Delay Discounting

Abstract

Delay discounting is a form of choice behavior in which the subjective value of a reward falls as the delay to receiving it grows, measured by choices between a smaller-sooner and a larger-later reward. This article contrasts the normative exponential discount function with the hyperbolic function that fits behavior, shows how the hyperbolic shape produces the preference reversals that define impulsivity and present bias, and reviews how discount rates are measured, represented in the brain, and elevated in addiction and psychiatric disorder. The central dispute is whether intertemporal valuation reflects two competing neural systems or one common-currency signal that already encodes discounted worth. Three interactive demonstrations let the reader compare discount functions, watch a preference reversal emerge as a shared delay grows, and titrate the subjective value of a delayed reward.

Keywords: delay discounting, intertemporal choice, hyperbolic discounting, preference reversal, impulsivity

Delay discounting is the process by which a delayed reward is worth less than the same reward received now, and it is the quantitative heart of the study of self-control. Every choice with consequences that unfold over time — to save or spend, to study or relax, to use a drug or abstain — pits a reward available sooner against a larger one available later, and how steeply a person discounts the future determines which they take. The appeal of the topic to cognitive science is that this trade-off can be reduced to a measurable function: present a series of choices between a smaller-sooner and a larger-later reward, find the amount at which the two are equally attractive, and the rate at which value decays with delay falls out as a number (Frederick et al., 2002). The literature surveyed here asks what mathematical form that decay takes, why its shape makes us change our minds, how the rate is measured and represented in the brain, and what a steep rate predicts about behavior and health.

What Delay Discounting Is

A delay-discounting task has a simple structure: on each trial the chooser picks between a smaller reward available at a shorter delay and a larger reward available at a longer one — classically framed as a choice between $50 today and $100 in a year. The quantity of interest is not any single choice but the indifference point: for a given delay to the larger reward, the amount of an immediate reward that the chooser finds equally attractive. That immediate equivalent is the subjective value of the delayed reward, and plotting it against delay traces the discount function, the curve that shows how value melts away as the wait lengthens (Mazur, 1987). A theory of discounting is a rule that gives the shape of that curve.

Two properties make the phenomenon more than a truism about preferring rewards now. First, discounting is graded and lawful: subjective value falls with delay in an orderly way that a small number of parameters capture well, so the impulsivity of a person or a species can be summarized by a single discount rate. Second, the shape of the function is not a detail but the whole story, because the two candidate shapes — exponential and hyperbolic — make opposite predictions about whether preferences stay consistent as time passes. The origin of the modern treatment is Paul Samuelson's discounted-utility model, which folded all of time preference into one exponential rate for the sake of tractability while explicitly doubting it described real behavior (Samuelson, 1937). The rest of the field has been an extended test of that doubt.

Discount Functions: Exponential and Hyperbolic

The normative benchmark is exponential discounting. Here the value of a delayed reward is its amount multiplied by a constant discount factor raised to the power of the delay, so that every additional unit of waiting shrinks value by the same proportion. This is the form Samuelson's discounted-utility model assumed, and it is the only form that keeps preferences time-consistent: because a common delay multiplies both options' values by the same factor, it cannot change which is larger (Samuelson, 1937). An exponential discounter who prefers the larger-later reward will keep that preference no matter how far off both rewards are pushed or how near they are brought.

The empirical discount function is not exponential. Across species, reward types, and procedures, subjective value is described far better by a hyperbolic function, in which value equals the reward amount divided by one plus the discount rate times the delay (Mazur, 1987). James Mazur derived this form, V = A / (1 + kD), from adjusting-delay experiments with pigeons, and its single free parameter k — larger k meaning steeper discounting — has become the standard index of impulsive choice. The hyperbola falls precipitously over the first moments of delay and then flattens, so the rate of discounting is not constant but declines with delay, which is exactly what breaks time consistency. George Ainslie first drew out the behavioral implications, arguing that this steeply-then-shallowly bowed curve is the mechanism of impulsiveness and the reason organisms repeatedly sabotage their own long-term interests (Ainslie, 1975). Economics later adopted a middle form, quasi-hyperbolic (beta-delta) discounting, which grafts a one-time drop for any delay at all onto an otherwise exponential curve, preserving present bias while remaining analytically tractable (Laibson, 1997). Table 1 sets the two functional forms side by side, and the demonstration below plots them together so the difference in their shapes, and its consequences, can be seen directly.

Table 1
Exponential and Hyperbolic Discount Functions Compared
Note. A is the undiscounted amount, D the delay, r the exponential rate, and k the hyperbolic rate. Only the exponential form keeps preferences time-consistent; the declining rate of the hyperbolic form is what allows preference reversal.

