2.1. Market-based approach to BCA

Our BCA focuses on anti-smoking policies that help correct individual failures in making utility-maximizing choices about cigarette consumption. The benefits to the consumer are measured by the compensating variation (CV) in income for the anti-smoking policy under consideration. Assume the consumer’s lifetime utility is given by $U=U(A,G,H,Y)$ , where $A$ is the quantity of cigarettes consumed, $G$ is the quality of cigarettes consumed, $H$ is health, and $Y$ is a composite commodity with its price normalized to one. Letting the subscript 0 indicate consumption choices made in the absence of the policy and ${\rm\Delta}$ indicate the policy-induced changes, the CV for the policy is defined to satisfy the equation:

In equation (1) the utility function that defines the CV is in terms of experienced utility, which may differ from decision utility (Kahneman, Wakker & Sarin, Reference Kahneman, Wakker and Sarin1997). In particular, the consumer’s experienced lifetime utility depends on the true health consequences of smoking even if these consequences are not fully incorporated into her decisions. The typical anti-smoking policy reduces the quantity ( ${\rm\Delta}A<0$ ) or quality ( ${\rm\Delta}G<0$ ) of cigarettes consumed over the lifetime. As a result, the consumer’s lifetime health improves ( ${\rm\Delta}H>0$ ). Because she spends less money on cigarettes her consumption of the composite commodity increases ( ${\rm\Delta}Y>0$ ).Footnote ⁵ If experienced utility and decision utility are the same so that the consumer’s choices $A_{0}$ and $G_{0}$ maximize her experienced lifetime utility, the changes induced by the policy make her worse off. The utility losses from ${\rm\Delta}A$ and ${\rm\Delta}G$ outweigh the utility gains from ${\rm\Delta}H$ and ${\rm\Delta}Y$ . The policy can make the consumer better off when experienced utility and decision utility are different so the consumer’s original choices fail to maximize lifetime utility. In this case the utility gains from ${\rm\Delta}H$ and ${\rm\Delta}Y$ outweigh the utility losses from ${\rm\Delta}A$ and ${\rm\Delta}G$ . Lifetime cigarette consumption might fail to maximize lifetime utility due to the consumer’s poor information about the health consequences of smoking, other decision-making errors that lead to imperfect optimization, or bounded self-control.

The CV places a dollar value on the net utility gain (or loss) from the anti-smoking policy. The definition of the CV in equation (1) does not require any assumptions about the consumer’s ability to make lifetime-utility-maximizing choices of $A,G,H$ and $Y$ . Adding the standard assumptions about rationality implies that the CV can be measured as an area of consumer surplus under the consumer demand curve (Deaton & Muellbauer, Reference Deaton and Muellbauer1980, pp. 184–189). Following the Australian Productivity Commission (2010), Weimer et al. (Reference Weimer, Vining and Thomas2009), Ashley et al. (Reference Ashley, Nardinelli and Lavaty2014) we assume that the observed market demand curve for cigarettes might not reflect fully optimizing decisions. Instead, we measure the area of consumer surplus that corresponds to the CV with respect to a hypothetical demand curve “which summarizes utility-maximizing decisions by consumers who are aware of all health costs and have fully incorporated them into consumption decisions” (Ashley et al., Reference Ashley, Nardinelli and Lavaty2014, p. 2). We tailor the Ashley et al. conceptual framework to our retrospective and prospective BCA as follows.

For the retrospective BCA our framework needs to allow for a situation where initially poorly informed consumers become much better informed over time. Our retrospective study period starts in 1964. This is only about a decade after the scientific research establishing the link between smoking and lung cancer began to receive major media attention and the new information was still diffusing to the public (Viscusi, Reference Viscusi1992). In a Gallup Poll conducted in 1960 only about half of the respondents agreed that smoking was one of the causes of lung cancer (Gallup, Reference Gallup1960). As will be discussed in more detail below in Section 3, the 1964 Surgeon General’s Report is widely seen as a landmark publication that directly provided information and indirectly helped launch other initiatives to provide information about the health consequences of smoking. By the late 1990s general population surveys found that virtually all U.S. consumers were aware of the links between cigarette smoking and the risks of lung cancer, heart disease and emphysema (Kenkel & Chen, Reference Kenkel, Chen, Jha and Chaloupka2000). A series of studies in the 1990s and 2000s provide evidence that most people in the US, including most smokers, overestimate the health risks of cigarette smoking.Footnote ⁶ While this evidence has attracted some controversy, it is not controversial to conclude that by the end of our study period in 2010 consumer information about the major health risks of smoking is very widespread.Footnote ⁷

