 2019). When it comes to healthcare, van der Aa et al. (2017) found among the Dutch population that care-seekers with lower financial capacities are considered more deserving in the allocation of collectively financed healthcare resources. Hence, we hypothesize:
H3: Higher-income cues increase the attribution of responsibility for health outcomes to the individual compared to lower-income cues.
H4: Higher income cues increase the attribution of responsibility for healthcare costs to the individual compared to lower income cues.
Although health outcomes and healthcare costs are intricately linked, it remains unclear to what extent public attribution of responsibility towards both dimensions aligns and is shaped by the same criteria. Since previous research has studied these components in isolation of one another,3 we are interested in whether health risks (i.e. the control criterion) and income (i.e. the need criterion) have the same effect on how citizens attribute responsibility for ensuring health outcomes as well as managing healthcare costs, or whether they have differential effects. For instance, citizens’ support for sharing resources in the healthcare system may be driven by a wide variety of motivations that go much beyond the nature of health risks to which those in need of care have been exposed. The question of who deserves healthcare may namely evoke diverse rationales, related to political attitudes and redistributive preferences. In contrast, health risks may have stronger direct implications for the question of who is to blame for health outcomes. Following this logic, one could expect that health risk cues have stronger effects on responsibility attribution for health outcomes than on responsibility attribution for healthcare costs. Such a gap in the causal effects of health risks may signal a sense of risk solidarity among the population. In a similar vein, the effects of income may not necessarily align when one considers health outcomes or healthcare costs. Our study’s design allows us to explore whether health risk and income cues affect attitudes towards responsibility for health outcomes and healthcare costs to the same extent.
Data and methods
Experimental design
To test how health risk and income cues shape public attitudes towards responsibility for health outcomes and healthcare costs, we designed a factorial survey experiment (Auspurg and Hinz, 2015). Respondents were asked to form judgements about descriptions of hypothetical persons (vignettes) with various attributes (dimensions). Each vignette described a fictive person in need of medical treatment and included an information cue about the nature of the health risk, the health problem, and the income level of the fictive person. Table 1 provides an overview of the vignette dimensions and levels. Vignette survey experiments have various advantages over item-based survey questions. The values (levels) of the dimensions are randomized across the different vignettes and allow us to assess the causal effects of each dimension on respondents’ judgments. Because of the multidimensionality of the evaluation task, vignette experiments reduce social desirability bias compared with direct questioning techniques, which are less likely to produce reliable results with sensitive topics such as health. Supplementary Table A1 provides an overview of the vignette dimensions and levels, and Supplementary Figure A3 shows an example of a vignette. Below, we describe the three dimensions that were randomized in the vignettes.
Vignette dimensions and levels.
| Dimension | Level |
|---|---|
| Health riska | Behavioural: diet/sun protection/physical exercise |
| Environmental: air pollution/sun exposure/green space | |
| Institutional: diagnosis failure | |
| Biological: family medical history | |
| Income | €1,400 |
| €2,100 | |
| €3,200 | |
| Health problem | Heart disease |
| Skin cancer | |
| Diabetes |
| Dimension | Level |
|---|---|
| Health riska | Behavioural: diet/sun protection/physical exercise |
| Environmental: air pollution/sun exposure/green space | |
| Institutional: diagnosis failure | |
| Biological: family medical history | |
| Income | €1,400 |
| €2,100 | |
| €3,200 | |
| Health problem | Heart disease |
| Skin cancer | |
| Diabetes |
aFull wording is provided in Supplementary Table A1.
Vignette dimensions and levels.
| Dimension | Level |
|---|---|
| Health riska | Behavioural: diet/sun protection/physical exercise |
| Environmental: air pollution/sun exposure/green space | |
| Institutional: diagnosis failure | |
| Biological: family medical history | |
| Income | €1,400 |
| €2,100 | |
| €3,200 | |
| Health problem | Heart disease |
| Skin cancer | |
| Diabetes |
| Dimension | Level |
|---|---|
| Health riska | Behavioural: diet/sun protection/physical exercise |
| Environmental: air pollution/sun exposure/green space | |
| Institutional: diagnosis failure | |
| Biological: family medical history | |
| Income | €1,400 |
| €2,100 | |
| €3,200 | |
| Health problem | Heart disease |
| Skin cancer | |
| Diabetes |
aFull wording is provided in Supplementary Table A1.