Preference Reversal and Present Bias

The decisive prediction that separates the two functions concerns what happens when a common delay is added to both options. Because a hyperbola discounts near delays much more steeply than far ones, adding the same waiting time to a smaller-sooner and a larger-later reward does not preserve their order: viewed from a distance, when both rewards are far off, the larger one has the higher discounted value and is preferred; as time passes and the smaller reward becomes imminent, its value shoots up along the steep near arm of the hyperbola and overtakes the larger-later reward, reversing the preference (Ainslie, 1975). This is a preference reversal, and it is the formal definition of present bias: an outsized pull toward whatever reward is available right now. An exponential discounter, by contrast, shows no such reversal, because the common delay scales both values identically.

Kris Kirby and Richard Herrnstein demonstrated the reversal directly in humans, having people choose between a smaller-sooner and a larger-later reward while both were moved together further into the future, and observing the predicted crossover from impulsive to self-controlled choice as the front-end delay grew (Kirby & Herrnstein, 1995). The finding matters beyond the laboratory because it dissolves a puzzle about willpower: a person who genuinely intends, in the morning, to skip dessert at dinner and then abandons that plan at the table has not necessarily been inconsistent or weak, but is behaving exactly as a hyperbolic discounter must. This is why commitment devices work — by removing the smaller-sooner option before its value spikes, a chooser can bind their future self to the preference held at a distance (Laibson, 1997). The demonstration below lets the two discount curves be shifted in time so the moment of crossover, and its disappearance under exponential discounting, is visible.

Measuring Discount Rates

Turning discounting into a number requires estimating the discount rate from a person's choices. The workhorse laboratory method is the adjusting-amount procedure: the delay to the larger reward is held fixed while the immediate amount is titrated up and down across trials until the chooser is indifferent, locating the indifference point for that delay; repeating across several delays traces out the discount function, whose curvature yields k (Mazur, 1987). Because the resulting curves can be non-linear and vary in scale, discounting is often summarized model-free by the area under the curve, a single number between zero and one that falls as discounting steepens and sidesteps commitment to any particular functional form (Green & Myerson, 2004).

For fast, reliable measurement outside the adjusting procedure, Kris Kirby and colleagues built the Monetary Choice Questionnaire, a fixed set of 27 choices between specific smaller-sooner and larger-later amounts whose pattern of responses pins each person's k to one of a graded set of values (Kirby et al., 1999). Its brevity and test-retest stability made discount rate a practical trait measure, and it was with this instrument that the clinical signal first became undeniable: heroin-dependent participants discounted delayed rewards far more steeply than matched non-users, and discounted the drug itself even more steeply than money (Kirby et al., 1999). Leonard Green and Joel Myerson then showed that the same hyperboloid framework accommodates both delayed and probabilistic rewards, unifying two literatures under one discounting mathematics (Green & Myerson, 2004). The discount rate is not perfectly constant across conditions: larger rewards are discounted proportionally less steeply than smaller ones, a systematic regularity known as the magnitude effect that any complete account of discounting must accommodate (Green & Myerson, 2004). The demonstration below titrates the immediate amount against a delayed reward to locate an indifference point and reports the resulting area under the curve.

Figure 1 shows the adjusting-amount logic: indifference points fall along a hyperbola as delay increases.

Figure 1

Indifference Points Tracing a Hyperbolic Discount Function

The Neuroscience of Intertemporal Choice

Where in the brain is a delayed reward valued? An influential early answer proposed a competition between two systems. Samuel McClure, David Laibson, George Loewenstein, and Jonathan Cohen scanned people making intertemporal choices and reported that limbic and paralimbic regions tied to the dopamine system responded preferentially when an immediate reward was available, while lateral prefrontal and parietal regions engaged for all choices regardless of delay, a dual-systems pattern they read as the neural face of the beta-delta split between an impulsive present-focused process and a patient deliberative one (McClure et al., 2004). The account was elegant and widely cited, and it gave present bias a candidate mechanism in the tension between two valuation systems.

A single-system alternative soon challenged it. Joseph Kable and Paul Glimcher recorded the neural signal for subjective value as participants chose among delayed rewards and found that activity in the ventral striatum, medial prefrontal cortex, and posterior cingulate tracked the discounted value of each option — scaled by each individual's own behaviorally measured discount rate — for immediate and delayed rewards alike, implying one valuation system that represents the already-discounted worth of any option rather than two systems competing (Kable & Glimcher, 2007). Jan Peters and Christian Büchel reconciled much of the debate by mapping the wider network and the factors that modulate it, showing how prospection and the vivid imagination of future outcomes, supported by medial temporal and prefrontal regions, can reduce discounting, and casting variability in intertemporal choice as the product of interacting valuation, prospection, and control systems (Peters & Büchel, 2011). The field now treats subjective value as computed on a common neural scale, with delay one of several variables that shape it.