For both our retrospective and prospective analyses it is also important that our framework allows for the possibility that consumers might make other decision-making errors that lead to imperfect optimization, or have bounded self-control. Psychological and behavioral economics research document decision-making errors across a range of domains (DellaVigna, Reference DellaVigna2009). However, Robinson and Hammitt (Reference Robinson and Hammitt2011, p. 2) point out that: “Much of the research involves small-scale laboratory experiments and needs to be supplemented by additional fieldwork to explore the real-world importance of the findings.” The real-world importance for the cigarette market is controversial. Survey evidence shows that many more smokers state that they intend to quit in the future than actually do so. This has been interpreted as evidence of the time-inconsistency predicted by behavioral economic models with hyperbolic discounting. However, statements about quit intentions can also be viewed with the same general skepticism economists view simple statements about preferences.Footnote ⁸ Econometric research into the market for cigarettes has not been able to distinguish rational addiction from a forward-looking model with quasi-hyperbolic discounting and time-inconsistency.Footnote ⁹ Studies of self-commitment devices provide another source of evidence that some smokers recognize that they face time-inconsistency problems (Khwaja, Silverman & Sloan, Reference Khwaja, Silverman and Sloan2007; Gine, Karlan & Zinman, Reference Gine, Karlan and Zinman2010). The available research provides evidence that the observed demand for cigarettes reflects individual failures to make lifetime-utility-maximizing decisions, but there is less evidence on the extent and precise nature of the individual failures.

To capture the issues discussed above, our conceptual framework assumes that initially consumers fail to make lifetime-utility-maximizing choices about cigarette consumption, but these failures are corrected over time with the help of anti-smoking policies. We take a reduced-form approach and do not specify the precise nature of the individual failures. Instead we assume that the individual failures stem from some combination of poor information about the health consequences of smoking, other decision-making errors that lead to imperfect optimization, and bounded self-control.

Figure 1 is a graphical illustration of the framework. The cigarette consumption observed in year $t$ is $A_{\mathit{Ot}}$ at the intersection of the market price $P_{t}$ and the observed demand curve $D_{\mathit{Ot}}$ . The counterfactual cigarette consumption that would have been observed in year $t$ in the absence of any anti-smoking policies is $A_{\mathit{CFt}}$ at the intersection of $P_{t}$ and the counterfactual demand curve $D_{\mathit{CFt}}$ . $A_{\mathit{CFt}}$ is also the starting point of the retrospective BCA, so the horizontal distance $A_{\mathit{CFt}}{-}A_{\mathit{Ot}}$ shows the impact of anti-smoking policies as of year $t$ . The rational utility-maximizing consumption is $A_{\mathit{Rt}}$ at the intersection of $P_{t}$ and the rational demand curve $D_{\mathit{Rt}}\mid A_{t-1}$ . The notation $D_{\mathit{Rt}}\mid A_{t-1}$ reflects that because cigarettes are addictive the rational level of demand at time $t$ depends on the addictive stock determined by past consumption at time $t-1$ .Footnote ¹⁰

Figure 1

Conceptual framework.

The vertical distances between $D_{\mathit{CFt}}$ and $D_{\mathit{Ot}}$ or $D_{\mathit{Rt}}\mid A_{t-1}$ show the dollar values of the individual failures to maximize utility. Individuals make errors and only internalize a fraction of the value of the health consequences of their cigarette consumption.Footnote ¹¹ Following what is becoming conventional usage we will call the uninternalized portion of the health consequences the “marginal internality” (Mullainathan, Schwartzstein & Congdon, Reference Mullainathan, Schwartzstein and Congdon2012).

Because Figure 1 can be used to show the welfare consequences of anti-smoking policies that partly or completely correct individual failures it provides the framework for our retrospective and prospective BCAs. The framework can be viewed as a modification of Harberger’s (Reference Harberger1971) first basic postulate for BCA, where the value of the marginal unit to demanders is given by the height of the rational demand curve $D_{\mathit{Rt}}\mid A_{t-1}$ . In our retrospective BCA, anti-smoking policies as of year $t$ reduce cigarette consumption from $A_{\mathit{CFt}}$ to $A_{\mathit{Ot}}$ . Consumer spending on cigarettes falls by an amount given by the area $J+K+L$ . The rational valuation of the foregone consumption is given by area $L$ under $D_{\mathit{Rt}}\mid A_{t-1}$ , yielding net consumer benefits given by area $J+K$ . In our prospective BCA we consider a hypothetical FDA regulation that further reduces cigarette consumption from $A_{\mathit{Ot}}$ to $A_{\mathit{Rt}}$ . By the same reasoning this reduction in consumption yields benefits given by the triangular area $G$ .

