Notes: PUBPOL 141 / Behavioral Economics for Public Policy

16 minute read


I took these notes for PUBPOl 141: Behavioral Economics for Public Policy taught by Dan Acland. Awesome class.

Relevant Freakonomics Episodes


Retirement savings

Consumer finance




  • From Push to Nudge - Q&A with Thaler and Sunstein on where to draw the “paternalism” line. It’s a written Q&A, but in my mind, students will have enjoyed the podcasts enough to give the print format a try.


  • How Stupid is Our Obsession with Laws? - Very stupid and selfish. A classic externality.
  • Show and Yell - The Prius effect: how the Prius’ distinct ability to “signal” a green commitment made it the environmentalist’s car of choice.


Benefit-Cost Analysis

  • When Helping Hurts - BCA applied to programs in education, criminal justice, and health, with surprising results.

Behavioral Phenomena

Know the following!

  • Present bias
  • Myopia
  • Complexity
  • Default bias
  • Loss aversion
  • Choice overload
  • Overoptimism
  • Salience bias
  • Mental accounting
  • Overweighting small probabilities
  • Anticipatory utility
  • Affective mis-forecasting
  • Psychology of portion size
  • Habit formation
  • Peanuts effect
  • Regret aversion
  • Intrinsic motiviation and crowding out
  • Social comparison
  • Reference dependence
  • Tunneling


The course is divided into sections: retirement, consumer finance, health, nudges, paternalism, environment, poverty and inequality, and benefit-cost analysis.


The standard model predicts consumption smoothing (you’ll spread out your money over time) and time consistency (you’ll follow through on your plans) from an assumption of exponential discounting. People are risk averse, so reducing uncertainty improves expected utility.

The classical policy analysis leaves two roles for governemnt:

  1. Adverse selection: In the annuities market, individuals know more about their longevity, so some will drop out and premiums will spiral upward.
  2. Politics of redistribution. The poor get out more than they put in. There’s a risk of moral hazard if people don’t have skin in the game.

Three behavioral issues:

  1. Present bias: Real-world discounting looks more like beta-delta discounting, with some immediate gratification. Present bias leads people to change savings plans, procrastinate, and undersave.
  2. Myopia: Even absent present bias/immediate gratification, people discount the future excessively. Perhaps due to future time perspective (a murky future), future anhedonia (weaker feelings in the future), or self-continuity (future you is a different you).
  3. Complexity: When decisions are complex, people rely on salient features, make calculation errors, and may make no choice at all. Retirement involves complex choices: how much money? what will you invest in?

Two current solutions:

  1. Social security: Force people to save and give them an annuity.
  2. Incentives to save: Tax deferred retirement plans, employer matches, savers credits.

Yet, people save too little. They know it too, and plan to increase savings. But they fail.

Clever design can take advantage of default bias. Evidence from organ donations (Johnson and Goldstein 2003) suggests that simply changing a program to opt-out can dramatically increase participation. A natural experiment in retirement savings (Madrian and Shea 2001) showed that contribution rates centered at the default rate (3%) in a opt-in scenario.

Escalating future defaults (Thaler and Benartzi 2004) also take advantage of present bias and loss aversion by delaying increases in contributions to align with the next paycheck. The policy is much more cost effective than incentives.

Mental accounting: The phenomenon of allocating income and expenditures to mental “accounts” and making financial decisions as if they were affected only by the constraints of the relevant account.

People engage in narrowly bracketed maximization, decisions based on separate budgets. A study on Japan (Ishikawa and Ueda 1984) suggests that people consume more of bonuses than regular income.

Debiasing—the removal of biases from judgement—seems to work in retirement savings. An experiment (Bryan and Hershfield 2012) triggered self-continuity and increased savings.

When too many choices are available, people experience choice overload and choose nothing. An experiment (Iyengar and Lepper 2000) with a small and large variety of jams demonstrated that way more people were willing to purchase from the smaller choice set. A follow-up experiment (Iyengar et al. 2004) showed that too many retirement funds also decreased participation. A solution could be to incentivize firms to limit the options for funds.

Consumer Finance

Subprime mortgages usually:

  • Defer cost, offering a low or zero down payment with an increasing payment schedule.
  • Are complex, with multiple rates taht may change and numerous, sometimes contingent fees.
  • Have low requirements, with some loans requiring no proof of ability to repay.

Subprime mortgages grew from 2001 to 2006, contributing to the 2008 financial crisis.

The standard model additionally requires correct predictions (an accurate probability distribution of the future) and unboumded rationality (correctly calculate/process info) and perfect attention (ability to find all relevant information).

In addition to present bias and myopia, borrowers suffer from:

  1. Salience bias: Paying attention to whatever stands out. Sometimes, the costs of a mortgage might not be obvious, whereas banks will highlight the benefits.
  2. Overoptimism: An erroneous positive forecast. An example is the planning fallacy, where students (Buehler et al 1994) underestimate their task completion times. People could overestimate their ability to pay a loan back.