Discounting as a Behavioral Marker

The clinical importance of delay discounting rests on a robust regularity: people with addictions discount the future steeply. Warren Bickel and Lisa Marsch laid out the behavioral-economic case that drug dependence is, in part, a disorder of excessive delay discounting — the drug's immediate reward dominating larger but delayed health, financial, and social rewards — and argued that discount rate offers a quantitative handle on the impulsivity that sustains addiction (Bickel & Marsch, 2001). The pattern is not confined to one substance: steep discounting appears across dependence on opioids, alcohol, tobacco, stimulants, and in gambling, and Michael Amlung and colleagues confirmed in a continuous meta-analysis that steeper discounting is reliably associated with the severity of addictive behavior across studies (Amlung et al., 2017).

The signal extends past addiction. A meta-analysis by Amlung and colleagues found elevated delay discounting across a broad range of psychiatric conditions — among them major depression, bipolar disorder, schizophrenia, and borderline personality disorder — leading them to describe steep discounting as a transdiagnostic process that cuts across the categories of the diagnostic manual rather than marking any single disorder (Amlung et al., 2019). This positions discount rate as a candidate dimensional marker of the kind the Research Domain Criteria framework seeks: a measurable, mechanistically grounded index of dysfunction that varies continuously and travels across diagnoses (Lempert et al., 2019). Whether such a marker can deliver on that promise — predicting outcomes or guiding treatment rather than merely correlating with diagnosis — is an open and active question.

Worked Example

The preference reversal can be made arithmetic with Mazur's hyperbola, V = A / (1 + kD), taking a discount rate of k = 0.06 with delay D in days. Consider a choice between option A, $50 in 5 days, and option B, $100 in 30 days. Their present subjective values are:

V_A = 50 / (1 + 0.06 × 5) = 50 / 1.3 ≈ 38.46, V_B = 100 / (1 + 0.06 × 30) = 100 / 2.8 ≈ 35.71.

Because V_A ≈ 38.46 exceeds V_B ≈ 35.71, the chooser prefers the smaller-sooner option A — the impulsive choice. Now add a common front-end delay of 20 days to both rewards, so option A pays in 25 days and option B in 50 days:

V_A = 50 / (1 + 0.06 × 25) = 50 / 2.5 = 20.00, V_B = 100 / (1 + 0.06 × 50) = 100 / 4.0 = 25.00.

Now V_B = 25.00 exceeds V_A = 20.00, so the chooser prefers the larger-later option B. Nothing about the rewards changed except a shared delay, yet the preference reversed — the hallmark of hyperbolic discounting. Under exponential discounting the reversal is impossible: with a constant factor the ratio V_A / V_B is fixed at roughly 2.35 to 1 in favor of A both before and after the shared delay, so the order never changes. The preference-reversal demonstration above computes exactly these crossing values for any two rewards and any front-end delay.

Discussion

The study of delay discounting has converged on a clear picture. The normative exponential model, adopted for tractability, predicts time-consistent preferences that real choosers do not show (Samuelson, 1937). The hyperbolic model that replaced it fits behavior across species and rewards with a single parameter and, crucially, predicts the preference reversals that define impulsivity and present bias (Mazur, 1987; Ainslie, 1975). Between them sits the quasi-hyperbolic compromise that let economics absorb present bias without abandoning its analytic machinery (Laibson, 1997). The reversal itself, demonstrated directly in human choice, reframed weakness of will as the lawful output of a curved discount function rather than a moral failing (Kirby & Herrnstein, 1995).

Measurement and neuroscience have made discount rate a usable quantity. Titration and questionnaire methods turn it into a stable individual difference (Kirby et al., 1999; Green & Myerson, 2004), and imaging locates the discounted value it reflects in identifiable valuation circuitry, whether read as two competing systems or one common-currency signal (McClure et al., 2004; Kable & Glimcher, 2007). That combination — a simple, reliable behavioral measure with a plausible neural basis and strong ties to addiction and psychopathology — is what has made delay discounting one of the most productive constructs in the science of self-control (Bickel & Marsch, 2001; Amlung et al., 2019).

Current Directions

One active question is whether steep discounting can be reduced, and whether reducing it changes behavior. Jillian Rung and Gregory Madden systematically reviewed and meta-analyzed experimental attempts to lower discounting — through episodic future thinking, working-memory training, framing manipulations, and related interventions — and found reliable short-term reductions from several techniques while cautioning that the durability of the effects and their transfer to real-world choice remain to be established (Rung & Madden, 2018). The therapeutic hope is that if discount rate is a modifiable driver of impulsive behavior rather than a fixed trait, it becomes a target for treatment.