First, the information cue about the health risk of the fictive person was either (i) behavioural, (ii) biological, (iii) environmental, or (iv) institutional. Note that these cues present respondents with information about the nature of the health risk to which a person was exposed rather than with factual statements about the cause of their health problems. The behavioural information cues refer to health-damaging individual behaviour. In contrast, the biological cues introduce a genetic predisposition for a health condition, while the environmental cues tap into health risks imposed on the individual from the physical environment. Finally, the institutional cues contain a diagnostic failure by the healthcare system to diagnose the first symptoms. We acknowledge that in the institutional scenario, a causal attribution of the health problem is less straightforward since a late diagnosis does not explain the very existence of a health problem. However, institutional risks may, nevertheless, worsen a person’s health status and subsequently increase the costs of medical treatment.
Second, the net monthly income of the fictive person in the vignette varies between three income levels, namely (i) €1,400, (ii) €2,100, and (iii) €3,200. These levels roughly correspond with the first quintile, the median, and fifth quintile of net equivalent incomes of the German population, respectively (Eurostat data code ilc_di01).4 By differentiating between a low-, medium-, and high-income level, we provide a more fine-grained test of financial capacity (i.e. the need criterion of deservingness) compared with previous research (Gollust and Lynch, 2011).
Third, the corresponding health problem in each vignette varies between (i) heart disease, (ii) skin cancer, and (iii) diabetes. We consider the inclusion of these three health problems as a robustness check for the generalizability of our findings because we are interested in general patterns of responsibility attribution rather than attitudes towards specific health problems. Previous studies often focus on only one health problem (such as obesity, diabetes, or heart disease in isolation) and do therefore not allow to test the generalizability of the findings across diverse health problems. Since the three health problems selected in our survey vignette differ in the extent to which they are perceived as severe (El-Toukhy, 2015) and perceived as problems that the government should handle (Jensen and Petersen, 2017), we believe that are suitable cases to test our hypotheses.5
The vignettes introduce respondents to a person with a mix of the three dimensions. Furthermore, each vignette has two fixed parameters. Following previous research (Gollust and Lynch, 2011), the fictive person is fixed to a 40-year-old man. Providing information about gender and age ensures that respondents are not making an assessment based on different assumptions about these attributes.
The experimental design allows us to assess the causal effects of the information cues compared with one another, thus giving insight into their relative effect sizes. The randomization of conditions over the survey vignettes ensures that the weight of a specific attribute can be tested. Moreover, by integrating different dimensions and levels, we are able to study how the effects of the health risk cues compare and interact with the effects of income cues. In summary, the experimental design tests how distinct health risk cues (i.e. behavioural, environmental, institutional, and biological) play out in combination with low-, medium-, and high-income levels and to what extent the effects are robust across different health problems.
Following each vignette, respondents were asked two questions that capture attitudes towards responsibility for health outcomes and healthcare costs. First, they were instructed to think about who is responsible for the fictive person’s health status:
Some people might think that this person is to blame for his own health problems. Others might think that this person is not to blame at all. What is your spontaneous impression about it?
Answers were measured on an 11-point scale ranging from ‘This person is not at all to blame’ (0) to ‘This person is completely to blame’ (10). This formulation of responsibility for health outcomes rather than for the disease itself allows for meaningful attribution to all four health risks, including the institutional scenarios. Second, respondents were instructed to express their opinion on who should be responsible for the fictive person’s treatment costs:
Some people feel that in a fair society, this person should pay for the entire cost of his medical treatment. Others think that in a fair society, all treatment costs of this person should be covered by the society through taxes or insurance contributions. What is your opinion?
Answers were measured on an 11-point scale ranging from ‘Society should pay all the costs’ (0) to ‘The person should pay all the costs’ (10). Higher scores indicate a stronger attribution of responsibility to the individual for health outcomes and healthcare costs.
Respondents were presented with two different descriptions of the 36 (4 health risk types × 3 income levels × 3 health problems) unique scenarios. The randomization was performed without replacement to ensure no respondent evaluated the same vignette twice. We confirmed that randomization was successful by comparing the demographic composition of the groups assigned to the different scenarios. Supplementary Table A5 indicates that correlations between the vignette levels and respondent’s characteristics are negligible. Prior to the main study, the survey experiment was tested in a qualitative pre-test based on 20 interviews and was part of a quantitative pre-test with approximately 350 completed questionnaires.
