Psychology of Learning

Module 10: Decision-Making 2

Part 2: Why We Fail at Self-Control

Looking Back

In Part 1, we explored decision-making under uncertainty through intuition & heuristics. The availability heuristic leads us to overestimate vivid or recent events. The representativeness heuristic produces base rate neglect & conjunction fallacies. These mental shortcuts evolved in ancestral environments where they generally worked well but often mislead in modern contexts. Now we turn to a pervasive decision-making challenge: self-control—choosing between immediate gratification & delayed benefits.

Real-Life Decision-Making: Self-Control & Social Cooperation

Most decisions we make are between two choices. Usually, one of these choices is easier or more fun at the moment, perhaps even illegal or immoral. The alternative choice is harder or less fun, but offers superior long-term consequences. Every dieter knows that dessert tastes better than the feeling of being slim, yet maintaining ideal weight requires repeatedly resisting immediate gustatory pleasure for delayed health & appearance benefits.

Self-control is control of behavior using internal controls (such as diligence or morality) rather than external controls (such as rules & laws). Self-control involves choosing larger-later rewards over smaller-sooner rewards, resisting immediate temptations for long-term benefits (Rachlin, 2000).

Self-control decisions pit present self against future self. Your present self wants dessert, wants to sleep in, wants to spend money frivolously. Your future self will bear the consequences—excess weight, failed exams, financial insecurity. Self-control means prioritizing future self’s welfare over present self’s immediate desires. This temporal conflict between selves pervades human decision-making, determining success in health, education, career, relationships, & finance. In single-player iterated decision-making situations, a player’s present self (trial N) is in competition with the player’s future self (trial N + 1). Choices that maximally benefit the present self often harm the future self (impulsive options), while choices that maximally benefit the future self are usually undesirable to the present self (prudent options). Even though both ‘players’ are the same person, it can be difficult for individuals to make effective choices because both the present & future consequences of decisions are rarely set in stone (Rachlin, Brown, & Baker, 2001).

In situations where the consequences of our actions fall to others, there is again usually two basic choices, one easy (often illegal or immoral as well) & one harder but with superior long-term consequences. These are called social cooperation problems. Should you litter (easy) or carry trash to a bin (effortful but benefits community)? Should you cheat on taxes (immediate financial gain) or pay honestly (maintains public services)? Should you overuse shared resources (tragedy of the commons) or restrain consumption for collective sustainability?

The problem of self-control can be considered a special form of the social-cooperation problem. In the social-cooperation problem, acts which benefit me now harm others later. In self-control, acts which benefit me now harm me later. The temporal structure is identical—immediate benefits versus delayed costs—but the affected party differs. Both require overriding immediate self-interest for longer-term welfare, whether personal or collective. As Rachlin, Brown, & Baker (2001) explained, in real-life self-control & social cooperation situations, as in prisoner’s dilemma games, the important question is: Will I (or will the other player) cooperate? In real-life social cooperation situations, the uncertain probability of reciprocation may lead to defection. In real-life self-control situations, the person is asking whether their future self will cooperate with the present self’s sacrifice—will the future payoff actually materialize?

Empirical Evidence: Self-Control Versus Social Cooperation

Brown & Rachlin (1999) provided direct experimental evidence for the relationship between self-control & social cooperation. Using a game board paradigm with red & green keys & doors, participants made choices with both immediate & delayed consequences. In the self-control condition, participants played alone—choosing the smaller immediate reward (opening the left door) yielded a red key that enabled access to better options on subsequent trials, while choosing the larger immediate reward (right door) yielded a green key limiting future options. In the social cooperation condition, the structure was identical, but the consequences of each participant’s choices were experienced by another player.

The results were striking: participants showed significantly more self-control when playing alone (48-67% cooperation across experiments) than social cooperation when playing with another person (19-24%). Importantly, there was little transfer between the two conditions—being good at self-control did not necessarily make someone good at social cooperation, & vice versa. This suggests that the crucial variable is the subjective probability of reciprocation—the perceived likelihood that one’s sacrifice will be matched. When playing alone (self-control), participants could be relatively certain their future self would benefit from their present sacrifice. When playing with another person (social cooperation), uncertainty about whether the other would reciprocate reduced cooperation. Increasing the reward magnitude increased self-control but not social cooperation, further supporting the role of reciprocation probability (Brown & Rachlin, 1999).