Another way to view the framework is in terms of the rule: the CV for a policy that changes cigarette consumption by ${\rm\Delta}A$ is given by ${\rm\Delta}A$ multiplied by the average value of the marginal internality. This rule is consistent with the general expression for the marginal welfare gain with behavioral agents in Mullainathan et al. (Reference Mullainathan, Schwartzstein and Congdon2012, equation 6, p. 521). In our retrospective BCA, because $A_{\mathit{CFt}}$ – $A_{\mathit{Ot}}$ shows the impact of anti-smoking policies as of year $t$ , the year $t$ value of those policies is $A_{\mathit{CFt}}$ – $A_{\mathit{Ot}}$ times the average of the vertical distance between $D_{\mathit{CFt}}$ and $D_{\mathit{Rt}}\mid A_{t-1}$ (the marginal internality at $A_{\mathit{CFt}}$ ) and the vertical distance between $D_{\mathit{Ot}}$ and $D_{\mathit{Rt}}\mid A_{t-1}$ (the marginal internality at $A_{\mathit{Ot}}$ ). This again corresponds to the sum of the area $J+K$ . In the case drawn in Figure 1 the policy benefits (area $J+K$ ) are necessarily smaller than the reduction in consumer spending on cigarettes (area $J+K+L$ ). However, this need not hold because depending on the size of the marginal internality the rational demand curve $D_{\mathit{Rt}}\mid A_{t-1}$ can be below the horizontal axis. This occurs if at $A_{\mathit{Ot}}$ given rational demand the value of the marginal cigarette is negative because of the health consequences.

A policy that completely corrects the individual failures would change cigarette consumption by $A_{\mathit{CFt}}$ – $A_{\mathit{Rt}}$ . The value of such a policy is $A_{\mathit{CFt}}$ – $A_{\mathit{Rt}}$ times the average of the vertical distance between $D_{\mathit{CFt}}$ and $D_{\mathit{Rt}}\mid A_{t-1}$ (the marginal internality at $A_{\mathit{CFt}}$ ) and 0 (the marginal internality at $A_{\mathit{Rt}}$ ), corresponding to the familiar triangular area $G+J+K$ .

Viewed from a static perspective, Figure 1 shows that the welfare gains from future anti-smoking policies diminish when past policies have already partly corrected the individual failures to maximize. For example, the area $J+K$ of consumer benefits from reducing $A_{\mathit{CFt}}$ to $A_{\mathit{Ot}}$ is much larger than the area G of consumer benefits from reducing $A_{\mathit{Ot}}$ to $A_{\mathit{Rt}}$ . This pattern is because the size of the welfare gain depends on the size of the marginal internality, which is much larger at $A_{\mathit{CFt}}$ than at $A_{\mathit{Ot}}$ .

However, BCA of anti-smoking policies involves important dynamics because addiction creates linkages between the demand curves across time. Consider a policy at time $t$ that helps consumers internalize more of the health consequences of smoking and thus causes the demand curve to shift from $D_{\mathit{CFt}}$ to $D_{\mathit{Ot}}$ . While the area $J+K$ measures the policy benefits at time $t$ , the reduced consumption level $A_{\mathit{Ot}}$ means that consumers at time $t+1$ have a lower addictive stock. In the time $t+1$ version of Figure 1 both the observed demand curve $D_{Ot+1}$ and the rational demand curve $D_{Rt+1}\mid A_{t}$ shift inwards toward the origin reflecting the lower addictive stock. In a way, the rational utility-maximizing level of consumption $A_{Rt+1}$ is a moving target: the reduced consumption at time $t$ reduces how much rational consumers would choose to smoke at time $t+1$ .

Importantly, the dynamics of cigarette demand reflect the utility costs of withdrawing from cigarette addiction. For a given addictive stock, much of the rational demand for cigarettes might simply be to avoid withdrawal costs. Hughes (Reference Hughes2007) reviews research on the effects of abstinence from tobacco and concludes (p. 315) that: “Anger, anxiety, depression, difficulty concentrating, impatience, insomnia, and restlessness are valid withdrawal symptoms that peak within the first week and last 2–4 weeks.” Hughes also mentions that other research reviews provide evidence for additional withdrawal symptoms that include craving, decreased heart rate, electroencephalographic changes, hunger and weight gain, impaired cognitive performance, increased monoamine oxidase levels, and changes in medication levels. Even with the aid of pharmaceutical smoking cessation products most quit attempts are unsuccessful, which provides more evidence that the withdrawal costs from cigarette addiction are high and also suggests that withdrawal is difficult to treat. Most smokers require multiple quit attempts before they succeed: in a typical example it might take a smoker roughly three years to quit, during which time they attempt to quit 6–11 times with the average episode lasting 1.5 weeks (Hughes, Keely & Naud, Reference Hughes, Keely and Naud2004). This implies that successful quitters spend on average 12 weeks of withdrawal symptoms during their serious attempts to quit. Rational cigarette demand might be mainly driven by the value of cigarettes as a means to reduce the utility losses from withdrawal.