Respondent sample
The experimental design required a large sample of respondents and was embedded in the Inequality Barometer organized by the University of Konstanz. This survey was fielded between 14 November 2022 and 2 December 2022 among a sample of respondents aged 18 and over in Germany. Respondents were drawn from an online access panel collected by Kantar. Quotas for age, gender, education, and region (with cross-quota for age, gender, and education) ensured that the sample was representative of the population’s demography.6 A detailed overview of the quota targets and actual sample by gender, age, education, and region is provided in Supplementary Tables A3 and A4. Prior to the survey questions, respondents were informed about the objectives of the study, and informed consent was obtained (Supplementary Figures A1 and A2). The anonymization of the survey data was processed by Payback, which scripted and hosted the survey. The panellists were actively recruited among all 31 million Payback members, one of the largest card communities in Germany. The panel management has information about the age, gender, education, and region of all their panellists and therefore can select suitable respondents who meet the specifications of a stratified random sample. For each respondent who participated in the survey, a respondent ID was generated that is attached to the survey answers. The survey data including these respondent ID’s is stored on separate servers of the Payback online panel without any link to personal data such as the email address of the respondent. Neither Kantar nor the research team at the University of Konstanz had access to any personal data. A total of 6,319 interviews were conducted. However, 301 respondents were removed from the dataset based on Kantar’s quality checks for interview duration and item non-response.7 We additionally removed observations with missing values on at least one of the dependent variables per vignette to ensure equal sample sizes for model comparison. These procedures resulted in a net sample size of 11,712 observations among 5,915 respondents and a minimum of 317 observations per unique vignette scenario. Descriptive statistics of the sample are presented in Supplementary Table A2.
Estimation strategy
The analysis tested our hypotheses that health risk and income cues affect respondents’ attitudes towards responsibility for health outcomes and healthcare costs. We performed ordinary least square (OLS) regression models with cluster-robust standard errors to account for the non-independence of outcomes from the same respondent. We discuss the main effects based on Models 2 in Supplementary Tables A6 and A7. Furthermore, we tested whether the effects of the health risk, income, and health problem cues are conditional on one another by interacting with them in the models (Supplementary Table A9). In addition to these analyses, we report a number of further analyses in Supplementary Appendix. First, we tested for heterogeneous treatment effects by respondent’s income because income is found to be associated with health outcomes (Mackenbach et al., 2019) as well as health policy preferences (Immergut and Schneider 2020). To test if the causal effects of the information cues differ significantly across respondents with different income levels, we report split sample analyses by respondent’s income (Supplementary Figure A5). Similarly, we test for heterogeneous treatment effects by respondent’s health insurance type (Supplementary Figure A6), since public and private health insurance systems are operating on somewhat different logics. Second, we provide a series of models with and without control variables (Supplementary Tables A6 and A7). The results are robust to controlling for the first vignette that respondents were presented with. We additionally controlled for respondent characteristics, including education, net equivalent household income, employment status (employed, retired, other), migration background, health insurance (public with or without top-ups, private), subjective health, region, children, age, and gender. The inclusion of these individual-level control variables did not affect our findings, as is to be expected from our randomization. All analyses were performed using STATA, Version 18.
Results
Effects of health risks and income on responsibility attribution
Figure 1 presents the predictive margins of health risk cues on attitudes towards responsibility for health outcomes in Panel A and for healthcare costs in Panel B. The dashed lines present the mean values for individual responsibility for health outcomes (3.58) and healthcare costs (3.32) in the pooled sample. Panel A indicates that the perceived individual responsibility was substantively higher in behavioural health risk scenarios (5.35) than in environmental (3.37), institutional (2.79), and biological (2.80) health risk scenarios. Furthermore, Panel B indicates that the perceived individual responsibility for healthcare costs was also highest for behavioural scenarios (4.22), followed by environmental (3.12), institutional (2.95), and lastly biological (2.97) scenarios. While both measures follow the same trend, these descriptive results indicate that the gap in responsibility attribution between behavioural and non-behavioural cues is larger when it comes to health outcomes compared with healthcare costs. Overall, respondents leaned towards collective responsibility attribution for healthcare costs, even in the presence of behavioural risk cues. This signals that a certain level of (behavioural) risk solidarity prevails within society.
