Cawley and Ruhm 2011 — The Economics of Risky Health Behaviors

Summary

A 162-page handbook chapter for the Handbook of Health Economics, Vol. 2 (NBER WP 17081), providing a comprehensive overview of the theoretical frameworks and empirical evidence on the economics of risky health behaviors — smoking, alcohol, drug use, diet, physical inactivity, and obesity. Uniquely organized by underlying economic concepts rather than by individual behavior, enabling cross-behavior synthesis. The chapter covers: (1) the scale and distribution of health behavior risks; (2) the Grossman health capital model and Theory of Rational Addiction; (3) non-traditional models (hyperbolic discounting, bounded rationality, visceral factors, dual-self); (4) economic consequences of risky behaviors (medical costs, education, wages, employment, crime); and (5) policy instruments (taxes, information, advertising restrictions, defaults).

Key Claims

Scale of the Problem

$\approx$ $38\%$ of US deaths in 1990 and 2000 are attributable to the combination of smoking, diet, physical activity, and alcohol consumption (McGinnis and Foege 1993; Mokdad et al. 2004).
In high-income countries: smoking responsible for $18\%$ of deaths and $11\%$ of disability-adjusted life years (DALYs); excess body weight for $8\%$ of deaths and $7\%$ of DALYs; physical inactivity for $8\%$ of deaths and $4\%$ of DALYs (World Health Organization (WHO) 2009).
Obesity more than doubled: $15\%$ (early 1970s) → $34\%$ (2003–2006); daily caloric intake rose $12\%$ for men, $23\%$ for women. Smoking prevalence nearly halved: $37\%$ (1974) → $20\%$ (2007).

SES and Health Behaviors

Strongest empirical regularity: higher socioeconomic status (SES) → healthier behaviors (lower smoking, obesity, inactivity; more screening). College vs. high school (HS) dropout: $-13.9$ percentage points (pp) smoking, $-8.7$ pp obesity, $-22.3$ pp physical inactivity.
The smoking SES gradient has widened dramatically: $11.6$ pp gap (1971–74) → $21.5$ pp gap (1999–2002); highly educated quit at far higher rates ( $80.7\%$ quit ratio vs. $45.7\%$ for those without HS diploma).
The obesity SES gradient has narrowed: higher SES groups have experienced faster body mass index (BMI) increases. Among non-Hispanic Black males, income and obesity are now positively correlated.
About $30\%$ of the education–health behavior gradient is accounted for by differences in cognitive ability, particularly high-level processing (Cutler and Lleras-Muney 2010); the rest is unexplained.

Grossman Health Capital Model

Foundational framework: individuals receive a health stock endowment at birth that depreciates with age, is raised by investments, and determines available healthy time. Health has both consumption value (enters utility directly) and investment value (determines capacity for market and non-market activity).
Unhealthy behaviors = negative investments in health: optimal when the marginal benefit (instantaneous pleasure) equals the marginal cost (monetary cost + reduced health stock + shorter lifespan).
Education improves health through allocative efficiency (choosing healthier inputs) and productive efficiency (getting more health from the same inputs).

Theory of Rational Addiction (TORA — Becker and Murphy 1988)

Addiction characterized by reinforcement ( $U_{CS} > 0$ : past consumption raises marginal utility (MU) of current consumption), tolerance ( $U_S < 0$ : past consumption lowers utility in harmful addictions), and withdrawal ( $U_C > 0$ : consuming the good always provides positive utility).
Addicts are forward-looking optimizers: addiction can be individually rational if the present discounted value of utility is maximized. Addicts need not be glad they are addicted.
Unstable steady states: for highly addictive goods, consumption distribution is bimodal (abstinence or addiction, few moderate users). Exogenous shocks (job loss, divorce) can trigger transitions between steady states.
Key empirical test: under TORA, future anticipated price increases reduce current consumption. Gruber and Koszegi (2001) find cigarette sales rise (stockpiling) but consumption falls in response to announced future tax increases — consistent with rational foresight.
Price elasticities of demand: cigarettes $-0.3$ to $-0.5$ (meta-analysis of 523 estimates, Gallet and List 2003); alcohol median $-0.44$ (Gallet 2007); food: small effects for most consumers but $3$ – $5\times$ larger for youths above the 80th BMI percentile (Auld and Powell 2009).
Income elasticities: cigarettes mean $0.42$ (Gallet and List 2003); alcohol median $0.69$ (Gallet 2007). Natural experiments (lottery winnings, Social Security notch, Earned Income Tax Credit (EITC)) yield small and mixed income effects on health behaviors.