Tragedy of the commons is a description of conflict where each individual, seeking out their own selfish best interest, utilizes common resources in a way that is ultimately destructive to the community as a whole. When everyone overuses a shared resource, the resource is depleted, harming all users (Hardin, 1968).

Classic example: Herders sharing a common pasture. Each herder benefits immediately by adding another animal to their herd. But overgrazing depletes the pasture, harming all herders. Individually rational decisions produce collectively disastrous outcomes. Modern examples include overfishing (each fishing company benefits from catching more fish, but overfishing collapses fish populations), pollution (each factory benefits from cheap disposal, but collective pollution harms everyone), & climate change (each nation benefits from unrestricted emissions, but collective emissions threaten civilization).

How Important Are Self-Control & Social Cooperation?

There was considerable variability in children’s ability to wait for the preferred reward in Mischel’s marshmallow test. Some children immediately chose the single marshmallow. Others waited the full 15 minutes for two marshmallows. This variability in childhood self-control predicted dramatic differences in life outcomes measured decades later.

Children that had been better able to wait for a marshmallow as preschoolers later exhibited: higher SAT scores (averaging 210 points higher), better stress management & emotional regulation, lower body mass index (healthier weight), lower rates of substance abuse, higher educational attainment, better career outcomes & financial stability, & more successful relationships.

These findings suggest that self-control capacity in early childhood—the ability to delay gratification—predicts success across virtually every life domain. Self-control isn’t just one skill among many; it appears to be a foundational capacity enabling long-term goal pursuit despite immediate temptations.

Success in an individual’s life always requires us to bypass momentary pleasures. If you want to be financially successful (rich!), you must bypass the impulsive desire to spend money now & instead delay that gratification by saving & investing. If you want to be physically fit, you must bypass the desire to eat junk food & instead exercise regularly & maintain healthy eating. If you want to excel academically or professionally, you must bypass entertainment & socializing to study & work diligently.

All success that a society has comes from social cooperation. When people are given the ability to choose between selfishness & social cooperation, those societies where people choose social cooperation thrive while those societies where people choose selfishness fail. Nations with high levels of trust, civic engagement, & cooperation—where people restrain selfish impulses for collective benefit—enjoy prosperity, low crime, effective governance, & social cohesion. Nations where selfishness dominates suffer poverty, conflict, corruption, & institutional failure.

Universal human morality is a human condition described by C.S. Lewis which is shared by all individuals, an internal control which helps keep selfish behavior in check. Lewis argued that despite cultural differences, all human societies recognize basic moral principles discouraging purely selfish behavior (Lewis, 1943).

Lewis noted that every culture has prohibitions against murder, theft, & lying—violations of social cooperation. Every culture values courage, justice, & mercy. These universal moral intuitions function as internal constraints on selfishness, making social cooperation possible. Without these internal controls, external enforcement alone couldn’t maintain cooperative societies. Self-control & moral restraint are foundational to civilization itself.

Why We Often Fail at Self-Control

Often, when we are making bad choices (choosing impulsively), we know we are making bad choices but we do it anyway. Why? Though occasionally we make poor choices because we genuinely don’t understand consequences, most self-control failures occur despite full knowledge of outcomes. The dieter knows that dessert will impede weight loss. The student knows that skipping studying will harm grades. The smoker knows that cigarettes cause cancer. Yet they choose immediate gratification anyway. Understanding self-control failure requires examining multiple contributing factors.

Failure to Understand Consequences

Even a baby knows that dessert tastes good; jars of Gerber desserts are much easier to feed a baby than jars of Gerber strained peas. However, the long-term consequences of eating nothing but desserts—nutritional deficiencies, health problems, obesity—are not immediately apparent. Young children lack understanding of these delayed consequences, making self-control especially difficult.

Many self-control failures in adults also reflect incomplete understanding—not of consequences themselves, but of probabilities & cumulative effects. The smoker may intellectually know that smoking causes cancer but fail to appreciate the magnitude of the risk, the cumulative nature of damage, or the difficulty of quitting later. The gambler may understand that casinos have an edge but fail to appreciate how small edges produce large losses over time. Education about consequences—especially vivid, concrete, emotionally engaging information—can improve self-control by making delayed costs more psychologically real.

