Judgment and Decision Making, Vol. 8, No. 3, May 2013, pp. 236–249 How to measure time preferences: An experimental comparison of three methods David J. Hardisty∗ Katherine F. Thompsony David H. Krantzy Elke U. Weberz Abstract In two studies, time preferences for financial gains and losses at delays of up to 50 years were elicited using three different methods: matching, fixed-sequence choice titration, and a dynamic “staircase” choice method. Matching was found to create fewer demand characteristics and to produce better fits with the hyperbolic model of discounting. The choice-based measures better predicted real-world outcomes such as smoking and payment of credit card debt. No consistent advantages were found for the dynamic staircase method over fixed-sequence titration. Keywords: time preference, discounting, measurement, method. 1 Introduction tial versus hyperbolic discounting models. The contin- uously compounded exponential discount rate (Samuel- Throughout life, people continually make decisions about son, 1937) is calculated as V = Ae−kD, where V is the what to do or have immediately, and what to put off un- present value, A is the future amount, e is the base of til later. The behavior of a person choosing an imme- the natural logarithm, D is the delay in years, and k is diate benefit at the cost of foregoing a larger delayed the discount rate. This is a normative model of discount- benefit (e.g., by purchasing a new car rather than sav- ing, which specifies how rational decision makers ought ing towards one’s pension) is an example of temporal to evaluate future events, but it has often been employed discounting (Samuelson, 1937). Similarly, choosing to as a descriptive model as well. The hyperbolic model avoid an immediate loss in favor of a larger, later loss (Mazur, 1987) is a descriptive model, calculated as V (e.g., by postponing a credit-card payment) is another ex- = A / (1+kD), where V is the present value, A is the ample of discounting. Laboratory measures of discount- future amount, D is the delay,1 and k is the discount ing predict many important real-world behaviors that in- rate. Although the hyperbolic and exponential models volve tradeoffs between immediate and delayed conse- are often highly correlated,2 the hyperbolic model often quences, including credit-card debt, smoking, exercise, fits the data somewhat better (e.g., Kirby, 1997; Kirby and marital infidelity (Chabris, Laibson, Morris, Schuldt, & Marakovic, 1995; Myerson & Green, 1995; Rachlin, & Taubinsky, 2008; Daly, Harmon, & Delaney, 2009; Raineri, & Cross, 1991). Because most recent psycho- Meier & Sprenger, 2010; Reimers, Maylor, Stewart, & logical studies of discounting have employed the hyper- Chater, 2009). At the same time, numerous studies have bolic model, our further analyses in this paper will focus established that time preferences are also determined by a on this model. number of contextual factors (for an overview, see Fred- In addition to these two popular discounting metrics, erick, 2003). others have been proposed and tested (for a recent review, Despite the growing popularity of research on tempo- see Doyle, 2013). In contrast, investigations of differ- ral discounting (Figure 1), there is relatively little consen- ent experimental procedures for eliciting discount rates sus or empirical research on which methods are best for are rare. A comprehensive review paper on discounting measuring discounting. Most of the theoretical and em- (Frederick, Loewenstein, & O’Donoghue, 2002) noted pirical efforts have been directed at testing rival exponen- the huge variability in discount rates among studies, and Support for this research was provided by National Science Foun- hypothesized that heterogeneity in elicitation methods dation grant SES-0820496. Supplementary material to this article is might be a major cause. Fifty-two percent of studies re- available through the table of contents of the journal, at the link on the top of each page. 1The units of delay in the hyperbolic model are unspecified, but in Copyright: © 2013. The authors license this article under the terms the experimental literature delay is often measured in days or years. of the Creative Commons Attribution 3.0 License. Measuring in years yields numbers that are easier to work with (e.g., ∗Graduate School of Business, Stanford University. Email: a yearly k of 0.3 rather than a daily k of 0.0008), so we will measure [email protected]. delay in years throughout this paper. yDepartment of Psychology, Columbia University. 2Researchers desiring to distinguish between hyperbolic and expo- zDepartments of Management and Psychology, Columbia Univer- nential models can design experimental stimuli to make them orthogo- sity nal, following the guidelines of Doyle, Chen, and Savani (2011). 