Time Preference

Victor Fuchs (1982) proposed rate of time preference as the common factor explaining the education–health correlation.
Evidence is mixed: BMI correlates with savings rates and willingness to delay gratification; but smokers and non-smokers have similar rates when measured via willingness to undergo colonoscopy (Khwaja et al. 2007).
Cutler and Glaeser (2005): within-person correlations across health behaviors are surprisingly low — most below $10\%$ , the highest (alcohol and smoking) at $16\%$ . Changes in health behaviors over time are also weakly correlated. Time preference is not the dominant explanation for clustering of risky behaviors.

Non-Traditional Models

Quasi-hyperbolic discounting (Laibson 1997): discount factor for immediate vs. near-future ( $\beta\delta$ ) is lower than for any two future periods ( $\delta$ ). If $\beta < 1$ , preferences are time-inconsistent. Gruber and Koszegi (2001) incorporate this into a smoking model; key implication is that standard TORA tests cannot distinguish rational addiction from time-inconsistent preferences.

Rational addiction with learning (Orphanides and Zervos 1995): individuals are uncertain about whether they are the "addict type." Overoptimists (underestimating their addictive tendency) are most likely to become addicted; all addicts regret their decision ex post. Policy can improve welfare by providing information that reduces divergence between subjective and objective risk.

Visceral factors (Loewenstein 2000): acute emotions — hunger, anger, sexual desire — create "hot states" where individuals appear to exhibit extreme discounting. In cold states, individuals systematically underestimate how much hot-state visceral factors will affect their choices.

Cue-triggered consumption (Laibson 2001): repeated pairing of environmental cues with addictive goods creates complementarity; the smell of food or sound of ice in a glass generates cravings. Strategic cue management (avoiding people, places, and smells associated with consumption) is a rational response.

Dual-self models (Thaler and Shefrin 1981): internal battle between a farsighted "planner" (prefrontal cortex) and myopic "doer" (limbic system). The planner can constrain the doer through costly precommitment, willpower expenditures, and mental accounting rules.

Bounded rationality: Akerlof (1991) shows small per-period biases toward present utility can accumulate into large errors; procrastination as repeated failures to follow through on intended behavior changes.

Economic Consequences

Macroeconomic fluctuations (Ruhm): In bad economic times, heavy drinking, smoking, obesity, and physical inactivity all decrease; diets improve. Two mechanisms: (1) income reductions lower some unhealthy consumption; (2) time-intensive healthy behaviors (exercise) increase when work hours fall. A $1$ pp increase in unemployment reduces total mortality by $0.3$ – $0.5\%$ , with large drops in coronary heart disease and traffic fatalities; cancer mortality is unaffected. Evans and Moore (2009) show the same causes of death exhibit large within-month variation (consistent with "full wallet" effects, not intertemporal optimization).

Medical care costs:

Obesity: OLS = $\$676$ /year higher medical spending; instrumental variables (IV) estimate (using child's obesity as instrument) = $\$2{,}826$ /year ( $41.5\%$ higher) — $4\times$ the ordinary least squares (OLS) due to attenuation bias in self-reported weight (Cawley and Meyerhoefer 2010). Total US obesity-related medical spending: $\$85.7$ B/year in 2008.
Smoking at age 24: $\$3{,}757$ / $\$2{,}617$ higher lifetime medical expenditures for women/men (Sloan et al. 2004).