When considering the rational demand for cigarettes at time $t$ , the rationality of smokers’ past decisions is irrelevant. Many smokers’ past decisions might have been irrational. Nevertheless, even these smokers’ current and forward-looking decisions about how much to smoke and whether to quit can be rational. A typical example might be someone who made a time-inconsistent decision to initiate smoking as an adolescent and became addicted. The past mistake to become addicted is a sunk cost. Given their current addictive stock, the question of rationality depends on whether they are making fully informed time-consistent decisions about continuing to smoke. Put differently, currently rational smokers are not necessarily “happy addicts” – they may very much regret their earlier time-inconsistent decision to start smoking. They can be currently rational in the sense that their current decisions to continue smoking are time consistent and do not impose internalities on their future selves.

Our retrospective and prospective BCAs of anti-smoking policies implement the framework summarized in Figure 1. We use a dynamic population model to compare observed cigarette consumption over time to consumption under simulated counterfactuals. To simulate rational consumption over time, we use the smoking rates of a group of consumers whose behavior tends to approximate fully informed rational decision making. The dynamic population model simulations build in the linkages between the demand curves across time. By combining the simulation results with a consensus price-elasticity estimate we construct a series of demand curves $D_{\mathit{CF}}$ , $D_{O}$ and $D_{R}\mid A_{t-1}$ over time and use them to calculate the areas $J+K$ and $G$ .

2.2. The health benefits approach and consumer surplus offset

In contrast to the market-based approach summarized in Figure 1, several recent BCAs of health-related regulations use what we term the health benefits approach. Although the approaches are not necessarily inconsistent and indeed Figure 1 can illustrate both, the approaches implicitly frame the policy problem quite differently. The market-based approach frames the problem as: we use BCA to determine if the policy leads to a market outcome that is closer to the socially optimal quantity. In a general version of Figure 1 the socially optimal quantity maximizes the sum of consumer and producer surplus, after taking into account market failures like externalities and, in a behavioral BCA, individual failures like internalities.Footnote ¹² The public health missions of federal agencies like the FDA make this framing of the problem seem foreign. From a public health perspective, the optimal level of cigarette consumption is obviously zero. Consistent with the agencies’ public health missions, the health benefits approach frames the problem as: we use BCA to measure the value of the policy-induced health gains and to see if the health benefits justify the costs. As reflected in Harberger’s (Reference Harberger1971) first basic postulate, consumer surplus is at the heart of the market-based approach to BCA. In contrast, in the health benefits approach consumer surplus is framed as a possible cost that offsets some of the health benefits. This frame is not necessarily inconsistent with OMB’s guidance for RIAs. After discussing approaches for valuing policy benefits, Office of Management and the Budget (2011, p. 10) advises but does not require that agencies should include gains or losses in consumer or producer surplus “where relevant.”Footnote ¹³

As we just mentioned, Figure 1 can be used to illustrate the health benefits approach to BCA. Consider an anti-smoking policy that corrects all individual failures and reduces cigarette consumption by $A_{\mathit{CFt}}$ – $A_{\mathit{Rt}}$ . The health benefits approach focuses on the health gains, as measured for example by quality-adjusted life years (QALYs), predicted to flow from the reduction in smoking. Using standard estimates of the dollar value of a QALY yields an estimate of the dollar value of the policy-induced health gains. Ashley et al. (Reference Ashley, Nardinelli and Lavaty2014) refer to this as the gross health benefits of the anti-smoking policy. Because the vertical distance between $D_{\mathit{CFt}}$ and $D_{R}\mid A_{t-1}$ is by definition the dollar value of the health consequences of smoking, the gross health benefits of a policy that decreases cigarette consumption by $A_{\mathit{CFt}}$ – $A_{\mathit{Rt}}$ is given by the area $E+F+I+G+J+K$ .

The health benefits approach to BCA calculates the anti-smoking policy’s gross health benefits and then calculates how much of the benefits are offset by lost consumer surplus from the reduction in cigarette consumption. Recall that the CV for the change in cigarette consumption $A_{\mathit{CFt}}$ – $A_{\mathit{Rt}}$ is given by the area $G+J+K$ . Ashley et al. (Reference Ashley, Nardinelli and Lavaty2014) refer to this as the net health benefits. As Cutler (Reference Cutler2008) points out, with linear demand curves like in Figure 1, the area of the triangle $G+J+K$ of net health benefits from this policy-induced change is exactly one-half the area of the parallelogram $E+F+I+G+J+K$ of gross health benefits. In other words, half of the gross health benefits are offset by lost consumer surplus. Ashley et al. (Reference Ashley, Nardinelli and Lavaty2014) emphasize that for other policy-induced changes and other assumptions the net-to-gross benefits or offset ratio can range from zero to 100%. Several recent RIAs of health-related regulations use offset ratios ranging from 10 to 99%.Footnote ¹⁴ The use of anything but a very small offset ratio near zero has been criticized in public comments on the RIAs, commentaries in public health journals and even in the popular press.Footnote ¹⁵