236 Judgment and Decision Making, Vol. 8, No. 3, May 2013 Time-preference measurement 237 directly. For example, it might ask the participant what Figure 1: Graph illustrating the rising popularity of amount “X” would make her indifferent between $10 im- temporal discounting as a research topic over the last mediately and $X in one year. twenty years. Results come from searching on ISI How do discount rates from these two elicitation meth- Web of Knowledge with Topic=(temporal discount*) OR ods compare? Several studies have concluded that match- Topic=(delay discount*) NOT Topic=(lobe) Search lan- ing yields lower discount rates than choice (Ahlbrecht guage=English Lemmatization=On Refined by: General & Weber, 1997; Manzini, Mariotti, & Mittone, 2008; Categories=( SOCIAL SCIENCES ), broken down by Read & Roelofsma, 2003). What are the reasons or year. mechanisms for this difference? One hypothesis is that, in choice, people are motivated to take the earlier re- ward, and pay relatively more attention to the delay (rather than the greater magnitude) of the later reward, whereas the matching methods, which typically ask for a match on the dollar dimension, focus them on the mag- nitude of the two rewards and thus results in a better at- tentional balance between the magnitude and delay at- tributes (Tversky, Sattath, & Slovic, 1988). This atten- tional hypothesis predicts order effects (specifically, that the first method will bias attention throughout the task), but unfortunately neither study investigated task order,3 so it is difficult to know whether participants’ experi- ence with one method influenced their answers on the other method. Frederick (2003) compared seven differ- ent elicitation methods (choice, matching, rating, “total”, sequence, “equity”, and “context”) for saving lives now or in the future. He also found that matching produced lower discount rates than choice, but again, order effects were not explored. He speculated that the choice task creates demand characteristics: offering the choice be- tween different amounts of immediate and future lives viewed used choice-based measures, 31% used matching, implies that one ought to discount them to some extent— and 17% used another method. “otherwise, why would the experimenter be asking the question” (Frederick, 2003, p. 42). In contrast, the match- ing method makes no suggestions as to which amounts 1.1 Measuring discount rates: Choice ver- are appropriate. sus matching Further evidence that characteristics of offered choice Choice-based methods often present participants with a options can bias discount rates comes from a pair of series of binary comparisons and use these to infer an in- studies that compared two variations on a choice-based difference point, which is then converted into a discount measure. One version presented repeated choices that rate. For example, suppose a participant, presented with a kept the larger-later reward constant with the amounts choice between receiving $10 immediately or $11 in one of the smaller-sooner reward presented in ascending or- year, chooses the immediate option, and subsequently, der, while the other version employed the same choice presented with a choice between $10 or $12 in one year, pairs, but with the smaller-sooner amounts presented in chooses the future option. This pattern of choices im- descending order. The order of presentation influenced plies that the participant would be indifferent between discount rates, such that participants were more patient $10 today and some amount between $11 and $12 in one (i.e., exhibited lower discount rates) when answering the year. For analytic convenience, we assign their indiffer- questions in descending order of sooner reward (Robles ence point as the average of the upper and lower bound, & Vargas, 2008; Robles, Vargas, & Bejarano, 2009). This which would be $11.50 in this case. This indifference suggests that observed discount rates are at least partly point can then be converted into a discount rate using one a function of constructed preference (Stewart, Chater, of the discounting models discussed above. For example, & Brown, 2006) and “coherent arbitrariness” (Ariely, using the continuously compounded exponential model, Loewenstein, & Prelec, 2003), rather than a stable in- this would yield a discount rate of 14%. The matching 3In Ahlbrecht & Weber (1997), participants always did matching method, in contrast, asks for the exact indifference point before choice. Read & Roelofsma (2003) does not state the task order. Judgment and Decision Making, Vol. 8, No. 3, May 2013 Time-preference measurement 238 dividual preference. One explanation offered by Robles based discount rates predict consequential decisions and and colleagues is the