Predictive margins of health risk cues on attitudes towards individual responsibility for (A) health outcomes and (B) healthcare costs (n = 11,712). Note: Horizontal bars indicate 95 per cent confidence intervals. Horizontal dashed lines indicate the mean.
Figure 2 presents the effects of the information cues of the health risk, health problem, and income on the perceived responsibility for health outcomes (black circles) and healthcare costs (grey diamonds). The results confirm our hypothesis that, ceteris paribus, cues about environmental, institutional, and biological health risks reduce the likelihood that respondents attribute blame for health outcomes to the individual, compared to the behavioural health attribution cues (H1). Respondents judged the responsibility of the individual 2.55 points lower on the 11-point scale when this person was exposed to biological health risks compared with behavioural risks. Similarly, environmental health risk cues reduced individual responsibility attribution for health outcomes by 1.98 points on the 11-point scale. This finding confirms that respondents believe environmental exposure at work or home influences one’s health outcomes. Furthermore, institutional health risks reduce individual responsibility attribution for health outcomes by 2.55 points; thus, people seem to acknowledge that the quality of health services and related medical errors by health professionals—for which a patient cannot be held responsible—can affect health outcomes.
Marginal effects of information cues on attitudes towards individual responsibility for health outcomes and healthcare costs. Note: Horizontal bars indicate 95 per cent confidence intervals. The points without bars denote the reference category for each dimension.
Alternative model specifications altering the reference category confirm that the negative effects of the institutional (H1b) and biological (H1c) health risk cues (compared with behavioural risks) are stronger than those of the environmental health risk (H1a).8 These findings suggest that the public perceives behavioural health risks as the most controllable, environmental health risks as less controllable than biological, and institutional and biological risks as the least controllable of all health risks considered. Interestingly, the effect sizes of the institutional and biological cues (relative to behavioural) are not statistically different from one another.
Next, we explored the perceived responsibility for healthcare costs (Figure 2, grey diamonds). Environmental, institutional, and biological cues all reduce attributed responsibility for healthcare costs to the individual compared with behavioural health risk cues (confirming H2). More specifically, responsibility attribution was 1.09 points lower on the 11-point scale in the scenarios with environmental cues, 1.27 points lower in scenarios with institutional cues, and 1.25 points lower in scenarios with biological cues. In other words, all three non-behavioural health risk cues made people more likely to think that the costs of medical treatment should be covered by society instead of the individual. The environmental health risk cue had the weakest effect size of the non-behavioural health risk cues.9 This could signal that environmental risk factors are perceived as relatively more controllable than institutional and biological risk factors.
The OLS models show that health risk factors had a stronger effect on attitudes towards responsibility for health outcomes than their associated healthcare costs. This difference in effect sizes is consistent across the different types of health risks included in the experiment. The exposure to various health risk factors was strongly decisive for the extent to which citizens blamed individuals for their health outcomes. However, risk factors are weaker predictors of solidarity in the healthcare system. This discrepancy is also reflected in the explained variance, which amounts to 18.5 per cent for the model predicting the attribution of responsibility for health outcomes compared with only 5.5 per cent for the model predicting the attribution of responsibility for healthcare costs (Supplementary Tables A6 and A7). In summary, public judgements about responsibility for healthcare costs appear to be less sensitive to health risk factors. This finding indicates that attitudes towards the responsibility of health outcomes and healthcare costs are distinct theoretical constructs.
Figure 2 further illustrates how the income level of the fictive person influences how respondents attribute responsibility for health outcomes and treatment costs. Respondents were significantly more likely to think that the individual is to blame for his health outcomes when he has an average or high income compared with a low income (confirming H3). Ratings for individual responsibility were 0.25 points higher on the 11-point scale for average income (€2,100) and 0.33 points higher for high income (€3,200) compared with low income (€1,400). However, the effect size difference between average and high income was statistically insignificant (Supplementary Table A8). Overall, these findings suggest that people are aware that the ability to take personal responsibility for one’s health is unequal across society. Respondents might have considered that one’s capacity to maintain health is threatened when one possesses few economic resources.
