The ex ante mechanism: subjective risk beliefs

Because individual’s true health risks are latent factors determined by genetics and health endowment (Smith et al., Reference Smith, Taylor, Sloan, Johnson and Desvousges2001; Clark and Etilé, Reference Clark and Etilé2002; Khwaja et al., Reference Khwaja, Sloan and Chung2006b), individuals must form subjective beliefs about their risks in order to make health-related decisions. For example, prospective smokers may classify themselves as relatively high or low health risk and then contemplate a trade-off between the discounted life-time utility from consuming tobacco and the downstream consequences that follow (i.e., health and financial losses).Footnote ¹

As described in Khwaja et al. (Reference Khwaja, Sloan and Chung2006b), weighing the costs and benefits of smoking is complicated by the uncertainty surrounding any personal consequences. A prospective smoker must re-weight the consequences of smoking according to her perceived likelihood of personally being afflicted with a smoking related illness. Those with beliefs of being relatively high-risk should be less likely to consume, which is implicitly assumed within the existing learning models for smokers.

The availability of health insurance can partly separate the smoking decision from current health risk beliefs and thereby allow a mix of risk types to use tobacco products. For example, full insurance eliminates the risks of future out-of-pocket (OOP) medical spending due to smoking. Thus, those choosing to smoke, who also presume having high intrinsic health risks, can then seek generous coverage, especially if premiums are exogenously set (e.g., community rating) and only vary with the level of financial risk protection sought. Absent liquidity constraints and perfect risk-rating by insurers, adverse selection can lead to greater coverage for high-risk smokers, who wish to simultaneously insure against their innate health risks and lifestyle choice – making market insurance and self-protection substitutes (Ehrlich and Becker, Reference Ehrlich and Becker1972). Removing the risks of financial consequences from the cost-benefit calculation also leaves only the difficult to quantify and perhaps incremental health consequences to offset the total expected utility from tobacco consumption (Koszegi and Szeidl, Reference Koszegi and Szeidl2013). Likewise, smokers confident in being lower-risk can forego purchasing generous coverage and divert the savings toward consumption of other goods.

Similar individuals, with divergent subjective risk beliefs, may be equally likely to consume tobacco but unlikely to select the same level of insurance – allowing differences in pre-shock coverage to partially reveal differences in baseline beliefs. Insofar as preferences over insurance products reflect priors about personal health risks, smokers selecting less financial risk protection ex ante could do the most updating ex post. Meanwhile, smokers who correctly presumed being high-risk (and accordingly purchased generous coverage) only have their priors reinforced after a health shock. This mechanism accordingly delivers two empirical predictions: (1) smokers with less coverage should be more likely to quit and (2) smokers with less coverage should have greater changes to their subjective beliefs following the shock.

The ex post mechanism: financial consequences

After a health shock, individuals with high financial risk exposure may experience substantive OOP medical expenditures – a common experience among older adults (Smith, Reference Smith1999; Meara et al., Reference Meara, White and Cutler2004; Paez et al., Reference Paez, Zhao and Hwang2009; De Nardi et al., Reference De Nardi, French and Jones2010; Babiarz et al., Reference Babiarz, Widdows and Yilmazer2013; Paccagnella et al., Reference Paccagnella, Rebba and Weber2013). Before the health shock, individuals are unlikely to know the type, scope and total costs of medical treatments made necessary by a given condition. Different shocks require different medical care, and the same shock can be treated differently across patients due to provider discretion and practice variance. Furthermore, the vector of prices belonging to all possible shock-and-treatment combinations cannot be known in the absence of perfect price transparency, which is likely rare (Anderson, Reference Anderson2007; Schoen et al., Reference Schoen, Osborn, Squires, Doty, Pierson and Applebaum2010). Whether cost-sharing varies by service type or is fixed across all services – as often the case (Chernew et al., Reference Chernew, Rosen and Fendrick2007) – the relevant prices and quantities of care are likely masked (and costly to know) before experiencing a specific illness. Owing to these particular features of medical care transactions, the financial consequences of poor health are learned after a shock requiring direct OOP expenditures – providing a second possible pathway between insurance coverage pre-shock and smoking behavior post-shock.