A second strand pursues discount rate as a formal marker for psychiatry. Building on evidence that steep discounting is transdiagnostic, Karolina Lempert and colleagues asked directly whether delay discounting can meet the standards of the Research Domain Criteria initiative — whether it has the reliability, construct validity, and predictive power to serve as a dimensional index of dysfunction across disorders — and mapped both the promise and the psychometric work still required before it can guide clinical decisions (Lempert et al., 2019; Amlung et al., 2019). Across both strands, the field is moving from establishing that discounting matters to asking whether it can be measured well enough, and changed reliably enough, to be acted on.

Common Misconceptions

Delay discounting just means being impatient or greedy.

It is a specific, measurable quantity — the rate at which a reward's subjective value falls with delay — not a vague personality trait. It is estimated from indifference points across delays and summarized by a discount rate or an area under the curve, which is why it can be compared across people, species, and rewards (Mazur, 1987; Green & Myerson, 2004).

Discounting is exponential, as in compound interest.

Rational-choice theory assumed a constant per-period rate, but behavior is described far better by a hyperbola whose discounting rate declines with delay. The difference is not academic: only the hyperbolic form predicts the preference reversals people actually show (Ainslie, 1975; Laibson, 1997).

Reversing a plan as a deadline nears is just irrationality or weakness.

A hyperbolic discounter must reverse preference as a smaller-sooner reward becomes imminent; the flip is the lawful output of the discount function, not a lapse. This is precisely why commitment devices help — they remove the tempting option before its value spikes (Kirby & Herrnstein, 1995; Laibson, 1997).

A steep discount rate is simply a symptom of addiction.

Steep discounting is associated with addiction, but it is elevated across many psychiatric conditions, which is why it is described as a transdiagnostic process rather than a marker of any one disorder (Amlung et al., 2019; Lempert et al., 2019).

Glossary

Adjusting-amount procedure.

A titration method that varies the immediate reward up and down at a fixed delay until the chooser is indifferent, locating the delayed reward's subjective value.

Area under the curve.

A model-free summary of discounting, the normalized area beneath the indifference points plotted against delay; it falls toward zero as discounting steepens.

Delay discounting.

The decline in the subjective value of a reward as the delay to receiving it increases; the standard measure of impulsive choice.

Discount rate (k).

The free parameter of the hyperbolic function that sets how steeply value falls with delay; larger values mean more impulsive discounting.

Exponential discounting.

Devaluation by a constant proportion per unit of delay; the normative model, which uniquely yields time-consistent preferences and no reversals.

Hyperbolic discounting.

Devaluation following V = A / (1 + kD), in which the discounting rate declines with delay; the descriptive model that predicts preference reversal.

Indifference point.

The immediate amount judged equal in value to a specified delayed reward; the raw datum from which a discount function is fitted.

Intertemporal choice.

Any choice whose outcomes are distributed across time, requiring a trade-off between rewards available at different delays.

Larger-later reward.

The bigger of the two options in a discounting task, available only after a longer delay; choosing it is the self-controlled response.

Magnitude effect.

The empirical finding that larger rewards are discounted proportionally less steeply than smaller ones, a systematic departure from a single fixed rate.

Monetary Choice Questionnaire.

Kirby's fixed 27-item set of smaller-sooner versus larger-later choices whose response pattern assigns each person a discount rate; a fast, stable trait measure.

Preference reversal.

The switch from preferring a larger-later reward to preferring a smaller-sooner one as a shared delay shrinks; the behavioral signature of hyperbolic discounting.

Present bias.

An outsized weighting of rewards available immediately relative to any delayed reward; formalized as the quasi-hyperbolic drop or the steep near arm of the hyperbola.

Quasi-hyperbolic discounting.

The beta-delta model that adds a one-time discount for any delay onto an otherwise exponential curve, capturing present bias while staying analytically tractable.

Smaller-sooner reward.

The smaller of the two options, available after a shorter delay; choosing it when the larger-later reward is objectively better is the impulsive response.

Subjective value.

The present worth a chooser assigns to a delayed reward, equal to its immediate equivalent; the quantity the discount function describes and valuation circuitry encodes.