Our market-based approach to the BCA of anti-smoking policies complements recent RIAs that use the health benefits approach. Our approach does not use estimates of the gross health benefits of anti-smoking policies, so it does not require an assumption about the appropriate offset ratio. Instead, as in the standard revealed preference approach to applied welfare economics, our approach analyses market behavior to infer the value of the policy-induced changes in cigarette consumption. Although it is not the main focus of our study, in Section 6 we return to a discussion of what our results imply for the health benefits approach to BCA and the consumer surplus offset controversy.

3.1. Background

The first step of our retrospective BCA is to define the set of anti-smoking policies and their impact on cigarette consumption. We consider the broad set of U.S. policies implemented over the period 1964–2010 that were designed to improve consumer information about the health consequences of cigarette smoking. As we mentioned above, the 1964 Surgeon General’s Report (SGR) is widely viewed as a landmark publication that launched modern anti-smoking policies in the United States. In its wake followed: additional SGRs further documenting the health consequences (almost annually); mandatory warning labels on cigarette packages and advertisements (enacted in 1965, strengthened in 1969, 1984, and pending); a ban on television and radio cigarette advertisements (1971); school-based health education about smoking (continuing); and mass media anti-smoking campaigns (continuing). School health education and mass media anti-smoking campaigns include new campaigns launched by state tobacco control programs in the 1990s and 2000s and the American Legacy Foundation’s ‘truth’ campaign funded by the 1998 Master Settlement Agreement between the tobacco industry and state attorneys general.

We do not attempt to isolate the impact of specific anti-smoking policies but instead consider their combined impact. Our approach is to attribute shifts in the observed cigarette demand curve $D_{\mathit{Ot}}$ over time to the combined impact of anti-smoking information policies. Because we focus on shifts in demand that reflect improved consumer information, as explained below we control for movements along the demand curve caused by tax-induced changes in cigarette prices over time.

Conceptually, we view cigarette excise taxes as a policy tool to correct for negative externalities. Externalities in the cigarette market include the health and utility costs of nonsmokers’ exposure to second hand smoke and the financial costs smoking-related illnesses impose on nonsmokers in collectively financed health insurance and retirement plans. Empirically, Sloan, Ostermann, Picone, Conover and Taylor (Reference Sloan, Ostermann, Picone, Conover and Taylor2004, Table 11.3, p. 255) estimate that circa 2000 cigarette taxes were not high enough to fully correct for externalities: the estimated externality per pack of cigarettes is $2.20 while the combined federal and average state cigarette tax per pack is $0.76.Footnote ¹⁶ However, by the end of our study period in 2010 subsequent federal and state tax hikes raise the combined tax to $2.18 per pack (Orzechowski & Walker, Reference Orzechowski and Walker2012). The externalities from second hand smoke are further corrected by the expansion of state and local bans on smoking in public places. From 2000–2009 the fraction of the U.S. population covered by smoke free worksite laws increased from 3% to 54% and the fraction covered by smoke free restaurant laws increased from 13% to 63% (Gonzalez, Sanders-Jackson, Song, Cheng & Glantz, Reference Gonzalez, Sanders-Jackson, Song, Cheng and Glantz2013). Some of the financial externalities of smoking are further addressed by the 50% health insurance premium surcharges for smokers allowed under the 2010 Patient Protection and Affordable Care Act. The policy trends suggest that over the period we study: one set of anti-smoking policies helped correct individual failures; while another set of policies helped correct externalities. To focus on the benefits of correcting individual failures, our approach abstracts from the net benefits of tax and other policies that correct externalities.Footnote ¹⁷

3.2. Dynamic population simulation model and data sources

We use a dynamic population model to compare observed cigarette consumption over time to cigarette consumption under a simulated “no-policy” counterfactual. Our simulation model is based on the model developed by Mendez, Warner and Courant (Reference Mendez, Warner and Courant1998), which they use to accurately predict smoking prevalence through 2010 (Warner & Mendez, Reference Warner and Mendez2012). Our model begins with estimates of the adult population by smoking status in 1960. The model uses estimates of birth rates, smoking status-specific mortality rates, and age-specific smoking initiation and cessation rates to simulate the number of adults smoking over time. The smoking population evolves as new cohorts initiate smoking while current smokers either continue to smoke, quit smoking and become former smokers, or die. Aggregate cigarette consumption in year $t$ is given by the size of the smoking population in year $t$ multiplied by an estimate of the average number of cigarettes consumed per smoker in year $t$ .