Delay Discounting: Rewards Lose Value Over Time

Delay discounting is the decrease in the subjective value of a reward as the delay to receiving that reward increases. Future rewards are worth less to us psychologically than immediate rewards, a phenomenon documented across species & reward types (Mazur, 1987).

Simple example: Would you rather have $100 today or $100 in a year? Almost everyone prefers $100 today. But would you rather have $100 today or $200 in a year? Now preferences vary. Some people prefer the immediate $100 (steep discounting), others prefer waiting for $200 (shallow discounting). The delayed reward must be larger to compensate for the delay—its subjective value has been discounted.

The hyperbolic equation V = A / (1 + kD) best describes delay discounting across species & reward types, where V = discounted value, A = actual value, k = discount rate (individual difference parameter), & D = delay. Larger k values indicate steeper discounting (more impulsive); smaller k values indicate shallower discounting (more patient).

Rachlin, Brown, & Cross (2000) provided strong empirical support for hyperbolic discounting using a cross-modality magnitude production method where participants drew lines proportional to the subjective value of delayed & probabilistic rewards. Their data confirmed that both delay & probability discount functions are hyperbolic, not exponential as assumed in standard economic models. For delay discounting, the hyperbolic function v = V / (1 + kD) fit the data far better than exponential alternatives. This finding has important implications: hyperbolic discounting, unlike exponential discounting, produces preference reversals—explaining why we can make excellent long-term plans but fail to execute them when the moment of choice arrives.

Critically, the shape of hyperbolic discounting means that discounting is steepest for short delays (the curve drops most rapidly near the origin) & more gradual for longer delays. This shape explains common self-control failures: A reward available ‘right now’ versus ‘in 1 hour’ shows enormous discounting, but the same reward available ‘in 100 days’ versus ‘in 101 days’ shows minimal discounting. The immediacy of temptation, not just its temporal proximity, drives impulsive choices.

Preference Reversals: Why Good Plans Fail

Preference reversals are changes in what a person prefers as time passes, even though objective consequences remain constant. Hyperbolic discounting produces preference reversals that explain why people make excellent long-term plans but fail to execute them.

Example: In the morning, a dieter plans to skip dessert after dinner. At that point, both dinner & dessert are hours away, so both are discounted substantially. The larger delayed benefit of weight loss exceeds the smaller delayed pleasure of dessert. But when dessert arrives—immediate, visible, aromatic—the pleasure is no longer discounted. The dessert’s value spikes while the weight loss remains distant. Preferences reverse: the morning’s planned restraint gives way to evening’s indulgence.

This pattern is universal & explains countless self-control failures. The student plans to study tonight but chooses video games when evening arrives. The saver plans to invest the bonus but spends it impulsively when the check arrives. The addict plans to quit tomorrow but continues using when withdrawal threatens. The key insight is that these aren’t failures of knowledge or intention—they’re predictable consequences of hyperbolic discounting & the resulting preference reversals.

Precommitment is a situation where the individual makes a decision well before the actual time where consequences would be given & removes the ability to change that decision later. By constraining future choice, precommitment helps overcome predictable preference reversals (Rachlin & Green, 1972).

Perhaps they could only bring enough money to the restaurant to afford dinner but not dessert. Or they could tell their dinner companion to stop them from choosing dessert, enlisting social enforcement of their morning commitment. Successful self-control often relies on such precommitment strategies: removing temptations from the environment, creating external constraints, or establishing social accountability before willpower is tested.

AUC as an Empirical Measure of Discounting

The discount rate can be derived from empirical data using a measure called Area Under the Curve (AUC), developed by Myerson, Green, & Warusawitharana (2001). This method plots subjective value against delay, creating a curve. The area under this curve quantifies discounting: larger AUC indicates less steep discounting (more self-control), smaller AUC indicates steeper discounting (less self-control).

AUC provides a single number summarizing an individual’s discounting rate across multiple delays, enabling comparison between people. Research using AUC has revealed that steep discounters (low AUC) show higher rates of substance abuse, obesity, gambling problems, & credit card debt. Delay discounting appears to be a trans-disease process—a common factor contributing to diverse self-control failures across domains.