Importantly, acute health shocks, such as myocardial infraction, can happen again as well as signal higher risk for future disease burden. The individual is then faced with the prospect of a series of health shocks and/or a sharper decline in health, which may also necessitate further OOP payments. With this in mind, the realized financial burden from entering a worse health state in the present may add incentive to reduce tobacco consumption going forward, but generous coverage mutes the incentive via moral hazard. If this mechanism is in operation, we should empirically see that individuals with greater financial risk are (1) less likely to smoke after a health shock (2) bear higher OOP expenses following the shock and (3) have a stronger smoking response if financially constrained.

Sample selection

We use the final release of the 2004 and 2006/2007 waves of the Survey of Health, Ageing and Retirement in Europe (SHARE), which surveys households with at least one member aged 50 or older across 11 European countries (Austria, Belgium, Denmark, France, Germany, Greece, Italy, the Netherlands, Spain, Sweden and Switzerland). The survey collects rich information on a wide range of domains, including health, health care utilization, health behaviors and wealth. Detailed information on the survey methodology can be found in Börsch-Supan and Jürges (Reference Börsch-Supan and Jürges2005). We focus our analysis on individuals aged 50 and over interviewed in both waves with complete information on smoking and who do not report any change of insurance coverage between waves (n=15,507). This last restriction is important for our analysis as it ensures that financial risk exposure did not vary following, or in anticipation, of a severe health shock.Footnote ² We further restrict the sample to respondents with complete information on all variables included in the analysis and obtain an analytic sample of 13,017 adults (respondents with missing information are older, less educated and more likely to be female).

Outcomes

The survey collects information on current and past use of tobacco products. Our main outcome of interest is a binary variable that indicates whether or not the respondent currently smokes cigarettes. In additional analyses, we focus on smokers and use the average number of cigarettes smoked per day as an outcome to examine changes in smoking intensity. In order to explore underlying mechanisms of behavior change, we exploit two additional variables that measure OOP spending and subjective mortality risk. Specifically, we use respondents’ self-reported OOP spending for inpatient care, outpatient care and prescription drugs (expressed in Euro-PPP) and exploit a question on subjective probability of survival where respondents are asked to report the likelihood, on a 0–100 scale, that they will live to a specific age (the relevant ‘target’ age depends on respondent’s age at the time of interview).Footnote ³

Health shocks

The survey includes information on a wide range of severe health events and diagnoses and on respondent’s medical history. In this paper, we focus on acute cardiovascular (CV) conditions – myocardial infarction, severe cardiac conditions and stroke. We construct a CV health shock indicator that equals one if the respondent reports at least one of these conditions in the second wave that was not reported in the first wave as existing or having previously existed. We focus on CV shocks because of their high incidence among older adults, link to tobacco use, less anticipated arrival (i.e., making them more ‘shock-like’ than the onset of chronic disease), and likelihood of requiring immediate and intensive use of costly medical services. In addition, these acute shocks may have implications for the individuals’ planning horizons over the remaining years of life [e.g., mortality risk and subjective life expectancy (SLE)], which can then influence health-related behaviors; and as previously stated, these health shocks can reoccur that makes them different from the onset of a chronic illness and makes prevention efforts in the ex post period of value to the individual.