Two types of responses:

  1. Heavy-handed: The Dodd-Frank Act prohibited balloon payments, recommending defaults, large late fees, and prepayment fees.
  2. Freedom-preserving: Preferences could be heterogenous. If consumers are
    • heterogenous and sophisticated, require disclosure only. APR disclosure solves compelxity partially, but more information is left to be categorized.
    • not heterogenous and not sophisticated, implement bans.
    • heterogenous and not sophisticated, it’s complicated. Maybe debiasing? Showing people the ways they are biased (future time perspective, future anhedonia, and self-continuity) might nudge them towards better decisions.

Credit cards a similarly manipulative, offering low introductory rates and minimum payments, with changing due dates and rates. Studies (Meier and Sprenger 2010, Kuchler and Pagel 2011) show that credit card users display present bias. Other behavioral issues include overoptimism, inattention, and boudned rationality.

Various bans and mandates could work. Some behavioral alternatives includes:

  • Disclosure of machine readable data and relavant statistics.
  • Opt-out default credit cards. Default payments could harm credit card holders bcause they choose not to make larger payments.
  • Financial education, which a study (Skimmyhorn 2016) of army recruits suggests is ineffective.

Poor poeple often undersave because they don’t have the minimum balance to open a bank. A standard solution is to change the system, requiring branches in low-income neighborhoods and payments via direct deposit.

Behavioral solutions include:

  1. Paycheck splitting: Harness default bias and mental accounting by splitting of parts of a paycheck automatically to a savings account.
  2. Prize-linked savings accounts (PLSs): Return interest on savings via a lottery. It draws on overweighting of small probabilities (due to availability bias ignorance prior, and coarse chance categories) and anticipatory utility (where people gain utility from anticipating events).
  3. Intentions and goal setting: setting an intention increased the rate of breast self-examination in a study (Orbell et al. 1997).


Americans want to lose weight/quit smoking but continue to die from related diseases. Behavioral explanations:

  1. Present bias and myopia: In studies, people with greater discounting buy unhealthy yogurt (De Marchi et al. 2016) and ready-to-eat/take-out food (Appelhans et al. 2012) while smoking more frequently (DeHart et al 2020).
  2. Over-optimism: People expect to quit smoking (Foulds 2002) and avoid health harms (Halpern-Felsher et al. 2004).
  3. Affective misforecasting: People fail to predict how they will feel in the future. In a smoking study (Chaloupka et al. 2019), subjects assumed that quitting would remain as difficult as it was at the beginning, an example of projection bias.
  4. Portion size: Greater portion sizes increase consumption without increasing satiation (Rolls et al. 2002).

Potential policies include:

  1. Labeling, information and defaults: Traffic light labels to decrease porportion of unhealthy foods (Thorndike et al. 2014), defaults to low-calorie sandwiches (Loewenstein et al. 2009), and graphic warning labels to overcome inattention and myopia (WHO 2014).
  2. Clever portion sizes: Decreasing portion sizes of french fries reduced consumption but maintained satiety (Varmote et al. 2018).
  3. Heavy handed alternatives: Bans on large sodas and internality taxes.

People have high elasticity for food (1.5-6), which suggests that subsidies should be effective. Some case studies:

  1. Insurance premium discounts for healthy habits do not work because they were delayed (Patel et al. 2016).
  2. Immediate incentives for gym attendance increased attendance temporarily, but results did not persist (Acland and Levy 2015).
  3. Commitment contracts with a voluntary commitment account had persistent wokring effects on gym attendance (Royer et al. 2012) and smoking cessation (Gine et al. 2010).
  4. Lotteries for home-based health monitoring were more effective when the lotter ywas small (Sen et al 2014), which is the peanuts effect.

Lessons: immediacy (rewards, now), salience, and mental accounting (rewards > reductions in paycheck) matter.

More on incentives:

  1. Non-monetary rewards: In a public service field experiment, subjects rewarded with gold stars were more motivated than those with a financial reward (Ashraf et al. 2014 ), especially for those with pro-social motiviation.
  2. Gamificiation: The use of game design elements in non-game contexts. Experiments with exercise apps (Chen and Pu 2014, Thorsteinsen et al. 2014) show that optimal games have evenly matched competitors and a greater proportion of friends.
  3. Intrinsic motivation: A classic experiment with a day care center (Gneezy and Rustichini 2000) found that after charging parents, late pickups increased. Crowding out may not be true for more integrated external forces, which involves relatedness, competence, and autonomy (Deci and Ryan 2000).


A nudge is a policy that changes behavior without restricting choice. Loewenstein and Chater offer a helpful framework. Nudges are cheap and don’t cause much unintended harms. Sometimes, nudges are particular effective, may be problematically manipulated, and could crowd out support for better alternatives (tradition and hybrid interventions/rationals).