We use standard data on birth rates and estimates of mortality rates by smoking status from Jha et al. (Reference Jha, Landsman, Rostron, Thun, Anderson, McAfee and Peto2013). We follow Mendez et al. (Reference Mendez, Warner and Courant1998) and assume that everyone aged 0–17 is a never-smoker, and some initiate smoking at age 18. We estimate sex-specific birth cohort smoking initiation rates from the Tobacco Use Supplements to the Current Population Survey (TUS-CPS). We also use these data to estimate sex- and age-specific smoking cessation rates. We use data on observed cigarette consumption from the Tax Burden on Tobacco (Orzechowski & Walker, Reference Orzechowski and Walker2012). An online Appendix provides more details about our simulation model and data sources.

It is useful to provide a brief overview of the trends in the input variables used in our dynamic population model. From 1964 to 2010 the adult population aged 18 and older grew from 120 to 205 million. The observed rate at which new cohorts initiate smoking falls fairly steadily from 52% in 1964 to 17% in 2010. The observed average cessation rate increases from about 1.5% per year in 1964 to 3% in 1990 to almost 6% by 2010. The observed average number of cigarettes per smoker increases from 26.5 cigarettes per day in 1964 to a high of over 30 cigarettes per day in 1983. It then drops to 23 cigarettes per day by 2010. Using the observed initiation and cessation rates, the model’s simulated smoking prevalence and cigarettes per day compare well to the observed (and incomplete) National Health Interview Survey and TUS-CPS data over this period.Footnote ¹⁸

We create a no-policy counterfactual simulation to predict what cigarette consumption would have been in the absence of the 1964 SGR and subsequent anti-smoking policies. For the counterfactual we assume that smoking initiation and cessation rates are fixed at their 1964 levels. We assume that the counterfactual number of cigarettes per smoker increases as observed through 1983, and then remains constant at the 1983 level of over 30 cigarettes per day through 2010. To control for movements along the observed demand curve $D_{\mathit{CFt }}$ due to price changes, we use price-elasticity to adjust for the differences in consumption due to the differences between $P_{1964}$ and $P_{t}$ .

The assumptions in the counterfactual simulations probably tend to overstate the impact of anti-smoking policies, for two reasons. First, even without the 1964 SGR and public informational campaigns, it is likely that consumers would naturally learn on their own about the medical research on the health risks of smoking.Footnote ¹⁹ If so, even without the policies smoking initiation rates would naturally fall and cessation rates would naturally rise. Second, the observed increase in the number of cigarettes per smoker from 1964 to 1983, which we assume would hold in the counterfactual, might not have occurred without the informational policies. Several econometric studies find that during this period smokers reacted to the new information about the health risks of smoking by switching to filter cigarettes and lower tar and lower nicotine cigarettes (Schneider, Klein & Murphy, Reference Schneider, Klein and Murphy1981; Ippolito & Ippolito, Reference Ippolito and Ippolito1984). If smokers compensated for these changes by smoking more, improved information paradoxically might explain the observed increase in the number of cigarettes per smoker.Footnote ²⁰ The no-policy counterfactual should not include any policy-induced increase in cigarettes per smoker, but we are unaware of data on this. Because these assumptions mean simulated counterfactual consumption is probably too high, our retrospective BCA can be seen as providing an upper bound estimate of the benefits of anti-smoking policies from 1964 to 2010.

Figure 2

Retrospective impact analysis: counterfactual vs observed smoking prevalence.

3.3. Trends in actual and simulated counterfactual cigarette consumption

Figures 2 and 3 compare the observed trends in smoking prevalence and cigarette consumption from 1960 to 2010 to the results of the no-policy counterfactual simulation. Counterfactual smoking prevalence slightly declines from about 40% in 1960 to 37% in 2010. In the counterfactual case, the rates of smoking initiation and cessation are constant at their 1964 levels, so this slight drop in prevalence is due to changes in the age composition of the population. In contrast, observed smoking prevalence drops steadily and falls about in half to 17% in 2010. Due to population growth, from 1964 to about 1980 both counterfactual cigarette consumption and observed cigarette consumption increase. Counterfactual consumption continues to grow through 2010, but after 1980 the drop in observed smoking prevalence (and after 1983 the drop in observed cigarettes per day) outweighs population growth, so observed consumption begins to fall. Observed consumption in 2010 is only 39% of the level predicted in the counterfactual simulation.

Figure 3

Retrospective impact analysis: counterfactual vs observed cigarette consumption.

