To measure how policy changes affect social welfare, economists typically look at how policies affect behavior. They then use a formal model to infer welfare consequences from the behavioral responses. The advantage of this approach is that many behaviors can be readily measured in easily available micro-data sets. The disadvantage is that the model used to make this inference is often empirically unverified. Since different models can map the same behavior to very different welfare impacts, it becomes hard to draw firm conclusions about many policies.
An excellent example of this conundrum is the taxation of addictive substances such as cigarettes. There is wide agreement that consumption of cigarette is fairly price sensitive (Chaloupka and Warner, 2001). But this fact is equally consistent with two very different models of why people smoke. Under the rational addiction model pioneered by Becker and Murphy, agents decide to smoke in the same way they decide on other things: they trade off the long-term costs of smoking against the immediate pleasures all the while taking into account the addictive properties of nicotine. In such a model, taxes will reduce smoking but will also make smokers worse off: the price of a good that they enjoy is more expensive.
An alternative class of models suggests that smoking decisions are not made optimally. For example, in the model of Gruber and Koszegi, time inconsistent smokers have self-control problems: they would like to quit smoking but cannot. In this model, a rise in taxes also reduces smoking. But now the reduction in smoking makes smokers better off: the higher taxes provide a commitment device that helps them deal with their self-control problem. fully
These models have very different policy implications. Under the rational addiction model, the only reason to tax cigarettes is the presence of interpersonal externalities. Under the more behavioral model, optimal taxes can be quite high, even absent interpersonal externalities, due to the self-control benefits of taxation. Critically, since consumption can be price sensitive under both models, existing evidence based solely on smoking behavior does not allow one to distinguish the correct model for welfare and policy analysis.
In this paper, we go beyond the existing literature to propose a new approach. We do so by drawing on a source of data that is sometimes used in other disciplines but rarely by economists: data on self-reported happiness. In principle, happiness is a direct welfare measure that can overcome the limitations of other approaches to welfare analysis of policies such as excise taxation. This measure has been repeatedly validated as a good correlate of well-being, using alternative psychological, physiological, and economic measures of well-being. Since the two models above make very different predictions of how taxes ought to affect happiness, this data allows us to distinguish between them in a way that traditional behavioral data cannot.
We use two independent data sets to examine the effect of cigarette taxes on happiness. These are the General Social Survey’s (GSS) that are carried out in the United States (since 1973) and in Canada (since 1985). Both surveys repeatedly ask a random sample of respondents to report on their well-being. In addition, the survey also contains information on a host of other demographic variables and, in many years, on smoking behavior.