  1. LeGrand and New: A policy is paternalistic if it is intended to replace the judgement of the individual and does so for the good of the individual. It is justifiable if and only if 1) it corrects judgement errors with respect to means, 2) there is rigorous research to show that it will make them better off, 3) the policy helps them more than it harms others, and 4) the net welfare benefits outweighs the infringement on autonomy.

Acland extends it to ends-related paternalism, as research demonstrates that people make judgement errors with respect to ends.

  1. Asymmetric Paternalism: A policy must pass the “mistakes” criterion (people conduct decision-making errors) and “minimal harms” criterion (little or no harms on those who are fully rational) (Camerer et al. 2003).

  2. Libertarian Paternalism: A policy must pass the “mistakes” criterian (people make inferior choices than if they were more rational) and “no restriction” criterion (no coercion).

What are “true preferences”? These are complicated by:

  1. Conflicting desires: People have different goals, perhaps a “delta self” and “beta self.”
  2. Observations: Experienced, real-time utility could provided an objective measure, but pleasure is only one part of happiness (Kahneman et al. 1997). Another approach is to debias people before eliciting choices (Allcott and Sunstein 2015).


A negative externality is a cost of production or consumption that is not borne by the producer or consumer of the good. It leads to a margional social cost above the marginal private cost, resulting in overconsumption.

Behavioral issues:

  1. The MPG illusion: People assume higher MPG means more gas is used.
  2. Energy savings are delayed, non-salient, and hard to compute.
  3. Electricity usage is similar, with additional inattention to real-time prices.

A positive externality or public good is underproduced for symmetric reasons. A larger subsidy for behavioral types causes over-consumption by rational types. So some believe in debiasing or nudging for behavioral types. Examples:

  1. Social comparisons: Social status affects happiness (Luttmer 2005) and labor supply (Neumark and POstlewaite 1998). By informing households about their relative electricity usage, people decreased both short and long term usage (Allcott and Rogers 2012).

  2. Salience: Real-time feedback timers reduce shower usage (Tiefenbeck et al. 2018) and electricity consumption (Shultz et al. 2015), but only with social comparison for the latter.

  3. Information: Providing informatino on fuel economy had no effect on which car was bought (Allcott and Knittel 2019). But this shouldn’t generalize to all information.

Poverty and Inequality

The standard model does not consider inequality for benefit cost analysis. But behavioral economics points the other direction:

  1. Distributional preferences: People have an interest in equal distribution of goods (Charness and Rabin 2002), particularly in the form of altruism. In contingency valuations, people are willing to pay more for equality.
  2. Reciprocity and fairness: People don’t display positive reciprocity, but they do show negative reciprocity and strong concern withdrawal (I won’t help you if you harm me).
  3. Reference dependence: Social comparisons can decrease well-being if you live in a relatively wealthier neighborhood (Luttmer 2005). Positional goods (think status symbols) set off an arms race, where everyone seeks a stronger relative status. Another example is the hedonic treadmill, where subjects revert to baseline subjective wellbeing (Wilson et al. 2000).

Scarcity is having less than you feel you need. It causes tunneling, a mangement of the scarcity at hand (Mullainathan and Shafir 2013). The poor are more focused on a tax at hand (the focus dividend) but at the expense of better planning. The wealthy benefit from slack, or extra budget, because it makes choices easier and cushions bad choices. Conversely, the poor are caught in poverty traps, as gains are likely to be lost quickly.

If scarcity impairs judgement, then perhaps a $100 cash transfer can provide more than $100 of value if it corrects irrationality. aAn experiment called SEED demonstrated more employment, self-insurance, and psychological ease. Yet, a smaller scale experiment with vendors in India and the Phillipines failed (Karlan et al. 2018).

Benefit Cost Analysis

BCA is the difference between aggregate willingness to pay (WTP) and aggregate wilingness to accept payment (WTA). To conduct BCA, predict and quantify the impacts of a policy in terms of dollars, often through QALYs, statistical lives, etc. Some problems:

  1. Loss aversion: The WTA is higher than a practically identical WTP scenario. One solution is to pick WTP or WTA depending on whether a gain or loss is at play. Another is to elicit indifference values between gains and losses.
  2. Present bias/temptation goods: Temptation goods may be overvalued. One approach is to assume present bias and attempt to correct for temptation. Another is to elicit WTP to stop engaging in a temptation good.
  3. VSL and risk preferences: A value of a statistical life is the WTP for an amount of risk reduction necessary to save one life, on average. VSL depends on risk context (planes are scary) and not on degree of risk reduction (overweighting of small probabilities).

Well-being analysis (WBA) involves measuring subjective well-being (SWB) through surveys and for different aspects of life creates an estimate of welfare impact. BCA is based on a preferentialist definition of welfare (people prefer things that provide more welfare) while WBA is a hedonic definition (good things provide greater subjective wellbeing).