An empirical investigation of heterogeneity in time preferences and smoking behaviors

Abstract

In this study I exploit the findings of a population survey in which I had the opportunity to introduce questions on time preferences. The study uses retrospective data to estimate the relationship between present-biased preferences and the decisions to start and quit smoking. Respondents stating present-biased preferences are not more prone to start smoking but quit later in life, and after more failed attempts. These preliminary results strongly suggest that smokers form a heterogeneous population and it can be argued that such heterogeneity means that taxes on cigarettes are a blunt and inefficient instrument of public health.

Introduction

Heterogeneity of preferences with respect to smoking has potentially important consequences for anti-smoking policies. While a tax on cigarettes can improve the welfare of individuals who are not fully rational in the sense that such individuals do not anticipate the consequences of their decision to start or to not quit smoking, the tax will harm perfectly rational individuals who know at the time they start smoking that they will quit before the risk of continued smoking become large (Warner et al., 1995, or Glied, 2002). Abstracting from externalities,¹ the only justifications for a tax in the rational addiction model derive from behavioral characteristics that generate imperfect rationality, such as changes in discount rate over the life cycle (Becker and Mulligan, 1997), an inability to benefit from perfect information about either the health consequences of cigarette consumption (Suranovic et al., 1999) or one's own level of addiction properties to smoking (Orphanides and Zervos, 1995, Wang, 2007). In the presence of such behavioral “anomalies”, a tax conveys the proper level of information on cigarette consumption to future or potential rational smokers² and is effective because the addictive nature of cigarette consumption means that consumption behavior is persistent (Gruber and Zinman, 2000, Glied, 2002, Gilleskie and Strumpf, 2005). A tax can also be justified if individuals are rational in the long run but not in the short run due to self-control problems such as an inconsistency in their time preferences called bias for the present (Laibson, 1994, Glazer and Weiss, 2007, Gruber and Köszegi, 2001): if individuals lacking self-control knew they would face a large tax on cigarettes they would not start smoking and the (anticipated) tax would then protect them against their impulsivity and would be beneficial in the long run (Geoffard, 2002, Gruber and Mullainathan, 2002, Laux and Peck, 2007; the same argument is developed by Cutler et al., 2003, in the case of obesity).³ This might explain the empirical result in Kan (2007) that smokers who intend to quit are more supportive of taxes and bans than the rest of the population (smokers with no intention to quit and non-smokers).

A uniform tax, however, will hurt rational individuals who are able to smoke when they enjoy it and stop before it gets detrimental by deterring some of them from accessing an enjoyable and safe⁴ consumption (Goldbaum, 2000, Suranovic, 2005). Studies in the epidemiology of smoking behaviors in most developed countries seems to show that that population of smokers who are able to quit before it is too late makes a growing proportion of the population who is smoking or has been smoking during their life: Pierce and Gilpin (1996) show that 50% of males smokers born in 1885–1889 were still smoking at age 75 but that the same percentage will be reached at age 35 for cohorts born 1975–1979.

In the presence of heterogeneity (some individuals are rational in their decision to start smoking and others not), anti-smoking policies able to target sub-populations, such as a commitment device targeted at present-biased individuals (O’Donoghue and Rabin, 2002), raising legal age limits (Suranovic, 2005, Sajjad and Billimek, 2007 and financing cessation programs Glied, 2002) might be more efficient, from a welfarist perspective, than a uniform tax rate on all smokers.⁵

The evidence on determinants of smoking suggests substantial heterogeneity in smoking behaviors: non-observable (idiosyncratic) heterogeneity in Gilleskie and Strumpf (2005), heterogeneity in daily consumption and ability to quit in Contoyannis and Jones (2001) and Etilé (2004), and heterogeneity in health status: Healthier smokers or smokers who do not suffer from adverse health shocks are less likely to quit (Jones, 1994, Contoyannis and Jones, 2001, Lahiri and Song, 2000, Clark and Etilé, 2002); however, non-smokers in poor health are more likely to start than non-smokers in good health (Adda and Lechene, 2004, Carbone et al., 2005). Other empirical evidence point toward psychological traits as potential sources of heterogeneity in smoking behaviors: attitudes toward the future (saving, home ownership, seat belt usage) in Evans and Montgomery (1994), feeling of helplessness and risk aversion in Hirschman et al. (1984) and risk aversion in Barsky et al. (1997), and Anderson and Mellor (2007). In the field of psychology, Chassin et al. (1984) show a link between smoking and various scales of attitudes and normative beliefs.

