Smoking, education and the ability to predict own survival probabilities
Introduction
Subjective survival probability (SSP) survey question – i.e., the probability a person assigns to the own likelihood to survive to a certain age – is a good predictor of mortality. Several studies show that SSP predicts mortality well even after controlling for mortality-related risk factors such as demographic and socioeconomic characteristics and objective health measures (Doorn & Kasl, 1998; Elder, 2012; Hurd & McGarry, 1995, 2002; Kutlu-Koc & Kalwij, 2017; Manski, 2004; Perozek, 2008; Siegel, Bradley, & Kasl, 2003; Smith, Taylor, & Sloan, 2001). This means that people know the effects of their characteristics and behaviors on their survival probability.
Methodologically speaking, subjective survival probability questions have also been shown to have good properties. For example, Dominitz and Manski (1997) pointed out that subjective probabilities can be checked for internal consistency using the laws of probability, and they can be directly compared across individuals.
The relevance of SSPs lies in the fact that they measure respondents’ estimate of the probability to live up to a certain age, incorporating individuals’ private and subtle information on mortality which could not be directly measurable through objective questions (Perozek, 2008) nor by standard life tables (Hurd, McFadden, & Gan, 1998). In turn, such expectations about the future are central in the decision making process that considers the dynamic implications of the individual’s choices (Wang, 2014). Studies of life cycle behavior based on the well-known variation in mortality rates by socio-economic status have acknowledged that the focus on SSP overcomes misestimations of standard life tables in savings and retirement models, as individuals learn over time and reach older ages with very different levels of knowledge and internalization. As Hurd et al. (1998) argue, such variation in decision making might depend on personality differences across individuals as well as on differences across sub-populations in perceived mortality risk that cannot be observed in standard life tables.
Therefore, understanding the variability of SSPs within a population is important because they may affect life-cycle decisions under uncertainty. Important decisions, such as when to exit the labor market, whether to buy a life insurance or to invest in consumption goods, and specific choices about the own savings are likely to be based on individuals’ longevity expectations. Also, expectations of own survival have been found to influence subsequent health-related behaviors (Adams, Stamp, Nettle, Milne, & Jagger, 2015).
The literature on SSPs using data on the USA from the Health and Retirement Study (HRS) shows that they are, in general, consistent with the observed survival patterns at the population and at individual level (Hurd & McGarry, 2002; Hurd, 2009; Novak & Palloni, 2013; Siegel et al., 2003; Smith et al., 2001). Yet, sub-groups within the population not only may display different survival probabilities depending on both observed and unobserved characteristics, but they may also be more or less able to predict own survival.
In this respect, several studies have assessed how SSPs vary across individuals (Bissonnette, Hurd, & Michaud, 2014; Khwaja, Sloan, & Chung, 2007; Ludwig & Zimper, 2013; Perozek, 2008), showing changes with observed characteristics such as health status, parental longevity, BMI, and smoking (Falba & Busch, 2005; Hurd & McGarry, 1995, 2002; Kutlu-Koc & Kalwij, 2017). In view of the high percentage of the American population that consists of current or past smokers, a percentage that reached 77% in some male cohorts, this study focuses on smoking behavior. The prevalence of cigarette smoking in the United States first rose and then fell during the 20th century, with the sex differential in smoking prevalence also first rising and then falling. While only about 6% of women smoked in 1924, the proportion increased to 16% by 1929. During the same period, more than 50% of men smoked. In 1955, smoking prevalence was 56.9% for men and 28.4% for women aged 18+. This sex difference subsequently declined to about 5 percentage points and has remained stable since the 1990s (Wang & Preston, 2009).
In addition to smoking behavior, we examine the variability in SSPs by education. Progress in health and life expectancy is closely associated with socio-economic development (Lutz & Kebede, 2018). Education, in particular, has been found to be a key factor in this respect. Research has demonstrated a predominant importance of education and the associated cognitive changes affecting risk perception, planning horizon, and access to information promoting health-related behaviors and use of health care facilities (Baker, Leon, Smith Greenaway, Collins, & Movit, 2011; Lutz & Skirbekk, 2014).