Financial risk exposure

The 11 countries included in our sample are typically a blend of public and private health insurance systems, which may overlap or supplement one another (Siciliani and Verzulli, Reference Siciliani and Verzulli2009). Although, Denmark, Italy, Spain and Sweden are predominantly tax-funded and publicly administered systems, which somewhat distances them from the other systems’ structures (Allin et al., Reference Allin, Bankauskaite, Dubois, Figueras, Golna, Grosse-Tebbe, Jemiai, McDaid, Riesberg, Schreyoegg and Thomson2005; Atella et al., Reference Atella, Brunetti and Maestas2012). A basic level of coverage is virtually universal within each country, and even Germany, which relatively recently mandated insurance for all its residents, had only a small portion of the population uninsured before its mandate (Docteur and Oxley, Reference Docteur and Oxley2003; Allin et al., Reference Allin, Bankauskaite, Dubois, Figueras, Golna, Grosse-Tebbe, Jemiai, McDaid, Riesberg, Schreyoegg and Thomson2005; Atella et al., Reference Atella, Brunetti and Maestas2012; Thomson et al., Reference Thomson, Osborn, Squires and Jun2013). Of note, Greece is an unusual case in that it has a publicly provided national health system, but at the same time, is characterized by considerable OOP spending burdens for its residents (Allin et al., Reference Allin, Bankauskaite, Dubois, Figueras, Golna, Grosse-Tebbe, Jemiai, McDaid, Riesberg, Schreyoegg and Thomson2005). As one explanation, a perverse custom of informal payments for care likely adds to the risks of personal medical spending in Greece and may account for the larger fraction of health care expenditures borne by individuals (Liaropoulos et al., Reference Liaropoulos, Siskou, Kaitelindou, Theodorou and Katostaras2008; Siskou et al., Reference Siskou, Kaitelindou, Papakonstantinou and Liaropoulos2008).

While not necessary in all countries, private insurance is required in Switzerland and is a key entity in Germany and the Netherlands. For the remaining SHARE countries, private insurance primarily functions to further reduce OOP expenses and/or extend the choice of providers and health care institutions for an enrollee (Docteur and Oxley, Reference Docteur and Oxley2003). Despite the rare absence of coverage within these countries, individuals generally face some level of OOP spending risk – a risk that has been ticking upward over time as many countries attempt to curb soaring health care expenditures (Atella et al., Reference Atella, Brunetti and Maestas2012). Cost-sharing often takes the form of set copayments for various services (e.g., provider visits and most commonly pharmaceuticals), but OOP expenses through coinsurance responsibilities are also possible (Allin et al., Reference Allin, Bankauskaite, Dubois, Figueras, Golna, Grosse-Tebbe, Jemiai, McDaid, Riesberg, Schreyoegg and Thomson2005; Thomson et al., Reference Thomson, Osborn, Squires and Jun2013). However, the bulk of these countries cap annual OOP payments for different components of care – providing a ceiling of sorts on personal financial risk. These system characteristics subsequently foster heterogeneity in OOP risk both across and importantly within countries due to individual differences in coverage attribute decisions and/or purchases of additional coverage. With these features in mind, we expect that SHARE respondents’ perceptions of personal financial risk exposure to medical spending are relative to their country of residence in terms of possible OOP costs and associated services.

Within the insurance portion of the SHARE survey instrument, questions ask if the respondent is ‘entirely,’ ‘mostly,’ ‘only somewhat’ or ‘not at all’ responsible for payments related to a list of medical services. We create a ‘high financial risk’ (HFR) indicator that equals ‘1’ if an individual reports being primarily (i.e., ‘entirely’ or ‘mostly’) responsible for medical spending related to at least one of these services: general practitioner visits, referred visits to specialists, public hospital care or prescription drugs, rather than a third party payer. We select this bundle of health services to reflect the medical care needs likely to follow the health shock during the acute and chronic phases as well as the varying potential OOP spending drivers facing individuals within these diverse health systems. Hence, the HFR variable is based on subjective perceptions of financial risk, as we cannot observe detailed insurance contracts, but individual perceptions can have informative value in this specific context.

Finally, we use demographic information on age, retirement status and wealth, unless otherwise indicated. The wealth measure is a sum of household income from respondent employment, employment of other household members, interest and dividend payments, public and private pension earnings, and savings. The composite wealth measure is then placed within the appropriate deciles of wealth for the respondent’s country of residence to reflect relative financial position of the household.