Key Researchers

George Ainslie (b. 1944). Psychiatrist affiliated with the Coatesville Veterans Affairs Medical Center and Temple University; he originated the hyperbolic account of impulsiveness and the theory of picoeconomics, in which successive motivational states bargain over time. Wikipedia - Homepage

Warren K. Bickel (1956-2024). Professor at the Fralin Biomedical Research Institute, Virginia Tech; he established delay discounting as a trans-disease behavioral marker of reinforcer pathology in addiction. Wikipedia - ORCID

Leonard Green (contemporary). Professor of Psychological and Brain Sciences at Washington University in St. Louis; with Myerson he developed the hyperboloid framework that unifies delay and probability discounting. Google Scholar - Faculty page

Joseph W. Kable (contemporary). Professor of Psychology at the University of Pennsylvania; with Glimcher he identified a common neural valuation signal that tracks the discounted value of immediate and delayed rewards alike. Wikipedia - Google Scholar - ORCID

Kris N. Kirby (contemporary). Professor of Psychology at Williams College; he built the Monetary Choice Questionnaire and demonstrated preference reversals arising from hyperbolic discounting in human choice. Faculty page

David Laibson (b. 1966). Robert I. Goldman Professor of Economics at Harvard University; he formalized quasi-hyperbolic (beta-delta) discounting and the analysis of present-biased preferences in economics. Wikipedia - Google Scholar - Faculty page

George Loewenstein (b. 1955). Herbert A. Simon University Professor of Economics and Psychology at Carnegie Mellon University; a founder of behavioral economics, he co-authored the definitive critical review of time discounting and time preference. Wikipedia - Google Scholar - ORCID

James E. Mazur (b. 1951). Professor Emeritus of Psychology at Southern Connecticut State University; he introduced the hyperbolic equation V = A / (1 + kD), the foundational model of delayed reinforcer value. Faculty page

Samuel M. McClure (contemporary). Professor of Psychology at Arizona State University; his dual-systems neuroimaging work provided evidence that distinct neural systems value immediate versus delayed rewards. Google Scholar - Faculty page

Amy L. Odum (contemporary). Professor of Psychology at Utah State University; she established delay discounting as a stable, trait-like individual difference that generalizes across reward types. Google Scholar - Faculty page - ORCID

Howard Rachlin (1935-2021). Emeritus Distinguished Research Professor of Psychology at Stony Brook University; he founded teleological behaviorism and pioneered the behavioral analysis of self-control and discounting over time. Wikipedia - ORCID

Frequently Asked Questions

What is delay discounting?
Delay discounting is the decline in the subjective value of a reward as the delay to receiving it grows. It is measured with choices between a smaller-sooner and a larger-later reward and summarized by a discount rate, making it the standard laboratory index of impulsive choice and self-control (Mazur, 1987).

What is the difference between exponential and hyperbolic discounting?
Exponential discounting devalues by a constant proportion per unit of delay and keeps preferences consistent over time; hyperbolic discounting devalues steeply over near delays and gently over far ones, so its rate declines with delay. Only the hyperbolic form predicts the reversals seen in behavior (Samuelson, 1937; Mazur, 1987).

Why does hyperbolic discounting cause preference reversals?
Because a hyperbola discounts imminent rewards far more steeply than distant ones, adding a common delay to both options does not preserve their order. A larger-later reward preferred from a distance is overtaken by a smaller-sooner one as the latter becomes imminent and its value spikes (Ainslie, 1975; Kirby & Herrnstein, 1995).

What is present bias?
Present bias is the outsized pull toward rewards available right now relative to any delayed reward. It is captured by the steep near arm of the hyperbola or, in economics, by the one-time drop of the quasi-hyperbolic (beta-delta) model (Laibson, 1997).

How is a person's discount rate measured?
Two common methods are the adjusting-amount procedure, which titrates the immediate amount to indifference across several delays, and the Monetary Choice Questionnaire, a fixed 27-item set whose response pattern assigns a discount rate. Both yield stable, trait-like measures (Kirby et al., 1999; Green & Myerson, 2004).

Where in the brain is intertemporal choice represented?
Valuation regions including the ventral striatum, medial prefrontal cortex, and posterior cingulate track the discounted value of delayed rewards. Whether this reflects two competing systems or one common-currency valuation signal has been actively debated (McClure et al., 2004; Kable & Glimcher, 2007).

Is delay discounting linked to addiction and other disorders?
Yes. Steep discounting is a robust behavioral marker of addiction across substances and is elevated across many psychiatric conditions, which is why it is described as a transdiagnostic process rather than a marker of any single disorder (Bickel & Marsch, 2001; Amlung et al., 2019).

Can delay discounting be reduced?
Experimental techniques such as episodic future thinking produce reliable short-term reductions in discounting, though whether these effects last and transfer to real-world choice is still being established (Rung & Madden, 2018).

References