While medium- and higher-income individuals are thus held more responsible for their health outcomes, Figure 2 indicates that respondents attributed more responsibility for the healthcare costs to a medium (0.26 points higher on the 11-point scale) or high-income earner (0.47 points) than a person with a low or average income (confirming H4). Additional tests confirmed that the difference between average and high income was statistically significant, providing evidence of a gradual effect of economic need on perceived responsibility for healthcare costs (Supplementary Table A8). Altogether these findings confirm that the depiction of a low-income person increases perceptions of deservingness associated with financial need for assistance.
We simultaneously tested the impact of cues about different health problems on the attribution of responsibility for health outcomes and healthcare costs. Heart disease served as a baseline scenario to compare the effect of skin cancer and diabetes (Figure 2). Ceteris paribus, persons with diabetes were held slightly more responsible for their health outcomes (0.27 points on the 11-point scale) and treatment costs (0.19 points on the 11-point scale) compared with those suffering from heart disease (Figure 2). People who developed skin cancer were, ceteris paribus, considered slightly more responsible for their health outcomes (0.18 points on the 11-point scale) but not for their treatment costs. These findings indicate that not all health problems are perceived as equal and that for some one is more likely to share the financial burden of medical treatment. However, the effect sizes of the health problems remain very small compared with the effect sizes of the health risk cues.
Robustness checks
We tested whether the effects of the health attribution and income cues held across different health problems. To this end, we performed split sample analyses (Figure 3) and interaction models (Supplementary Table A9). Figure 3 shows that the effects of health risk cues were generally more moderate in diabetes than in skin cancer and heart disease scenarios. Interaction models confirm that the effects of all non-behavioural cues (compared with the behavioural baseline) are significantly weaker in the diabetes scenarios. In particular, the environmental cue had a relatively weak effect in the diabetes vignettes. This difference in effect size could have been related to the perceived impact and severity of the environmental risk. The diabetes environmental health risk scenario referred to the lack of green space to exercise. The absence of green recreational space may not have been considered a legitimate excuse for neglecting physical exercise or not considered impactful on the risk of developing diabetes. In contrast, the environmental risks that were presented in the scenario of heart disease (i.e. living in a neighbourhood with high air pollution) and skin cancer (i.e. excessive exposure to UV radiation during work hours) may have been perceived as more severe threats to maintain good health (compared with their behavioural baseline of unhealthy dietary choices and unprotected skin exposure, respectively). The weaker effects of the environmental health risk cues in the pooled sample (Figure 2) were thus largely driven by the diabetes scenarios. Finally, Figure 3 also shows that the income effects operate similarly across the different health problems. Split sample analyses by income are provided Supplementary Figure A4 and provide further evidence for the robustness of our findings.
Marginal effects of health risk cues and income cues on attitudes towards individual responsibility for health outcomes and healthcare costs by health problem.
Furthermore, we are interested in whether the effects of the health risk cues varied significantly with the income of the fictive person. To facilitate the interpretation of the interaction model (Supplementary Table A9), Figure 4 displays the predictive margins of the health risk cues on responsibility attribution by income. Panel A shows that low-income persons with risky health behaviour and exposure to environmental risk factors were held slightly less responsible for their health status than high-income persons with the same health risks. For example, a low-income person exhibiting risky health behaviour is on average rated 0.47 points less responsible for his health outcomes (on the 11-point scale) than a high-income person exhibiting risky health behaviour. Similarly, a low-income person exposed to environmental risks is evaluated as 0.56 points less responsible for his health status compared with a high-income person. Although this pattern is not found for the institutional and biological risks, the differences are too small to be substantively meaningful. Furthermore, Panel B illustrates that the effects of the health risk cues on responsibility attribution for healthcare costs operate very similarly across different income levels. We observe that, in the behavioural and environmental risk scenarios, the differences in responsibility attribution are somewhat more conditional on the income level of the fictive person, but these differences are again only marginal and not statistically significant. For example, the extent to which citizens attribute responsibility for healthcare costs to the low-income sick who exhibited risky health behaviour compared with the high-income sick exhibiting such behaviour differs only by 0.45 points on the 11-point scale. Overall, these findings provide no conclusive evidence for the thesis that responsibility attribution for health outcomes and healthcare costs in the face of certain health risks depends on the income level of this hypothetical person. Although lower-income people often have fewer exit options and possibilities to escape behavioural and environmental risks, citizens may not be fully aware of it or do not give strong weight to it when attributing responsibility for health outcomes and subsequent healthcare costs.