Observed cigarette consumption from 1964 to 2010 totals 1.13 trillion packs; under the no-policy counterfactual cigarette consumption totals 1.58 trillion packs over the same period. We use these simulation results to estimate that the overall impact of anti-smoking informational policies from 1964 to 2010 is to reduce cigarette consumption by 28%. This does not include reductions in cigarette consumption due to price increases over the period.

The policy-induced reductions grow over time, both in levels and in percentage terms. Observed consumption in 1974 is about 10% lower than counterfactual consumption. The reduction in observed consumption compared to counterfactual consumption grows to 21% in 1984 to 38% in 1994 to 48% in 2004 to 61% in 2010. The time pattern reflects wider dissemination of health information and the dynamics of the smoking population. Changes in the flows into and out of the smoking population due to changes in initiation and cessation are small relative to the stock of smokers. Their impact accumulates over time.

4.1. Constructing cigarette demand curves

To calculate the areas in Figure 1 that measure the consumer benefits of anti-smoking policies we construct a series of demand curves $D_{\mathit{CF}}$ , $D_{O}$ and $D_{R}\mid A_{t-1}$ for 1964–2010. The observed cigarette consumption levels and counterfactual simulation results provide points on the observed demand curves $D_{O}$ and the counterfactual demand curves $D_{\mathit{CF}}$ .

To find points on $D_{R}\mid A_{t-1}$ we use our dynamic population model to simulate consumption over time based on the smoking rates of a group of consumers whose smoking behavior tends to approximate fully informed rational decision making. We use data from the 2011–2012 TUS-CPS on the distribution of various groups of consumers across smoking statuses. We restrict the sample to those aged 30–45 to focus on adult consumers who made the decision to start smoking well after the health consequences were widely publicized through the 1964 SGR and the anti-smoking campaigns that followed. We further restrict the sample to highly educated consumers to focus on consumers most able to align their behavior with their well-informed preferences. Based on the smoking behaviors by consumers in this group, in our simulations of the rational demand for cigarette consumption we assume: an initiation rate of 5%; the age- and sex-specific cessation rates of smokers in this highly educated group; and average consumption of 16 cigarettes per smoker per day.Footnote ²¹

Figure 4 compares simulated rational cigarette consumption to observed consumption from 1964 to 2010. In 1964 rational consumption is estimated to be 18 billion packs, which is 79% of observed consumption in that year. Rational consumption is relatively constant over time. As noted above, observed consumption rises to a peak in the early 1980s and then steadily declines. In 2010 rational consumption is 11 billion packs, which is 72% of observed consumption.

Figure 4

Retrospective BCA: rational vs observed cigarette consumption.

A key feature of our approach is that the rational demand curve captures the linkages across time created by addiction. To find points on the rational demand curves $D_{\mathit{Rt}}\mid A_{t-1}$ our simulation of rational consumption at time $t$ reflects the addictive stock determined by observed consumption at time $t-1$ . As can be seen in Figure 4, persistence in the addictive stock means that the level of rational demand does not change quickly over time. The differences in the flows into and out of the smoking population due to differences in rational and observed initiation and cessation rates are small relative to the addictive stock. The large differences between rational and observed consumption that had started in 1964 and persisted over time are mainly driven by the differences in the rational and observed number of cigarettes smoked per day. Although our approach assumes that the size of the smoking population does not change quickly, we make the simplifying assumption that rational smokers would immediately cut their daily cigarette consumption. This assumption is driven in part by data limitations: research on nicotine withdrawal symptoms typically focuses on quit attempts, not on what happens when smokers cut down. Because our assumption overstates the difference between observed and rational demand, our approach tends to overstate the consumer benefits of anti-smoking policies.

To conduct our retrospective BCA we construct a series of demand curves $D_{O}$ , $D_{\mathit{CF}}$ , and $D_{R}\mid A_{t-1}$ over time. The observed market price and observed, simulated counterfactual, and simulated rational consumption at time $t$ provide points on the respective demand curves at time $t$ . Based on a consensus estimate that the price-elasticity of cigarette demand is $-0.3$ (Gallet & List, Reference Gallet and List2003), we estimate a common slope for the demand curves.Footnote ²² We then use the point-slope formula to construct a set of linear and parallel demand curves for the years 1964–2010.

