Do Discount Rates Affect Behaviors Like Saving and Smoking?

Researchers have long been interested in understanding how people make decisions about behaviors that have long-term consequences for their well-being, like saving or smoking. These decisions require individuals to consider how they value costs and benefits that occur in the future versus those in the present- for example, saving requires sacrificing consumption today in order to have higher consumption in the future. Economic theory predicts that the discount rate - the rate at which individuals discount future costs and benefits - will be a critical factor in these decisions.

Different people are likely to have different discount rates, since some people are more patient (low discount rate) while others are more impatient (high discount rate). Do individuals' discount rates help to explain their decisions about behaviors like saving and smoking? This question is examined in a new study, "Individual Laboratory-Measured Discount Rates Predict Field Behaviors" (NBER Working Paper 14270), which is authored by an interdisciplinary team of economists and psychologists including researchers Christopher Chabris, David Laibson, Carrie Morris, Jonathon Schuldt, and Dmitry Taubinsky.

The authors use a laboratory task to compute an individual-specific discount rate and then estimate the effect of the discount rate and demographic factors on behaviors such as saving and smoking. While other studies have demonstrated a relationship between laboratory measures of discounting and various behaviors, this study is unique for its use of a large, diverse sample to examine a wide range of behaviors (fifteen in all) and compare the predictive strength of the discount rate to that of demographic variables in explaining these behaviors.

The authors begin by estimating discount rates for over 500 subjects using a laboratory task. This sample includes participants from three different studies; each study examines a different set of behaviors, but all subjects are given the same laboratory task. Participants are asked to choose between an immediate reward and a long-term reward that would be paid after a specified number of days - for example, whether they would prefer $40 today or $55 received in 62 days. They are asked to make 27 such choices, where the size of the immediate and long-term rewards and the number of days of delay are varied for each. To give them an incentive to answer questions honestly, participants have a 1-in-6 chance of receiving one of the rewards they selected. The answers to these questions are used to calculate an individual-specific discount rate. The authors assume a hyperbolic discount function, as this has been shown to predict behavior better than a constant discount rate.

The first study examines health-related variables associated with making tradeoffs between the present and future, including body mass index (BMI), exercise frequency, dieting, and smoking. The authors find that the discount rate is a significant determinant of BMI, exercise, and smoking and that it can explain 15 to 20 percent of the variation (or differences in these variables across people) in each of these measures. Interestingly, no other variable explains as much of the variation as the discount rate. When the authors create an index of these four health variables, the results are even more striking - the discount rate explains one-quarter of the variation in the index, while no other variable explains more than one-tenth. A second study that looks at BMI and exercise obtains similar results.

The third study examines a much larger set of behaviors, including additional health-related behaviors such as dental check-ups, flossing, and selection of healthy food as well as financial behaviors such as gambling, late payments on credit cards, and saving. This study had the largest sample size, but was administered over the internet rather than in a controlled laboratory setting. The authors find mixed results when they look at the effect of the discount rate on individual behaviors. However, when the behaviors are combined in an index, the discount rate has a significant effect on behavior, though the share of variation explained by the discount rate is smaller than in the other two studies; its effect is smaller than that of age, but larger than that of sex or education.

Next the authors present a theoretical framework to explore how much of the variation in behavior we would expect discounting to explain. Using an example where there are many factors that may have some influence on a behavior (for example, smoking may be affected by exposure to cigarette advertising, seeing celebrities smoke, a desire to use smoking for weight loss, etc.), they show that even if the discount rate was measured perfectly and was a more important determinant of behavior than the other variables, it would still account for only a small share of the total variation. A second insight of the theoretical model is that discounting will be a stronger determinant of an index of behaviors than of any single behavior.

At first glance, the key findings of the paper appear to be at odds with each other. On the one hand, discount rates are only weakly correlated with individual behaviors - no correlation is above 0.3 and many are close to zero. On the other hand, demographic factors have even less predictive power, despite being measured more precisely, and the relative superiority of the discount rate increases when looking at an index of behaviors rather than individual behaviors. The authors' theoretical results help to reconcile these findings.

The paper's results support two broad conclusions. First, "there exists a domain-general behavioral disposition towards impatience/impulsivity" and second, "a discount rate estimated through a set of intertemporal monetary choice questions constitutes a useful, though noisy, measure of this disposition." The authors suggest that future research could use discount rates as phenotypes in genetic studies designed to identify the molecular mechanisms of intertemporal choice.

The authors acknowledge financial support from a NARSAD Young Investigator Award and DCI Postdoctoral Fellowship awarded to Christopher Chabris, an NSF ROLE grant to J. Richard Hackman and Stephen Kosslyn, and NIA (P01 AG005842, R01 AG021650) and NSF (0527516) grants to David Laibson.