How well do any of the models previously discussed capture the behavior of addicts? In this section, the econometric literature on addictive behaviors are discussed.
An advantage of the rational addiction model is that it provides a framework in which to develop statistical models of the consumption of addictive goods (Becker et al., 1994). A key insight from this framework, and a testable implication, is that past, current, and future prices will all affect consumption behavior. From models that estimate the size of these effects researchers can predict how policy changes such as tax increases or decreases will impact across people and across time. These models are estimated using either aggregate or individual-level data on consumption of addictive goods, prices, incomes, and other determinants of consumption. The addictive good under scrutiny varies across studies: There are many studies of tobacco and smoking behavior; other possibly addictive goods that have been empirically examined in this framework include alcohol, marijuana, cocaine, gambling, and even coffee.
The key and oft-replicated finding from the empirical literature on addictive goods is that people, even addicts, respond to an increase in the current price of addictive good by decreasing current consumption (Chaloupka and Warner, 2000; Gallet and List, 2002; DeCicca et al., 2008; Sen et al., 2010). If the consumption of addictive goods were an entirely irrational behavior, then consumption would not vary systematically and predictably with prices. Contrary to irrationality, it is well established that consumption of addictive goods responds to price incentives. This implies that the consumption of addictive goods is, at least to some degree, rational.
Much of the empirical literature considers one addictive good or activity in isolation, but some work attempts to model joint consumption of multiple addictive goods. Generally, a change in the price of one addictive good will affect consumption of all addictive goods. Examples include Dinardo and Lemieux (2001), who present statistical evidence suggesting that youths substitute alcohol and cannabis, and Cameron and Williams (2001), who estimate own and cross-price effects in demand for alcohol, tobacco, and cannabis and find that alcohol and cannabis may be substitutes, whereas alcohol and cigarettes are complements. Jofre-Bonet and Petry (2008) document a complex pattern of substitutes and compliments between various addictive substances for heroin and cocaine addicts. They find that heroin and cocaine addicts use marijuana, valium, and cigarettes as substitutes.
The intertemporal influence of prices on behavior constitutes the main estimable difference between nonaddictive goods and addictive goods: The consumption of nonaddictive goods is not influenced by past or future prices. Using this testable hypothesis, many papers claim to find strong evidence of rational addiction, even for goods such as coffee (Olekalns and Bardsley, 1996). However, Auld and Grootendorst (2004) demonstrate that using aggregate data (e.g., total cigarette sales by the US state over time) to estimate addiction models tends to yield spurious evidence in favor of addiction; these methods are biased in favor of finding evidence of addiction even when the good under scrutiny is actually nonaddictive. This problem can be avoided by using individual-level data or using quasi-experimental empirical strategies. For example, Gruber and Koszegi (2001) use the preannouncement of state excise taxes on tobacco and show that smokers are forward looking in their behavior.
Similarly, statistical models show that past consumption affects future consumption in the manner predicted by rational addiction models, with an effect that diminishes over time (Gilleskie and Strumpf, 2005). The effect of past consumption on current behavior has also been found to vary markedly across people in the manner predicted by economic theory (Auld, 2005). Keeler et al. (1999) find that smokers respond to price incentives and that smokers with higher socioeconomic status are more likely to quit, all of which is predicted by the rational addiction model.
The empirical literature has had less success in cleanly distinguishing between different models of addiction. Goldfarb et al. (2001) note that commonly used empirical methods in this literature cannot be used to support or refute rational models over nonrational models. In particular, all economic models of addiction predict the observed responsiveness to prices. Levy (in press) extends the empirical literature by deriving the conditions under which the perfectly rational model of addiction can be tested against models that exhibit present bias and utility projection bias. Further, he derives estimating conditions that allow him to distinguish between the two forms of bias. Using data from the US National Health Interview Survey he finds that observed behavior strongly rejects perfect rationality, and estimates of projection bias and utility bias are strong and consistent with previous studies of nonaddictive behaviors. Consistent with the existence of these biases, Gruber and Mullainathan (2005) find that tobacco taxes increase self-reported happiness for people with a high propensity to smoke. This is suggestive that taxes are correcting for an internality.