Before turning to BCA results, we note an intermediate result from the constructed demand curves: a set of estimates of the vertical distances between the observed, counterfactual, and rational demand curves. These estimates are interesting in their own right because the vertical distances between $D_{\mathit{Ot}}$ and $D_{\mathit{Rt}}\mid A_{t-1}$ or $D_{\mathit{CFt}}$ and $D_{\mathit{Rt}}\mid A_{t-1}$ measure the marginal internality, i.e., the health consequences of cigarette consumption that are uninternalized due to individual failures to optimize (see Figure 1). Based on the constructed demand curves, at the start of our study period in 1964 the marginal internality is $1.07 per pack. By 2010 the marginal internality increases to $3.61 per pack.Footnote ²³ In terms of the graphed curves, at first the vertical distance between the other demand curves and $D_{\mathit{Rt}}\mid A_{t-1}$ increases over time because $D_{\mathit{Rt}}\mid A_{t-1}$ shifts inwards toward the origin more quickly. In terms of the economic forces at work, this pattern reflects the moving target nature of rational demand: as cigarette consumption falls over time, the reductions in the addictive stock reduce the level of rational consumption.

Our estimates of the marginal internality are much smaller than previous estimates of the average value of the health consequences per pack of cigarettes. Gruber and Koszegi (Reference Gruber and Koszegi2001), Cutler (Reference Cutler, Auerbach and Martin2002), Sloan et al. (Reference Sloan, Ostermann, Picone, Conover and Taylor2004) estimate that the average value of the health consequences of smoking is in the range of $20–$30 per pack; Viscusi and Hersch (Reference Viscusi and Hersch2008) estimate that the average value is even higher at $94 per pack for women and $222 per pack for men. Analogous to the health benefits approach to BCA, these estimates are based on estimates of the lifetime health consequences of smoking combined with estimates of the dollar value of health. The lifetime value of the health consequences is then averaged over the lifetime number of packs. The lifetime average value is greater than the marginal internality in part because of the sunk cost nature of the addictive stock. In addition, the marginal internality reflects the impact of the anti-smoking policies that help correct for individual failures to internalize the health consequences of their cigarette consumption.Footnote ²⁴

4.2. BCA of anti-smoking policies from 1964 to 2010

For each year from 1964 to 2010 we use the constructed demand curves $D_{\mathit{CF}}$ , $D_{O}$ and $D_{R}$ to calculate the CV measured by the area $J+K$ in Figure 1. The sum of the CVs is our measure of the consumer benefits of the anti-smoking policies from 1964 to 2010.

At a discount rate of 3%, from the perspective of 1964 the present value of benefits of the anti-smoking policies is about $573 billion (in 2010$). At a discount rate of 7% the 1964–present value is $191 billion. To provide some context for these large numbers: at a discount rate of 3% our estimate of the 1964–present value of benefits is actually larger than the 1964–present value of cigarette sales from 1964 to 2010. Another way to provide context is to calculate the benefits per smoker-year. Based on the present value of the number of smoker-years in the simulated no policy counterfactual, at a discount rate of 3% we estimate that the anti-smoking policies yield consumer benefits of $369 per smoker-year.

We are unable to develop a hard estimate of the costs of anti-smoking policies from 1964 to 2010. The estimated consumer benefits flow from a broad set of federal, state, local and private-sector anti-smoking policies including: publication of SGRs and other government reports; mandatory warning labels; a ban of television and radio cigarette advertisements; school-based health education about smoking; and mass media anti-smoking campaigns. We were unable to find systematic data on the costs of these policies. Many of the informational policies involve fairly minor resource costs. The costs of school-based health education and mass media campaigns could be more substantial.

We use data on the costs of anti-smoking policies from 1985 to 2010 to make illustrative calculations of the costs over the full period 1964–2010. Starting in the late 1980s, states began to centralize their anti-smoking efforts in state tobacco control programs (Farrelly, Pechacek & Chaloupka, Reference Farrelly, Pechacek and Chaloupka2003). The 1998 Master Settlement Agreement between the tobacco industry and state attorneys general provided an additional stream of revenues for these programs and also funded the Legacy Foundation’s national “truth” mass media anti-smoking campaign. In our illustrative calculation we assume that from 1964–1998 the annual cost of anti-smoking policies equals the average program costs over the period 1985–1998. The admittedly loose justification is that prior to 1985 the costs of public sector anti-smoking policies might have been roughly comparable but were not centralized in specific state programs. This assumption probably overestimates the costs. For the years 1999–2010 we assume the annual cost of anti-smoking policies is given by the sum of state tobacco control program expenditures and the expenditures of the Legacy Foundation.Footnote ²⁵

Based on those assumptions, our illustrative calculation is that at 3% discount rate the 1964–present value of costs is $6.5 billion. At 7% the 1964–present value of costs is $2.9 billion. Putting these cost estimates together with our estimates of the consumer benefits implies that anti-smoking policies from 1964 to 2010 yield net benefits of $566 billion (at 3%) or $188 billion (at 7%). Although the illustrative calculation is far from a hard estimate of the policy costs, the illustrative calculations suggest that the consumer benefits of the anti-smoking policies are likely to substantially outweigh the policy costs.