Temporal, Social, & Probability Discounting Share Cognitive Resources

If we think of the human as existing at multiple points in time, we can talk about a present self & a series of future selves (my tomorrow self, my next-week self, my next-year self, my 20-years-from-now self). Delay discounting means I care less about my future selves than my present self—their welfare is psychologically distant, so I discount their interests.

Similarly, I care less about socially distant others than about myself & close others. A stranger’s welfare is psychologically distant, so I discount it. Research reveals that temporal discounting is social discounting: People who discount future rewards steeply also discount socially distant others’ welfare steeply. The same psychological process—discounting psychological distance—operates across temporal & social dimensions.

Probability discounting refers to the loss of value (utility) of a reinforcer that is to be received only probabilistically. A $20 prize that is certain is worth more than a $20 prize with 50% probability. Rachlin, Brown, & Cross (2000) demonstrated that probability discounting follows the same hyperbolic form as delay discounting. They showed that probability can be expressed as equivalent waiting-time, unifying delay & probability discounting into a common framework. People who discount delayed rewards steeply also discount uncertain rewards steeply. This suggests that distance in time, social space, & probability all reduce subjective value through common psychological mechanisms.

These relationships suggest that a single cognitive system handles all forms of psychological distance. Training someone to be more patient (reduce temporal discounting) might also increase prosocial behavior (reduce social discounting) & increase willingness to take calculated risks (reduce probability discounting). Interventions targeting one form of discounting may generalize to other forms.

Time Horizons: The Ultimate Form of Discounting

Time horizons are points in time in the future beyond which we do not understand or plan. Time horizons are like the ultimate form of delay discounting. However, beyond the horizon line (whether temporal, social, or probabilistic), we aren’t just discounting value—we’re completely failing to represent or consider consequences.

A teenager’s time horizon might extend only a few years into the future. They can’t meaningfully imagine or plan for age 40 or 60. Consequences beyond their time horizon might as well not exist. This explains why adolescents engage in risky behaviors despite long-term dangers: The long-term consequences fall beyond their representational horizon. They’re not just discounting distant outcomes; they’re failing to represent them at all.

Rachlin, Brown, & Cross (2000) found that time horizons (the H parameter in their probability discounting model) systematically varied with reward amount—larger amounts were associated with longer time horizons. This finding connects to broader questions about how people mentally represent the future & why some consequences fail to influence present behavior at all.

Expanding time horizons—helping people vividly imagine & emotionally connect with distant future states—can improve self-control. Interventions that have people write letters to their future selves, create detailed visualizations of future outcomes, or interact with age-progressed images of themselves can extend time horizons & reduce impulsive choices.

Uncertainty of Consequences

The consequences of our actions, especially the long-term consequences, are uncertain. If it were possible to know that smoking was guaranteed to give you lung cancer & make you die a painful death, no one would smoke. This is not how the world works, of course. Smoking increases lung cancer risk, but many smokers don’t develop lung cancer. This uncertainty makes the long-term costs feel less real, less compelling.

Brown (2006) investigated how uncertainty affects prudent decision-making using a computer game with both present consequences (points received) & future consequences (keys determining future options). The study manipulated two types of uncertainty: present-uncertainty (whether points would be received on the current trial was probabilistic) & future-uncertainty (whether the key received would enable good future options was probabilistic). The results were clear: both present-uncertainty & future-uncertainty decreased prudent decision-making compared to control conditions with no uncertainty. Moreover, the combination of both types of uncertainty decreased prudent decision-making even further than either type alone.

These findings help explain real-world self-control failures. For a smoker contemplating quitting, there is present-uncertainty about today’s consequences (will I experience severe withdrawal?) & future-uncertainty about long-term consequences (will quitting actually prevent disease?). This dual uncertainty provides justification—or perhaps an excuse—for impulsive behavior (Brown, 2006; Brown & Lovett, 2001). When the world feels unpredictable, people are less likely to make decisions driven by long-term consequences.