This study aims first at identifying differences in the subjective perceptions of survival between individuals who currently smoke, those who have smoked in the past, and those who have never smoked. In doing so, we further distinguish between three levels of education and compare SSPs across sub-groups of the population defined by their educational attainment and smoking behavior. We use data from the HRS.
The second aim is to assess whether the results obtained from the analyses on the subjective survival probabilities reflect the real data, i.e. the actual survival patterns of the sub-populations considered. Because this information is not available in life tables, which at best contain average survival probabilities for a few population sub-groups, we have estimated survival probabilities from the longitudinal sample of HRS. Therefore, in a second step, we use a Gompertz survival model to study how educational attainment and smoking behavior affect the objective survival probability (OSP). Finally, we compare subjective and objective survival probabilities in order to assess whether sub-groups defined by smoking behavior and education differ in their ability to predict own survival.
Section snippets
Background
This study draws on previous literature from various strands, including research on subjective probabilities, smoking, and education.
The health capital investment framework (Grossman, 1972) considers health capital as a form of human capital in which individuals may decide to invest more or less, for example by not smoking or smoking respectively, according to their preferences as well as to individual-specific expected costs and benefits (Hersch, 1996). Individuals take on a safe behavior if
Data
We use data from the Health and Retirement Study (HRS), an age-cohort–based longitudinal panel survey of persons aged 50 years and older in the United States. In particular, we consider respondents interviewed for the first time in 2000, 2002, 2004, 2006, 2008, 2010, and 2012 waves.
Our analysis applies to older adults aged 50–89 years. It excludes respondent who are living in nursing homes as their number is limited to carry out separate analyses, yet they might substantially differ from the
Results
In the first step of our multivariate analyses we investigated how SSPs vary by sub-groups defined by education (low, medium, high) and smoking behaviors (never smoked, smoked in the past, currently smoking), adjusting for the control variables previously described. Fig. 2 shows the predicted SSPs for the 9 considered sub-groups with the respective 95% confidence intervals. These predictions are derived from linear regression models where the outcome variable is the individual SSP, the
Discussion and conclusion
Previous literature on health-related behaviors has demonstrated that smoking negatively impacts on health and survival (Jayes et al., 2016; Taghizadeh et al., 2016). Other strands of research focusing on education have also highlighted education effects on health as well as on health behaviors and life expectancy (Bijwaard, Poppel, van, Ekamper, & Lumey, 2015; Brunello, Fort, Schneeweis, & Winter-Ebmer, 2016; Cacciani et al., 2015). Drawing on the importance of subjective survival
References (51)
- et al.
Smoking, health, risk, and perception
Journal of Health Economics
(2005) - et al.
SmokeHaz: Systematic reviews and meta-analyses of the effects of smoking on respiratory health
Chest
(2016) - et al.
The role of time and risk preferences in smoking inequalities: A population-based study
Addictive Behaviors
(2013) - et al.
The 21st century hazards of smoking and benefits of stopping: A prospective study of one million women in the UK
The Lancet
(2013) - et al.
Life expectancies of cigarette smokers and nonsmokers in the United States
Social Science & Medicine
(1991) The analysis of survival (mortality) data: fitting Gompertz, Weibull, and logistic functions
Mechanisms of Ageing and Development
(1994)- et al.
Anticipated survival and health behaviours in older English adults: Cross sectional and longitudinal analysis of the English Longitudinal Study of Ageing
PLoS ONE
(2015) - et al.
The education effect on population health: A reassessment
Population and Development Review
(2011) Survival expectations, subjective health and smoking: Evidence from SHARE
Empirical Economics
(2014)- et al.
A theory of rational addiction
Journal of Political Economy
(1988)