Education: Binge drinking reduces high school graduation probability by up to $14.5\%$ for males (Renna 2007). High school frequent drinkers complete $2.2$ fewer years of college (Cook and Moore 1993).

Employment: Problem drinking reduces employment probability by $50\%$ for men and $40\%$ for women in Finnish data (Johansson et al. 2007). Obese job applicants $6$ – $8$ pp less likely to receive interview invitations in audit study (Rooth 2009).

Wages: Moderate alcohol consumption raises income by $+10\%$ ("drinker's bonus"); smoking reduces income by $-24\%$ (Auld 2005). White female wages begin declining at BMI $\approx 23$ (well within healthy range), suggesting a beauty premium rather than pure obesity penalty (Gregory and Ruhm 2011). Two-standard-deviation weight increase → $-18\%$ wages for white females (Cawley 2004 IV).

Crime: Crack cocaine introduction into a city raised urban crime rates by $\approx$ $10\%$ (Grogger and Willis 2000). Higher alcohol taxes → lower crime (Carpenter and Dobkin 2010).

Policy

Taxes: Pigouvian rationale for internalizing externalities. Manning et al. (1991) calculates net external cost of smoking including disability benefits and lost wage tax revenue; some of the "cost" is offset by lower Social Security payments from early death.
Behavioral paternalism: When behaviors harm one's future self (internalities rather than externalities), paternalistic intervention may increase social welfare even without market failures — a key implication of hyperbolic discounting models.
Information: 1964 Surgeon General's report → immediate $5\%$ smoking decline; effects larger for more educated (cognitive ability channel). Dupas (2011) Kenya experiment: information that HIV risk is higher for older men reduced teen pregnancy by $61\%$ .
Advertising bans: Comprehensive bans reduce tobacco use (Saffer and Chaloupka 2000) but limited bans have little effect; Nelson (2010) finds no alcohol advertising ban effects.
Calorie labeling: NYC 2008 law had minimal effects on fast food purchases (Elbel et al. 2009); but $-6\%$ calories per transaction at Starbucks (Bollinger et al. 2010).
Default architecture: Making low-calorie items the default option promising; empirical evidence just beginning to accumulate as of 2011.

Concepts Introduced or Extended

Entities Mentioned

Quotes

"By far the strongest results are that higher socioeconomic status (SES), as proxied by educational attainment or family income, is generally correlated with healthier behaviors."

"Cutler and Glaeser (2005) find that the correlations across health behaviors are 'surprisingly low' (p. 238); most below 10%, with the highest (alcohol and smoking) at 16%... suggesting that time preference is not a major determinant of health behaviors."

"In bad economic times, heavy drinking and drunk-driving, smoking, obesity, and physical inactivity decrease and diets improve... a one percentage point increase in unemployment reduces total mortality by 0.3 to 0.5 percent."

My Take

The paper's most durable contributions are methodological: the cross-behavior organization reveals that the TORA, hyperbolic discounting, and visceral factors are all special cases of a more general framework for modeling self-control failures; and the Ruhm macroeconomic findings give the most credible causal identification available for the health-behavior–mortality relationship. The Cawley-Meyerhoefer obesity medical cost result (IV = $4\times$ OLS) is one of the field's best illustrations of why self-reported weight data leads to severe attenuation bias. For the wiki, the most important connections are: (1) the procyclical mortality literature directly informs DI application cyclicality and Cawley et al. 2009 (next paper); (2) the SES behavioral gradient literature provides context for the widening mortality-education gradient documented in Bound et al. (2015), Case and Deaton (2015/2017), and Waldron (2007); and (3) the disability-externalities section of the taxation discussion connects to the Disability Insurance (DI) moral hazard literature throughout the wiki.