Brown & Lovett (2001) demonstrated this in a modified prisoner’s dilemma paradigm. When information about future consequences & past behavior was removed from the game, cooperation (analogous to self-control) decreased substantially. Even more dramatically, when outcomes were made probabilistic rather than deterministic, participants’ ability to ‘solve’ the game dropped to near-chance levels. The qualitative nature of learning also changed: with certain outcomes, participants showed ‘insight learning’ with sudden strategy adoption, while with uncertain outcomes, they shifted to gradual ‘trial & error’ learning. This suggests that uncertainty doesn’t just make self-control harder—it fundamentally changes how people approach decision-making.

Feedback & Probability Estimation

A crucial finding from Brown’s research program is that feedback can counteract the negative effects of uncertainty on self-control. In Brown (2006), participants who received feedback about their past behaviors & outcomes showed significantly higher levels of prudent decision-making when future consequences were uncertain. The mechanism appears to be feedback’s ability to help people construct accurate mental probabilities about behavior-consequence relationships.

When facing uncertain situations, people must estimate probabilities. Without feedback, these estimates may be wildly inaccurate, leading people to equate ‘uncertain’ with ‘unpredictable’ or even ‘random.’ But probabilistic does not equal random. As Brown (2006) concluded, ‘anything that reduces the uncertainty people feel in a world full of uncertainty will increase their ability to make prudent decisions.’ Feedback provides the denominator (past behaviors) & numerator (outcomes) needed to construct meaningful probability estimates.

Brown, Taylor, & Bazaldua (2008) replicated & extended these findings, showing that as predictability of long-term consequences decreased, self-control decreased—but feedback about behaviors & outcomes almost entirely negated the negative effects of unpredictability. Participants who received no feedback showed a steady decline in self-control as predictability dropped (from ~65% at 100% predictability to ~25% at 62.5% predictability). But participants who received feedback maintained relatively stable self-control across all predictability levels (~50-60%).

The practical implication is profound: in situations where long-term consequences are uncertain—which describes most real-world self-control challenges—keeping records of behaviors & outcomes can dramatically improve self-control. The dieter who tracks daily food choices & weight changes, the student who tracks study hours & exam scores, the saver who tracks spending & account balances—all are creating the feedback needed to understand the probabilistic but non-random relationship between present behavior & future outcomes.

Willpower: A Limited Cognitive Resource

Baumeister refers to willpower, which is the cognitive resource required for self-regulation. Willpower is like a muscle: the more we exercise it, the stronger it gets, but it also gets fatigued with use. Exercising self-control depletes willpower temporarily, making subsequent self-control attempts more likely to fail. This is called ego depletion.

Willpower might best be defined as the ability to resist impulses, or the ability to do the things that don’t come naturally. Every act of self-control—resisting temptation, forcing yourself to study, suppressing emotional outbursts, making difficult decisions—draws from a limited willpower reservoir. As this reservoir depletes, self-control capacity diminishes.

Research on ego depletion shows that people who exert self-control on one task (resisting cookies, suppressing emotions, making choices) subsequently perform worse on unrelated self-control tasks. Their willpower has been depleted. This explains why self-control is hardest at the end of long days, after making many decisions, or when already stressed. The willpower reservoir is low, leaving little capacity for additional self-control.

However, like a muscle, willpower can be strengthened through regular exercise. People who consistently practice self-control in one domain (maintaining exercise routines, sticking to budgets, practicing meditation) show improved self-control in other domains. The willpower ‘muscle’ grows stronger with training. This suggests that self-control is a trainable capacity, not a fixed trait—an encouraging finding for anyone struggling with self-control challenges.

Looking Forward

We’ve explored self-control as a foundational decision-making challenge where immediate gratification conflicts with long-term welfare. Multiple factors cause self-control failure: incomplete understanding of consequences, delay discounting (hyperbolic function showing steepest discounting for short delays), preference reversals, uncertainty of consequences, & willpower depletion. Experimental evidence from Brown & Rachlin (1999) shows that self-control is structurally similar to social cooperation, with probability of reciprocation as the crucial variable. Temporal, social, & probability discounting share cognitive resources, following the same hyperbolic form (Rachlin, Brown, & Cross, 2000), suggesting common underlying mechanisms. Crucially, feedback can counteract the negative effects of uncertainty on self-control by enabling accurate probability estimation (Brown, 2006; Brown et al., 2008). In Part 3, we’ll explore applications: interventions to improve self-control, strategies for managing willpower, & connections to life history theory, examining how to quit smoking as a prototypical self-control challenge.