Predictive margins of health risk cues on individual responsibility for (A) health outcomes and (B) healthcare costs by income level of the fictive person.
Conclusion
The contribution of this article is threefold. First, while previous studies on healthcare deservingness have largely failed to acknowledge the diversity of health risk factors, our study clarified how diverse health risks shape responsibility attribution for health outcomes and healthcare costs. Our vignette experiment fielded in Germany reveals that environmental, institutional, or biological health risk cues made citizens less inclined to attribute responsibility for health outcomes to the individual, compared with behavioural risk cues. In addition, environmental, institutional, and biological health risk cues increased the belief that the costs for medical treatment should be covered entirely by society compared with behavioural health risk cues. These causal effects were found across different health problems (heart disease, skin cancer, diabetes) and income levels (low, medium, high) in the vignettes. However, not all non-behavioural health risks are treated equally; the relatively weaker effects of the environmental health risks suggest they are perceived as more controllable than institutional and biological risks.
Second, the survey experiment revealed that persons with a high income were blamed somewhat more strongly for their health outcomes and were deemed more responsible for covering all the costs of medical treatment themselves, compared with those with a low income. These findings suggest that citizens are aware that certain health risks are non-random but related to socio-economic status. Taking into account that economic resources increase agency over one’s health status, citizens may reason that it is unfair to hold poor and rich members of society accountable to the same extent for their health outcomes and subsequent healthcare costs. Given that a flat rate for insurance contributions applies in the German public health insurance system, this finding points towards support for stronger redistribution between rich and poor individuals within the healthcare system. Overall, these results confirm that besides control, (economic) need is a relevant deservingness criterion that citizens employ when attributing responsibility in the healthcare domain (Gollust and Lynch, 2011; van der Aa et al., 2017).
Third, integrating two dimensions of responsibility attribution—health outcomes and healthcare costs—allowed us to observe differential patterns and better understand the nature and intensity of health solidarity. More specifically, we found that the nature of health risks was more important in explaining health attribution than in explaining policy preferences. Consider the behavioural health risks, which were measured in this experiment by unhealthy dietary choices, unprotected sun exposure, and a lack of physical activity. While respondents are more likely to blame the sick for their health status if they exhibited any of these risk behaviours, these risks matter much less when it comes to who should cover the financial costs of medical treatment. Even in the presence of behavioural risks, respondents on average prefer that society covers most of the medical treatment costs of healthcare beneficiaries. This finding signals a certain level of (behavioural) risk solidarity in society and is consistent with previous research, which finds that respondents remain overwhelmingly supportive of government aid for the sick even in the face of explicit information that sick individuals are in control over their fate (Jensen and Petersen, 2017). The observed gap between outcome versus cost responsibility attribution could be due to a potential policy feedback effect of the prevailing healthcare system on citizens’ policy preferences. The German healthcare system is characterized by a high share of public funding, which might have fuelled strong preferences for collective responsibility in financing medical treatment costs. Healthcare policies are often universal in their coverage and typically provide medical care irrespective of the cause of the disease.