The present study belongs to that field of research of associations between attitudinal traits (heterogeneity in attitudes toward risk and future) and smoking behaviors. It provides new evidence on the link between heterogeneity in inconsistencies in time preferences (as shown in Benzion et al., 1989) and heterogeneity in smoking behaviors. It is assumed that preferences for present are attitudinal traits that do not vary with time or context (at least for adults) and influence the ability of individuals to resist addiction to cigarette; it is also assumed that individuals are not fully aware of their inconsistencies in time preferences. These two assumptions are not tested in the present study which, as a result, aims at measuring statistical associations between time inconsistencies and addiction to smoking. The rationale for studying the link between time-inconsistencies and smoking is that, as previously stated, taxes on cigarettes can help time-inconsistent smokers but always hurt time-consistent ones.

I exploit a unique data set from France that includes individual-level information on both time preferences (and their inconsistencies) and smoking history. I particularly focus on the relationship between present-biasedness in time preference and the decisions to start and quit smoking. Present-biased individuals discount more heavily than the usual exponential discount rate, at least in the short run. As detailed in Section 2 the measure used for present-biasedness is a crude one: it does not allow one to distinguish between purely “hyperbolic discounting” (Laibson, 1994, Ainslie, 1991) and any behavior weighing current decisions more than future ones (described by O’Donoghue and Rabin (1999) as a coefficient β < 1 applying to all periods but the current one in a lifetime utility otherwise characterized by exponential discounting at rate δ: $U^t(x_t,x_{t+1},\mathrm{...})=u(x_t)+\beta {\sum }_{s\ge 1}{\delta }^{s}u(x_{t+s})$.⁶ What really matters is to be able to separate individuals who are consistent in the short and long run from those who are not (Ainslie, 1974, O’Donoghue and Rabin, 1999). In that sense, I use inconsistencies in time-preferences to reveal self-control issues (modeled as dual selves behaviors as in Thaler and Shefrin, 1981, irrational urges as in Benzion et al., 1989, or impulsivity as in Mitchell, 1999); the analogy between self-control and inconsistencies in time preferences has been pointed out in Frederick et al., 2002 and Fudenberg and Levine, 2006); I then test empirically whether such self-control problems correlate with smoking addiction and generate preference reversals and regret for past choices, such as starting smoking. The rationale for capturing self-control issues through present-biasedness is that it allows me to separate (consistent) impatience from inconsistencies as a determinant of smoking behaviors: impatience as a factor of smoking has been studied (Kwaja et al., 2007) but not separated from inconsistent time preferences.

The theoretical model is quite straightforward: I expect that present-biased individuals will experience regret more often than individuals with consistent time preferences, either patient or impatient. For that reason, a tax on cigarettes could help inconsistent individuals not to start a behavior they will regret. The focus here is on regretting the decision to have started smoking in the past, independent of the underlying taste for smoking. The theoretical model assumes a lifetime constant taste for smoking independent of time inconsistencies (albeit not necessarily of consistent impatience or patience). Individuals are not aware of their status (consistent or inconsistent) and start smoking based on their tastes for smoking and their anticipation of the health hazards of smoking (this is in line with the standard rational addiction model). When comes the time to quit (as rational smokers anticipated they would) time consistent smokers do as planned but present-biased ones cannot quit and remain stuck with their detrimental habit.

The research questions to be empirically tested are therefore:

Research question 1: Preferences for smoking are independent of inconsistencies in time preferences: I study the differences in starting behaviors across present-biasedness to control that the relationship between present bias and quitting later does not come from the fact that present-biased individuals also happen to have a stronger taste for smoking: the absence of a difference in taste for smoking is tested through the absence of difference in the starting process across time preference categories.

Research question 2: There is an association between present-biasedness and quitting behavior and I expect present-biased to quit later, make more (unsuccessful) attempts at quitting, and be less price sensitive in their decision to quit.

To analyze the careers of smokers (e.g., the probabilities of starting and quitting, the duration in the state of smoker, and the price elasticity of the duration in the smoker state), I follow a well-established methodology of duration models applied to retrospective data, as found in Douglas and Hariharan (1994), Douglas (1998), Forster and Jones (2001), Madden (2001), López Nicolás (2002), and Grignon and Pierrard (2004).

This study adds to the literature on heterogeneity of smoking behaviors in that the survey used elicit time preferences, including present-biased preferences, participation and duration in smoking behaviors at the individual level and, as a result, allows to describe heterogeneity in time discounting and how it correlates with the decisions to start and quit smoking. It is the first empirical study of the link between heterogeneity in present-biasedness and heterogeneity in smoking behaviors.

The results indicate that self-reported present-biased discounters are not more likely to start smoking or to start younger but quit later in life, and after more failed attempts. This provides support to the idea that present-biasedness is associated to “regret” and more difficulties to quit a detrimental behavior. Non-present-biased individuals, therefore, seem to have better chances of controlling their smoking behavior and might be hurt by a tax on cigarettes. However, empirical results are less clear about the prediction that present-biased individuals are less price-sensitive in their decision to quit.