When interpreting the findings of this study, some limitations should be kept in mind. First, while the findings suggest that health risks with lower controllability trigger stronger perceptions of deservingness, the study did not include direct measures of the perceived controllability of the behavioural, environmental, institutional, and biological health risks. Future research could measure whether there is a universal ranking order in the perceived risk controllability for such diverse health risk factors. Second, the operationalization of the control criterion focussed on the presence of a health risk rather than a causal attribution of the health problem of the fictive person. As a result, the effects may have been even stronger if we used direct causal language between the risks and the health problems. However, as in real-life situations, health problems can often not be simply attributed to one specific cause, such as resulting from either the person’s diet, exposure to air pollution or genetics. Future research could explore how the combination of various health risks may shape citizens’ responsibility attribution. Third, our study is limited to the German population. Therefore, an important pathway for future research concerns a comparative perspective on how health risk factors and income cues shape attribution of responsibility for health and healthcare costs. Welfare states vary in the extent to which they mediate health risks (Bambra, 2011, 2013; Beckfield et al., 2015) and in the degree to which social disparities in health exist (Lago et al., 2018; Barnes, Hall, and Taylor, 2023). As a result of these diverging policy contexts, cross-national variation may exist in how health risks are perceived and the implications they hold for health solidarity among populations. For instance, one could expect that the gap between the attribution of responsibility for health outcomes and healthcare costs varies depending on the public–private mix of healthcare financing. Finally, our research design does not allow us to identify changes in responsibility attribution for health outcomes and healthcare costs over time. As many European healthcare systems are under strain due to increased healthcare expenditure and demographic ageing, debates about the deservingness of healthcare beneficiaries are likely to intensify. Longitudinal research could provide insight into whether the criteria of control and (economic) need are gaining or losing importance, relative to one another as well as towards other deservingness criteria that could not be included in the current study.
Sharon Baute is an Assistant Professor of Comparative Social Policy at the Department of Politics and Public Administration at the University of Konstanz and the Principal Investigator at the Cluster of Excellence ‘The Politics of Inequality’. Her research focuses on social policy, European integration, and international solidarity.
Luna Bellani is a lecturer at the Department of Economics at the University of Ulm and a research associate at the Cluster of Excellence ‘The Politics of Inequality’ at the University of Konstanz. Her research interests include intergenerational income and education mobility, the political economy of redistribution, and household and gender economics.
Acknowledgements
We gratefully acknowledge valuable feedback from Marius Busemeyer, Claudia Diehl, Nanna Lauritz Schönhage, Guido Schwerdt, the reviewers and editors of ESR, as well as contributors at the 2023 ASA Annual Meeting in Philadelphia and the CPE workshop at the University of Konstanz.
Author contributions
Sharon Baute (Conceptualization [lead], Formal analysis [lead], Investigation [lead], Methodology [lead], Project administration [equal], Visualization [lead], Writing—original draft [lead]), and Luna Bellani (Conceptualization [supporting], Formal analysis [supporting], Funding acquisition [lead], Methodology [supporting], Project administration [equal], Visualization [supporting])
Funding
This work was supported by the Deutsche Forschungsgemeinschaft (DFG) under the Excellence Strategy of the German federal and state governments—EXC-2035/1 – 390681379.
Data availability
The data underlying this article will be shared upon reasonable request to the corresponding author.
References
Footnotes
While we acknowledge other influences on health, such as education, gender, and social networks, we focus on behavioural, environmental, institutional, and biological risks as major determinants of health inequality identified in the literature (see Bartley, 2016 and Mackenbach, 2019 for overviews).
The environmental attribution in this study focussed on the natural environment and not the food (i.e. high-calorie food advertisements, unhealthy food availability, healthy food affordability) or social (i.e. educational, family, and community) environments.
See Gollust and Lynch (2011) for an exception.
The exact income levels amounted €1,360.67 (20 per cent lowest income), €2,084.58 (median income), and €3,171.33 (20 per cent highest income). The different income levels in the vignettes were consistently rounded up to the first full hundred to account for the fact that the latest available Eurostat figures (accessed on 8 September 2022) may slightly underestimate real income figures at the time of the survey fieldwork.
El-Toukhy (2015) finds that diabetes is perceived as less severe than skin cancer, which is in turn perceived as less severe than cardiac problems (including cardiac arrest, heart failure and stroke). Jensen and Petersen (2017) show that diabetes is viewed as less deserving of government help compared with cardiac problems, while cancer is perceived as most deserving of government help.
The targets were reached in all groups except for two, with smaller than 5 per cent between the target and the actual sample.
Respondents who completed the survey in less than 40 per cent of the median duration and those with more than 30 per cent item non-response were deleted.
An additional model was estimated using environmental risk cues as a reference category instead of the behavioural risk cue (Supplementary Table A8). The results confirm that the institutional and biological health risk cues significantly reduce the attribution of responsibility to the individual compared to the environmental health risk cue.
Alternative model specification using the environmental risk cue as a reference category confirmed that the institutional and biological health risk cues significantly reduce the attribution of responsibility to the individual compared with the environmental health risk cue (Supplementary Table A8).
